2,696 491 11MB
Pages 402 Page size 612 x 792 pts (letter) Year 2010
JBR Special Issues 1999 - 2009
The La Londe Conference in Marketing Communications and Consumer Behavior
Contents
• 1999 Lalonde conference Yves Evrard, Wayne D. Hoyer and Alain Strazzieri Introduction to the special issue on marketing communications and consumer behavior
• 2001 La Londe conference Christian Derbaix , Lynn R. Kahle, Dwight Merunka and Alain Strazzieri Introduction to the special issue on marketing communications and consumer behavior
• 2003 La Londe conference Gilles Laurent and Judy Zaichkowsky Preface to La Londe 2003 special issue: communications and consumer behavior
• 2005 La Londe conference Curtis P. Haugtvedt, Dwight R. Merunka and Luk Warlop Marketing communications and consumer behavior: Introduction to the special issue from the 2005 La Londe seminar
• 2007 La Londe Conference Søren Askegaard, Dwight R. Merunka and M. Joseph Sirgy Marketing communications and consumer behavior: Introduction to the special issue from the 2007 La Londe conference
• 2009 La Londe Conference Virginie De Barnier, Chris A. Janiszewski, Dwight R. Merunka, Stijn M.J. van Osselaer Marketing communications and consumer behavior: Introduction to the special issue from the 2009 La Londe conference
Journal of Business Research 58 (2005) 339 – 340
Editorial
Introduction to the special issue on marketing communications and consumer behavior$ Abstract This special issue includes a selection of papers presented at the 1999 La Londe Seminar. After a description of the conference, this introduction shortly presents the papers included in the special issue: two guest papers from Sidney Levy and Robert Peterson, and five competitive papers dealing with marketing communications and consumer behavior. D 2003 Elsevier Inc. All rights reserved.
1. A small and ‘‘interactive’’ conference The seminar is unlike many other conferences in that it is small (72 participants this year) and is characterized by a friendly and informal atmosphere of exchange between researchers from all over the world (14 countries represented this year). In order to facilitate fruitful exchanges, each presentation lasts 45 min. Participants are scholars with an interest in developing a scientific understanding of consumer behavior in order to better design and implement advertising and other marketing communication tools.
2. An outstanding scientific committee IAE and the conference organizers are very grateful to the respected colleagues who served as members of the scientific committee. These individuals reviewed and constructively commented on the 83 papers submitted. Thirtyone papers were accepted and included in the proceedings. Heartfelt thanks are given to the following scholars: Gerald Albaum (University of Oregon), Rajeev Batra (University of Michigan at Ann Arbor), Russell W. Belk (University of Utah), Christian Derbaix (FUCAM, Mons), Peter Doyle (University of Warwick), Alain Jolibert (ESA, Universite´ Pierre Mende`s-France, Grenoble), Lynn R. Kahle (University of Oregon), Michel Laroche (Universite´ Concordia, Montre´al), Siew Meng Leong (National University of Singapore), Claude R. Martin (University of Michigan at $ In 1974, Eric Langeard created the IAE ‘‘International Research Seminar in Marketing’’, primarily known as the ‘‘Se´nanque Seminar’’. Eric passed away in November 1998. At the beginning of the 1999 seminar, Pierre Eiglier, Bob Peterson, Yves Evrard and Sid Levy gave a tribute to the founding father of the seminar. Adieu Eric! All your friends are grateful.
0148-2963/03/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00099-7
Ann Arbor), Hans Mulbacher (Universita¨t Innsbruck), Robert A. Peterson (The University of Texas at Austin), Rik Pieters (Tilburg University), Christian Pinson (INSEAD), Bernard Pras (Universite´ de Paris IX-Dauphine, ESSEC), Don E. Schultz (Northwestern University, Medill), Jan-Benedict Steenkamp (Catholic University at Leuven), Alice M. Tybout (Northwestern University, Kellogg), Mark Uncles (University of New South Wales),W. Fred van Raaij (Erasmus University) and Arch G. Woodside (Tulane University). This year, for the first time, we requested other respected colleagues to be ad hoc reviewers. We appreciated strongly their fruitful cooperation. The following are the names of those who helped: Dana Alden (University of Hawaii), Mark Alpert (University of Texas at Austin), Philippe Aurier (Universite´ de Montpellier 2), Michelle Bergadaa (Universite´ de Gene`ve), Peter Bloch (University of Missouri), Susan Broniarczyk (University of Texas at Austin), Daniel Caumont (Universite´ de Nancy), Jean-Charles Chebat (HEC Montreal), Terry Childers (University of Minnesota), Bettina Cornwell (University of Memphis), Ayn Crowley (Drake University), Rene´ Darmon (ESSEC), Kalpesh Desai (SUNY-Buffalo), Pierre Desmet (ESSEC and Paris-Dauphine), Marie-Laure Gavard-Perret (Universite´ de Savoie), Pierre Gre´gory (Universite´ de Paris I), Pam Henderson (Washington State University), Jeff Inman (University of Wisconsin) Richard Ladwein (IAE Universite´ de Lille), David Mazursky (Hebrew University of Jerusalem), Dwight Merunka (IAE Universite´ d’Aix-Marseille), Ashesh Mukherjee (McGill University), Pascale Quester (University of Adelaide), Peter Reed (Monash University), Elyette Roux (ESSEC), Eric Spangenberg (Washington State University), Pierre Valette-Florence (Universite´ de Grenoble), Bjorn Waliser (Universite´ de Strasbourg), Judy Zaichkowsky (Simon Frazier University), Monique Zollinger (Universite´ de Tours).
340
Editorial
3. A vivifying ‘‘heritage assessment’’ session Since this conference was the last of the millennium, three distinguished scholars were requested to participate to a heritage assessment session. We asked these individuals to provide a panoramic view of the evolution of research and to evaluate the contributions on topics relevant for the conference. Frank Bass presented brightly a history of diffusion theory in marketing (not included in this special issue). Sid Levy impassioned the attendants with his probably impartial and surely erudite narration of the battles between ‘‘quantitativists’’ and ‘‘qualitativists’’ in the Consumer Behavior research community. He advocated convincingly for more realistic research and more open-minded researchers. Bob Peterson presented published and forthcoming metaanalyses results showing convincingly that the variance accounted for in several areas of consumer behavior research was very weak. He argued that one of the causes of that unpleasant fact was that the data collected by questionnaires are no more than on the spot constructions.
4. From Nelson’s theory to internet communication through advertising affective reactions effects The five competitive papers selected for inclusion in this special issue address various issues, from testing an old advertising theory to exploring the new wold of Internet communication. Sridhar Moorthy and Scott A Hawskins present the results of a laboratory experiment aimed at verifying if advertising repetition (real or told) influences more the perceived quality of experience (vs. search) goods as argued by Nelson (1970). They find limited support to the Nelson’s theory. June Cotte, Robin Higie Coulter and Melissa Moore study the effects of credibility and of the perception of manipulative intent on the ad’s ability to induce the intended emotion (specifically guilt). Their experiment shows notably the disrupting effect of the perception of manipulative intent. Mark I. Alpert, Judy I. Alpert and Elliot N. Malz study the effects of background music on audience responses
(moods, attitudes and intentions), using structural profiles of music elements and introducing purchase occasion congruity. Their experiment supports the positive effect of congruent music. Majorie Dijkstra, Heidi E.J.J.M. Buijtels and W. Fred van Raaij study the effects on consumer cognitive, affective and conative responses of single-medium and multiplemedia campaigns (TV, print, Internet). Among other results, their laboratory experiment suggests that Internet alone is hardly effective, but can play a complementary function. David R. Fortin and Ruby Roy Dholakia study the effect of interactivity (specifically three levels for online advertising) on involvement with the ad and arousal. Their experiment with 360 self-recruited web users indicates notably a mediating role of the perception of social presence and a plateau effect of interactivity on involvement.
5. Final words Finally, as cochairs and coordinator of this conference, we strongly wish to express our gratefulness to Arch Woodside, editor in chief of the Journal of Business Research, for instigating in 1995 a special issue devoted to a selection of the papers. This action was a determinant factor of the success of this biennial meeting of worldwide scholars involved in consumer behavior and marketing communications research.
Yves Evrard HEC, Paris, France Wayne D. Hoyer University of Texas, Austin, USA Alain Strazzieri* IAE Aix, Universite´ d’Aix-Marseille, Clos Guiot, 13540 Puyricard, France E-mail address: [email protected] * Corresponding author.
Journal of Business Research 58 (2005) 341 – 347
The evolution of qualitative research in consumer behavior Sidney J. Levy Department of Marketing, University of Arizona, Room 320U, McClelland Hall, Tucson, AZ 85721-0108, USA
Abstract This article is my response to an invitation to prepare a ‘‘heritage assessment’’ for presentation to the International Research Seminar at La Londe les Maures in June 1999. Such an assessment is, according to Alain Strazzieri, an authorized view of what is worth remembering from the literature about topics in consumer behavior research. This charge is an open one and I will execute it freely. My presentation of this view has been authorized by the seminar’s Scientific Committee. Otherwise, you will have to judge my authority on its merits. It does have the weight of my advanced years, giving me the advantage of having started formal study of consumers when research into their behavior was still young. My main themes are the intellectual battling and intellectual cycling that have gone on in our field, especially with reference to the role of qualitative research. D 2003 Elsevier Inc. All rights reserved. Keywords: Qualitative research; Consumer behavior; Intellectual battling; Intellectual cycling
1. Intellectual conflict: theme and variations The perspectives I derived in the late 1940s came mainly from the topics and methods of the behavioral sciences that were the fashion of the period and that I studied at the University of Chicago. Some of my teachers were outstanding names in the behavioral sciences. I studied with or was exposed to the thinking of Robert Redfield, W. Lloyd Warner, Everett Hughes, Herbert Blumer, Edward Shils, David Reisman, William E. Henry, Don Campbell, and others. The so-called ‘‘Second Chicago School’’ of Sociology was flourishing (Fine, 1995) and my home base, The Committee on Human Development, provided a multidisciplinary and eclectic education. Among my fellow students and friends were Herbert Gans, Lee Rainwater, and Erving Goffman. There was great intellectual stimulation, with argumentation over philosophies, subject matter, and methods. Some of these controversial topics show how the modern roots of our present concerns reach back to the 1920s and 1930s. For instance, criticism of the work of Piaget (1926) still goes on because he generalized grandly from observing small samples of children. In basic scientific tradition, he drew inferences and conclusions from his observations and other researchers dispute his hypotheses and try to refute them. The behavioral science disciplines
E-mail address: [email protected] (S.J. Levy). 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00107-3
grew vigorously in the 1940s and 1950s. Much of the study that goes on nowadays in the consumer research field adds some basic knowledge to the early learning; it also refines, elaborates, and, most strongly, applies what we have learned. However, the fundamental intellectual battle still goes on between the partisans of nomothetic approaches and the partisans of idiographic study. The scholars reading this article are surely familiar with much of the history of our discipline, but I will remind us of some highlights that stand out in my mind. Major phenomena in the research area are the growth of qualitative methods and the resistance to them by people who prefer to rely on quantitative methods. Gary Fine (1995) gives an excellent account of this struggle at Chicago, describing the conflict between the advocates of quantitative methods and those committed to the prewar (World War II) emphasis upon qualitative methods and field research methodologies. He notes that Blumer criticized ‘‘those who were attempting to achieve exactitude in social science at the price of direct and naturalistic study’’ (p. 145) and cites David Reisman’s recollection that the enmity was a problem for graduate students who worked with Hughes and with me, as well as for nontenured colleagues, particularly with Nelson Foote and Anselm Strauss. . .Unrealistically, if understandably, translating acidulous comments by faculty members into actual prescriptions of what would pass muster, some able graduate students feared to write a dissertation
342
S.J. Levy / Journal of Business Research 58 (2005) 341–347
without tables in it. . .I sometimes had the dismal experience of having as a doctoral candidate someone who had been a spirited undergraduate and watching that person become more timid and less original as time went by. (Reisman, in Berger, 1990, p. 63) More than 50 years later, when I ran a workshop on qualitative research, one of the doctoral students responded with this fretful statement: I suppose what I mean is that there has to be a certain consistency in epistemology, ontological assumptions, and the use of techniques of data collection and presentation. Why do we have to constantly justify the use of certain techniques, and why do we willingly or unwillingly participate in perpetuating the dominance of a particular discourse of ‘‘doing research;’’ in other words, why are we ashamed to be purely qualitative, just as we are not ashamed to be purely quantitative? I find that I am constantly grappling with this in my own research, and would like to find a way of dealing with this issue, so that we can have other ways of collecting, analyzing, and presenting data. One wonders about the persistence of this contentiousness, and the lack of objectivity shown by so many academicians with doctorates to their names. These remarks are true for fields other than marketing: a professor of finance recently raved in my presence that he hated the behavioral people he believed were ruining his field. This struggle has a certain one-sidedness to it as the quantitative people have the dominant paradigm and the greater power. Scholars who are interested in qualitative research usually understand the role of measurement and do not deny its value and even the need for it in the large place it occupies in the activities of the research world. They believe that they are bringing new and useful insights to the marketing field. These scholars mainly would like to be free to do their kinds of research, to get it published, and to be hired as regular members of a faculty. However, generally, the dominant paradigm people resist, show great hostility, and at many schools refuse to hire any faculty who are qualitatively oriented. They behave defensively, foolishly acting as though their livelihoods are threatened by the projective techniques and ethnographies that will replace surveys, regressions, and multivariate methods. At the 1989 conference of the Association for Consumer Research, they complained fearfully that qualitative researchers might be taking over the conference. They also attack, looking down on the qualitative people, and sneer at them as if they were chiropractors or dentists who could not succeed as physicians. At a recent conference of the American Marketing Association, one participant implied that qualitative researchers are like modern artists who do unrealistic, distorted work because they are not competent at drawing: that is, unlike Picasso who we knew could really draw, they
supposedly cannot properly measure and hold their work up to the criteria that govern scientific research. The oddity here is that an ethnography or thick description surely captures a situation more realistically than a particular statistic, no matter how large the sample or whatever the level of confidence.
2. Conflict and cycling of ideas However, the ebb and flow of intellectual conflict is an ordinary thing, in the nature of science, and is a requirement for ideas to demonstrate their viability. It does not afflict qualitative consumer researchers only. The prewar sociologists at Chicago resisted the incoming proponents of quantitative methods, just as historians who pursue the ancient tradition of qualitative study of the past sputtered and fumed when the cliometricians arrived with their threatening statistical techniques for measuring history and their use of tables as well as tales. Often, of course, the emotional component of the critics’ reactions is so great as to have little to do with the science of the matter. Real scholars are calmer, having a comprehension of the variety of ways science goes on and understanding that ultimately it should address itself to rival hypotheses and whether they can be shown to be false. Like the presumption of innocence in law, any generalization—whether drawn from qualitative or quantitative study—may be taken as valid until there is evidence that it is not true; then, the exception does not prove the rule (except in the sense of testing it) but provides an occasion for fresh theory or insight. All researchers make observations of some kinds, draw inferences from those observations to arrive at their preferred ways of explaining the phenomena they have observed and how generally they occur. However, there are many ways of doing that. I am reminded of a research meeting at which a psychologist told the head of a pharmaceutical company that some consumers were stomach-oriented and others were anal-oriented when it came to taking laxatives. The executive asked ‘‘How many are there of each type?’’ and the psychologist replied, ‘‘Enough of them, sir!’’—and undoubtedly there are. The differences in approaches became apparent in the 1930s when the influx of European scholars brought to America and to the business world quantitative methods as in the survey and panel methods of Paul Lazarsfeld (Lazarsfeld, 1940) and depth techniques as in the psychoanalytic interpretations of Ernest Dichter. Hal Kassarjian (1994) describes in detail these European roots. His informative chapter appears in the valuable overview of research traditions in marketing provided by Laurent et al. (1994). The study of consumer behavior gained momentum in the late 1940s and 1950s, especially as its value was embraced by advertising agencies on behalf of their clients. A lot of that work was proprietary and did not appear in journals—and the Association for Consumer Research and
S.J. Levy / Journal of Business Research 58 (2005) 341–347
the Journal of Consumer Research did not show up until the 1970s. However, there was the large-scale survey work by Alfred Politz and a variety of so-called motivation or qualitative research studies—many of which were reported on in the trade press. The 1950s were a kind of heyday of communications research (Klapper, 1960) and studies in persuasion (Hovland et al., 1953). The lively character of the work in that post-World War II period is especially evident in the two volumes on Consumer Behavior edited by Lincoln H. Clark (1954 – 1955). These volumes report on two conferences held at the University of Michigan that brought together scholars from several research organizations. The disciplines of sociology, psychology, and economics were well-represented, and although there were tables and statistics reported—and the University of Michigan Survey Research Center had a dominating role—the papers had a markedly thoughtful and discursive quality, suggesting some breadth of thinking and conversation that went on about the lives of consumers, their life cycles, and their decision making. Such earlier subjects and methods are like the trunk of a tree that in time grew many branches and leafed out into the variety of sessions and the flowering richness of specific topics apparent in modern journals and in the contemporary conferences of the Association for Consumer Research, the Society for Consumer Psychology, and the International Research Seminar. I will not repeat here some of the historical developments that I have noted on other occasions (Levy, 1991, 1994, 1996). The ‘‘motivation research’’ of the 1950s was subsequently perceived to have died of its excesses, and the rise of the computer assisted in pushing positivism and its methods to the fore. But all along, the protagonists of qualitative research persisted, partially aided (and partially disserved) by the business world’s huge embrace of focus groups. By the 1970s, the growth of consumer research generally, in both business and academic studies, was so great that there was room for qualitative work and need for the ideas it could engender. Since then, importantly, qualitative research scholars have made names for themselves as leaders in the application of the various methods drawn from the various behavioral sciences—ethnographies, semiotic studies that include literary analyses and examination of rhetoric, inquiries eliciting projective materials, etc. Simply put, these kinds of studies contrast with surveys and experiments by turning from the measurement of variables to intensive description, interpretation of situations, and the search for meaning.
3. The need for qualitative research Reasonable scholars—presumably meaning those who agree with me—may have read the great essay by Kurt Lewin (1935) on The Conflict Between Aristotelian and Galileian Modes of Thought in Contemporary Psychology, or the more recent perspectives provided by Morgan and
343
Smircich (1980) and Hunt (1991). Lewin points out the lingering Aristotelian influence that gives a valuative color to modes of thought and searches empirically for the lawfulness of events through their regularity and frequency. The result is a preference for understanding events that possess a certain persistence and stability and the placement of less common events and individuality as outside the realm of science. This underlying idea that the lawful and the individual are antithetical shows itself when the critics want to know if qualitative findings are generalizable. As Lewin (1935, p. 15) says, ‘‘If one shows a film of a concrete incident in the behavior of a certain child, the first question of the psychologist usually is: ‘Do all children do that, or is it at least common?’’’ Lawfulness as frequency means that repetition is a major criterion, and leads to the commanding role of statistics, ‘‘the most striking expression of this Aristotelian mode of thinking’’ (p. 16). As a field theorist, a leader in the Gestalt school of psychology, Lewin was less concerned with frequent repetition than with the nature of the whole situation. Since all events must in truth be lawful, interest in the situation and the holistic attitude that underlies its study mean that we perceive the events of consumer behavior as dynamic outcomes of multiple forces. As Lewin sums up, ‘‘The tendency to comprehend the actual situation as fully and concretely as possible, even in its individual peculiarities, makes the most precise possible qualitative and quantitative determination necessary and profitable’’ (1935, p. 25). Historically, more and more scholars have come to seek that comprehension of the situation, or at least to approximate it. Thus, despite the resistance that still occurs, this aspiration has gained a large number of adherents, more visibility, and a fairly loud voice. The more fully researchers want to understand consumer behavior, the more they are motivated to use methods that allow the interaction of multiple forces to show itself. Were this to be done ideally, a consumer event would be intensively scrutinized by a team of thinkers representing every discipline, explaining every possible antecedent and current element with any possible effect on the action at issue. Short of that, we engage in the varieties of research activity called qualitative research. That means going beyond the exponents of behaviorism (unlike behavioralists) who want to limit their data and thinking to explicit and plainly observable acts and events, to stimulus and response; and who focus on the degree to which the phenomena occur and the level of confidence we may have in getting the same results were the study to be repeated. Being more broadly behavioral means the unleashing of all we can do to find out what consumers’ lives are like, especially with reference to the situations that interest us. Getting reports on their actions may seem the easier part, although some skeptical scientists want to watch the behavior rather than be told about it in an interview or a laboratory. That means going out into the field, mingling with the subjects as if being one of them, making detailed notes, creating the methods of case studies, participant
344
S.J. Levy / Journal of Business Research 58 (2005) 341–347
observation, autodriving, ethnography, and thick descriptions. We are also challenged to deal with consumers’ inner lives, including their introspections, with all the hazards entailed in their self-expression, their truth and lies, their ignorance, uncertainty, face work, contradictions, and mechanisms of defense—all in all, their being complex human beings. Methods arise to gather this sort of information, such as depth interviews, focus groups sessions, and projective techniques. That also means we have to interpret the complexity and make inferences from what is observable to what is underlying and theoretically operative. Despite their necessity, introspection, and interpretation arouse the inevitable demon of subjectivity, on the part of the subjects, the researchers, and certainly their critics. We enlarge our vocabularies, having to learn words such as phenomenology, hermeneutic, semiotic, emic and etic, hegemonies, rituals, myths, symbolic consumption, and postmodernity. That irritates a lot of people who forget they also had to learn what partial derivatives are. However, what are we to do? We are people thinking about people and giving a lot of emphasis to how they perceive themselves and their relations to the outside world and the products they consume. Regardless of the long history I am describing here, it is a sign of the irregular situation of qualitative research that examples of its application still turn up in the business press as if it were some remarkable newcomer. Recently, The Wall Street Journal (1999) reported Chrysler’s ‘‘first vehicle designed entirely through an unconventional market-research process known as ‘archetype research.’ (Ah, Carl Jung!). . .overseen by a. . .French-born medical anthropologist named G. Clotaire Rapaille.’’ The research involved poring over focus group protocols in which participants were asked to drift back to their childhoods and jot down the memories invoked by the prototype of the vehicle. A project about Procter & Gamble’s Folgers Coffee was also mentioned in which ‘‘Dr. Rapaille concluded that aroma sells coffee more than taste does because aroma invokes feelings of home.’’ If this was news to Procter & Gamble, perhaps the ‘‘unconventionality’’ and ever-renewed sense of novelty about qualitative research is due to the persisting naivete´ and capacity for astonishment of corporate personnel. Nevertheless, in the face of continuing contention, acrimony, and defensiveness on all sides, qualitative scientists have persisted and amplified their numbers. Their stream of work seems to be well established now. It has taken root and a substantial number of scholars are devotedly pursuing the content and methods of qualitative study and publishing their work. Although qualitative research is a subfield of a relatively small field, it has radiated out into the world. Sage Publications sells an active list of materials about qualitative research and such work has partisans around the globe at schools and companies. Most marketing research organizations claim to do qualitative research, even if only or mainly focus groups; the Burke Institute and the A.C. Nielsen Center for Marketing Research expose their students to
the topic. Perhaps another sign of the degree of recognition and acceptance that has occurred are publications that reflect back on particular contributors and their work. Along the way, a notable example was Elizabeth Hirschman’s (1985) analysis of Scientific Style and the Conduct of Consumer Research. More recently, Hope Schau (1998) discussed the character of Russell Belk’s ideas, Stephen Brown (1999) compared Morris Holbrook’s thinking and style with Theodore Levitt’s, and Sage Publications has a forthcoming volume collecting 50 years of my writing, edited by Dennis Rook (1999). Alain Jolibert is planning to edit a French volume of the intellectual biographies of 11 contributors to marketing thought. However, looking back and summing up in this field are not far advanced and there is a need for more grand integrations and overviews. A few signs of such maturation are the volumes by McCracken (1988), Hirschman and Holbrook (1992), and by Firat and Dholakia (1998). Otherwise, the closest we come is with reviews of the literature on focal topics in single articles, in proceedings, and in edited collections of chapters written for books such as John Sherry’s (1995, 1998) Contemporary Marketing and Consumer Behavior and his ServiceScapes. Textbooks serve this summary function by harvesting the findings of many individual studies, although the applied orientation of textbooks is usually so great that considerations of theory and method are less apparent. As academic consumer research studies were gathering steam in the 1960s, the landmark text by Engel, Kollat, and Blackwell that emerged in 1968 organized its contents by using the common categories that continue today. These are general cultural and environmental forces, reference group memberships, and psychological components that affect consumer decision-making. The early books relied a lot on fashionable theoretical currents, such as personality and learning theory, cognitive dissonance and its reduction, involvement, and social stratification. In a doctoral seminar I taught in 1977, Kenneth Wisniewski wrote a paper on the diffusion of innovation and Deborah RoedderJohn wrote one on perceived risk; both papers suggested that the concepts seemed to have been largely exhausted. Two outstanding books provided overviews of the basic theoretical resources available to researchers: Behavioral Science Foundations of Consumer Behavior, edited by Joel Cohen (1972), and Consumer Behavior: Theoretical Sources, edited by Ward and Robertson (1973). Early textbooks often included the identical material and anthologies collected many of the same articles, but in time, there was a proliferation of studies, with fresh and diverse examples. The goals of communicating research findings, assessing them and applying them, were emphasized by Engel et al., but in their sixth edition in 1990, they added this fifth objective: To make the field of consumer behavior exciting, interesting, and relevant to both students and faculty. With advanced textbook technology, the modern consumer behavior books are gorgeous with their varied fonts, colors, pictures, charts, and boxes. Along this line, the
S.J. Levy / Journal of Business Research 58 (2005) 341–347
volume by Michael R. Solomon (1992, 1st ed.) is remarkable for the lavishness and variety of its illustrations and its comprehensive coverage. Texts on advertising relate a lot to consumer behavior and show similar opulence of design. The effect is sometimes broken up and distracting, probably in keeping with the fragmentation characteristic of postmodern organization and the students’ experience with MTV, the Internet, and the provocative visualizations in the fast moving world of special effects.
345
sumers as qualifying the theories of economists and psychologists by showing the role of consumer behavior under certain conditions, such as branding and variety seeking. 4.2. Too much psychology Another respondent criticizes the consumer research orientation as overly dominated by psychology and idealizes the earlier days as more comfortable with qualitative research, which was hardly the case; but lends support to my perception of the viciousness that still goes on.
4. A little study Recent and contemporary qualitative consumer research has clearly made a place for itself and contributed to a great richness of detail in exploring consumer situations. It is so varied and ramified, it would be difficult to enumerate here all that has specifically been learned. Rather than do that, I will be a dutiful qualitative researcher and report on a modest research project I carried out in preparation for this article. To share the challenge of the assignment, I asked a few outstanding individuals in the consumer research field what they regarded as important contributions. Although I asked for accomplishments and what we have learned, I got some negative views as well. I will quote the most salient comments. 4.1. The dark side Well, here’s a dark side point of view. You might not want to hear this, but one thing I’ve been thinking is that marketing, at least the part about ‘‘listening to the customer,’’ has been so enormously successful that we only have ourselves to blame for the current situation we’re in. What do I mean? I think marketing helped to bring about postmodernism, where everyone thinks his or her own point of view is equally valid (cf. ‘‘the customer is always right’’); that diversity is more important than quality, that quality is not a truth but an opinion. . .Rather than pursuing knowledge and beauty and goodness for all, we have numerous tiny special groups ranting ‘‘listen to us, we have special needs’’ as if we all do not have needs, and as if those needs don’t turn out to be rather common. This respondent is pointing out that research approaches have philosophical implications, that the emphasis on phenomenology that underlies the famous marketing concept is overdone and has adverse consequences. She points to the contemporary conflict of the traditional pursuit of absolutes, such as truth and virtue, versus the rampant individualism and relativism some people find so destructive. Her thought echoes the criticism of the broadening concept of marketing specified by Laczniak and Michie (1979) as promoting social disorder. In additional remarks, she continues in a more positive vein to see the benefit of marketing study of con-
A fascinating element of consumer research is that, since consumers are people and being involved in a consumption or exchange situation requires interacting with other people, our discipline should always have identified with all of the social sciences. My perception is that it did—in the early stages of development when scholars were well versed in a variety of disciplines. But, during the years that marketing departments responded to the pressures of business schools needing to develop an image of being ‘‘research-oriented,’’ our discipline had to pick one science to pattern itself after, and, for reasons I’m not sure about, that discipline was psychology. We seem to have been struggling ever since to return to a world that is more reflective of the complexities of consumers, consumption and exchange processes, but the struggle has really hurt the discipline as a whole. . .My perception is that the field was very comfortable with qualitative methods early in its development, but along the way, someone decided that for either speed of analysis or rigor we needed to move to surveys, with scales to complete, or experiments with lots of control. Then, people got interested in qualitative methods again, but only as defined by the rigor identified as ethnography. As each wave builds, in order to defend the value of whichever method someone prefers, everyone seems to have criticized the other methods or worse, assert that the other methods are of no value. The academic world has been most vicious about this. 4.3. Going to extremes This respondent also thinks that current study has been distracted away from the middle road of inquiry by being either too abstract and unrealistic or too concrete and descriptive. We have, as academics tended to study things that are totally abstracted from reality in order to develop control in the research process (almost all experimental research until the past few years) or we have moved to observing only real world events/contexts (rafting on a river, watching a specific movie or baseball team) which are, by definition, so specific as to provide little room for generalization.
346
S.J. Levy / Journal of Business Research 58 (2005) 341–347
4.4. Interpreting patterns This next respondent sees development toward studying complexity and recognizes that the main merit of qualitative study is the field approach. There the goal is to explain the nature of situations, learning their characteristics and ‘‘deeper’’ relationships rather than, or in addition to, surveying (measuring and correlating) the frequency of subjects’ verbalizations. When Engel, Kollat, and Blackwell first came out, there was significant reliance on survey research which analyzed patterns in what people told us about themselves. But now, the focus is on mathematical models, experiments, and ethnographic observations to find patterns that consumers are not able to articulate about themselves. For example, Grant McCracken’s article on ‘‘Homeyness’’ dug into people’s home furnishings to lay out the characteristics that contribute to making an environment homey: layering, embracing, variable, engaging, mnemonic, etc.. . .there is much greater awareness of and healthy questioning of the scientific assumptions made in pursuit of that contribution to knowledge. We now acknowledge that we all start out by making particular assumptions, and then proceed to see what new contribution can be made after that. . .and give more attention to finding tensions that consumers operate within rather than point estimate of their preferences. . .some people call this postmodern, others just call it appreciating complexity. 4.5. Cycles, extending methods, and theories Another respondent associates freely to the evolution of consumer research, noting its cyclical character and its increasingly wide range of theorizing, use of various methods of research, and diversity of topics. Some study is seen as oriented to capturing consumers’ realities, while others’ look for underlying or more basic processes. A stream of consciousness response: I discovered that all of the hoo-hah about product symbolism some of us stumbled upon in the 1980s had been written about in the 1950s by yourself and a few others. . .perhaps the true hallmark of consumer behavior research is rediscovering the wheel??. . .Haire’s shopping list methodology. . . shopping/farmers’ market, etc., and ethnography (Odyssey) to capture phenomenology and real world experiences of consumers. . .methodologies ranging from the information display board to autodriving that capture the predominance of the visual channel. . .constructive memory processes (Bettman) and prospect theory. . . the symbolic renaissance of the 1980s. . .deeper descriptions: cultural meaning transfer (McCracken), aesthetics/ experiential consumption (Holbrook), extended self (Belk), rituals and structuralism (Rook and Levy)
. . .attitude models (Fishbein). . .deep study of subcultures, (Pen˜aloza) Hispanic immigrants, (Schouten) Harley riders. . .etc. 4.6. Fleshing out the details The final respondent provides a relatively detached summary that implies the big ideas have all been had, but are still guiding specific studies. The major contribution of the ’60s were the comprehensive models such as Howard and Sheth. They identified the basic influences and processes involved in CB. I don’t think much has been added in terms of scope. Since this time, we have been fleshing out the details. . .Attitude research of the 70s provided an important understanding of belief-based attitudes, a framework still useful today. . .Social influence work of the 70s. . .the human information processing paradigm. . .early research strictly cognitive. . .expanded to include emotional responses, low-involvement decision making and judgment, and even non-conscious processes. . .‘‘revolt’’ of the late 80s, most notably, the Odyssey project re-introduced the macro perspectives that had been part of the comprehensive models of the 1960s. . .focusing on the meanings of brands and consumption. . .included symbolic, ritual, emotional, social, and sensation aspects of consumption. . .qualitative methods used in sociology and anthropology. The respondents’ associations show the sense of change and developments over time. They recognize the richness and variety of the field of consumer research, and they show some sharp individual differences in awareness and evaluation of what has been achieved. Their views reflect the coming and going of qualitative research as it has contended for its place in the sun of the research world. In my view, having been part of this intellectual fray and survived it for over 50 years, I must say it has been enjoyable despite the many anxious moments in trying to cope with dubious clients, skeptical students, rigid reviewers, and supercilious colleagues. I am grateful that our heritage is sometimes funny, even ridiculous, so that we can laugh rather than cry about it. I am grateful to know and to have known so many of the participants— especially the really bright ones—who have collectively created and continue to work on our great common endeavor. References Berger B, editor. Authors of their own lives: intellectual autobiographies by twenty American sociologists. Berkeley (CA): University of California Press; 1990. Brown S. Marketing and literature: the anxiety of academic influence. J Mark 1999;63(January):1 – 15.
S.J. Levy / Journal of Business Research 58 (2005) 341–347 Clark LH, editor. Consumer behavior, vols. I – II. New York: New York Univ. Press; 1954 – 1955. Cohen JB, editor. Behavioral science foundations of consumer behavior. New York: Free Press; 1972. Fine GA. A second Chicago school? Chicago (IL): University of Chicago Press; 1995. Firat AF, Dholakia N. Consuming people: from political economy to theaters of consumption. New York: Routledge; 1998. Hirschman EC. Scientific style and the conduct of consumer research. J Consum Res 1985;12(September):225 – 39. Hirschman EC, Holbrook MB. Postmodern consumer research. Thousand Oaks (CA): Sage Publications; 1992. Hovland CI, Janis IL, Kelley HH. Communication and persuasion. New Haven (CT): Yale Univ. Press; 1953. Hunt S. Modern marketing theory: critical issues in the philosophy of marketing science. Cincinnati (OH): Southwestern; 1991. Kassarjian HH. Scholarly traditions and European roots of American consumer research. In: Laurent G, et al., editors. Research traditions in marketing. Boston (MA): Kluwer Academic Publishers; 1994. p. 265 – 79. Klapper JT. The effects of mass communication. New York: Free Press; 1960. Laczniak GR, Michie DA. The social disorder of the broadened concept of marketing. J Am Acad Mark Sci 1979;7, 3(Summer):214 – 32. Laurent G, Lilien GL, Pras B. Research traditions in marketing. Boston (MA): Kluwer Academic Publishers; 1994. Lazarsfeld P. Radio and the printed page. New York: Duell, Sloan and Pearce; 1940. Levy SJ. President’s column: a brief history. Assoc Consum Res Newsl (March) 1991;2 – 6. Levy SJ. Commentary by Sidney J. Levy. In: Laurent G, et al., editors. Research traditions in marketing. Boston (MA): Kluwer Academic Publishers; 1994. p. 283 – 7. Levy SJ. Stalking the amphisbaena. J Consum Res 1996;23, 3(December): 163 – 76. Lewin K. The conflict between Aristotelian and Galileian modes of thought
347
in contemporary psychology. A dynamic theory of personality. New York: McGraw-Hill; 1935. p. 1 – 42. McCracken G. Culture and consumption. Bloomington (IN): Indiana Univ. Press; 1988. Morgan G, Smircich L. The case for qualitative research. Acad Manage Rev 1980;5(October):491 – 500. Piaget J. The language and thought of the child. New York: Harcourt Brace; 1926. Rook D, editor. Brands, consumers, symbols, and research: Sidney J. Levy on marketing. Thousand Oaks (CA): Sage Publications; 1999. Schau HJ. Discourse of possessions: the metatheory of Russell W. Belk. In: Alba JW, Hutchinson JW, editors. Adv Consum Res, vol. XXV; 1998. p. 37 – 44. Sherry Jr JF, editor. Contemporary marketing and consumer behavior: an anthropological sourcebook. Thousand Oaks, CA: Sage Publications; 1995. Sherry Jr JF, editor. ServiceScapes. Chicago (IL): NTC Business Books; 1998. Solomon M. Consumer behavior. Boston (MA): Allyn & Bacon; 1992. Wall Street Journal. But how does it make you feel? 1999;B1 [May 3]. Ward S, Robertson T, editors. Consumer behavior: theoretical sources. Englewood Cliffs (NJ): Prentice-Hall; 1973.
Further reading Merton R. Mass persuasion. New York: Harper & Brothers; 1946. Murray HA, et al. Techniques for a systematic investigation of fantasy. New York: Oxford Univ. Press; 1937. Warner WL, Henry WE. The radio day time serial: a symbolic analysis. Genet Psychol Monogr 1948;37:3 – 71. Warner WL, Meeker M, Eells K. Social class in America. Chicago (IL): Science Research Associates; 1949. p. 273. Wilder FF. Radio’s daytime serial [pamphlet]. New York: Columbia Broadcasting System; 1945.
Journal of Business Research 58 (2005) 348 – 353
Response construction in consumer behavior research$ Robert A. Peterson* Department of Marketing, Mc Combs School of Business, The University of Texas, Austin, TX 78712, USA
Abstract To date, researchers have been relatively unsuccessful in accounting for a substantial proportion of the variance in the measures of consumer behavior that have been investigated. It is posited here that one of the primary reasons for this lack of success is that most studies of consumer behavior use self-reports—answers or responses to research questions—that are often very labile. It is further posited that responses to research questions are not generally revealed (retrieved directly from memory) but rather are constructed at the time a question is asked and answered. Because they are derived from processes that are inherently constructive, self-reports are susceptible to a variety of contaminating influences that collectively constrain the ability of researchers to explain or predict consumer behavior. Several suggestions are offered for addressing response construction processes and their effects. D 2003 Elsevier Inc. All rights reserved. Keywords: Consumer behavior; Response construction; Consumer behavior research
1. Introduction More than two decades ago, Jacoby (1978) authored a scathing ‘‘state-of-the-art review’’ of consumer behavior research. Jacoby began his review, which coincidentally received the prestigious Harold H. Maynard Award from the American Marketing Association for its contribution to marketing theory, by stating that ‘‘too large a proportion of the consumer (including marketing) research literature is not worth the paper it is printed on or the time it takes to read’’ (p. 87). A major theme throughout his review was that researchers had produced relatively little substantive knowledge of consumer behavior. If ‘‘substantive knowledge’’ can be equated with ‘‘variance accounted for,’’ it would appear that Jacoby (1978) was correct in his assessment of consumer behavior knowledge produced. Following an analysis of 70 different behavioral data sets (including but not limited to consumer behavior data sets), Cote and Buckley (1987) found that, of the variance accounted for in a variety of construct measures, less than 42% was due to the traits studied; the
$ An earlier version of this article was presented at the 1999 Marketing Communications and Consumer Behavior Seminar in La Londe les Maures, France. Appreciation is expressed to Gerald Albaum, Steven P. Brown, and Wayne D. Hoyer for their comments on a preliminary draft of the article. * Tel.: +1-512-471-9438. E-mail address: [email protected] (R.A. Peterson).
0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00100-0
remainder was accounted for by the methods employed and by error. Although Peterson and Jolibert’s (1995) meta-analysis of 1520 country-of-origin effects revealed that, on average, the presence of a country-of-origin cue accounted for 26% of the variance in perceptions and purchase intentions, their results appear to be somewhat of an anomaly. In a general meta-analysis of the variance accounted for in consumer behavior experiments over the time period 1970 – 1982 (which included the publication of Jacoby’s 1978 review), Peterson et al. (1985) found that, across 118 independent experiments containing 1036 effects, on average, 5% of the variance in the dependent variables was accounted for by experimental manipulations. An identical percentage was obtained by Wilson and Sherrell (1993) in their metaanalysis of the effect of message source manipulation on persuasibility. More recently, a meta-analysis of 580 surveybased regression analyses (a majority of which were carried out in consumer behavior studies) conducted by the author covering the period 1964 – 1994 revealed that, on average, the variance accounted for in a dependent variable by an independent variable was slightly less than 1% (the average zero-order correlation coefficient was .08). Thus, in general, the amount of variance accounted for in measures of consumer behavior would seem to be relatively minor. A question then arises as to why this is so. The amount of variance accounted for is a function of many factors, including the theory employed (or the lack of
R.A. Peterson / Journal of Business Research 58 (2005) 348–353
theory), the research procedures and techniques utilized, the individuals and populations studied, and even the phenomena and constructs investigated. However, there is another plausible explanation for the minimal variance typically accounted for in consumer behavior research. This explanation is based on the data collected and analyzed in consumer behavior research. 1.1. Reliance on self-report data Much of what is known about consumer behavior, and, indeed, human behavior in general, is based on self-reports or, more generally, answers or responses to questions (e.g., Peterson and Kerin, 1981). While the theories, research procedures and techniques, individuals and populations studied, and phenomena and constructs investigated vary greatly in consumer behavior research, a common thread is the use of self-report data. To illustrate, examination of the articles contained in the Journal of Consumer Research since its inception reveals that fully 73% have relied on self-report data for their conclusions. Self-report data are usually accepted at ‘‘face value,’’ regardless of their source (oftentimes college students) and despite the fact that there is virtual consensus as to their fallibility. One need only peruse such classic articles as ‘‘Telling More Than We Can Know: Verbal Reports on Mental Processes’’ (Nisbett and Wilson, 1977) or ‘‘Verbal Reports as Data’’ (Ericsson and Simon, 1980) to appreciate the difficulties that can arise when using self-report data in consumer behavior research. A possible reason for the fallibility of self-report data relates to the notion that answers to research questions are constructed as they are elicited. This notion is the subject of this article.
349
were no more than chaotic swirls of thought.’’ More recently, in the context of attitude measurement, Wilson and Hodges (1992) concluded that attitudes are ‘‘temporary constructs’’ formed at the moment a question is asked and answered. This opinion is consistent with that of Zaller and Feldman (1992, p. 582), who wrote that ‘‘people do not merely reveal preexisting attitudes on surveys; to some considerable extent, people are using the questionnaire to decide what their ‘attitudes’ are’’ (i.e., they construct their attitudes and question answers simultaneously). Even more recently, Schwarz et al. (1998, p. 150) observed that participants in a survey answering an attitude question . . .may either retrieve a previously formed attitude judgment from memory, or they may ‘‘compute’’ [emphasis added] a judgment [response] on the spot.. . .[To do so] they will also need to retrieve or construct [emphasis added] some standard against which the target is evaluated. In brief, although the ‘‘traditional view [of the attitude response process] holds that evaluative responses are cognitively represented in memory and may be directly (and automatically) activated’’ (Lavine et al., 1998, p. 359), there is an increasing belief among researchers that many attitudes are temporary constructions. Consequently, to the extent that attitudes are constructed, as seems to be the case for a variety of attitudes, answers to questions reporting attitudes must also be constructed (e.g., Lavine et al., 1998; Tourangeau and Rasinski, 1988; Zaller and Feldman, 1992). There is no logical alternative to response construction in such a case. 2.1. Response construction is pervasive
2. Responses are constructed The fundamental premise of this article is that the selfreport data employed in consumer behavior research consist of answers that have been constructed as responses to questions. Although answers to certain questions (‘‘Are you male or female?’’) may almost be reflexive, it is proposed that, in general, even the answers to factual questions, such those asking about age and income, are constructed ‘‘on-line,’’ ‘‘on-the-fly,’’ or in ‘‘real time’’ when the questions are asked. Rather than simply retrieving a response from memory—merely recalling information from memory or revealing a psychological characteristic and reporting that information or characteristic in response to a question, it is hypothesized that consumers generally construct their response when answering a question. The notion that responses are subject to inherently constructive cognitive processes is neither new nor novel. For example, more than 30 years ago, Bogart (1967, p. 335) wrote that sometimes asking a question ‘‘forces the crystallization and expression of opinions where [previously] there
This article goes beyond responses to attitude questions in that it argues that responses to most questions are constructed, not only those to attitude questions. As such, the present position is consistent with that of Schwarz et al. (1998, p. 150), who observed that if study participants are answering a question about behavior, they ‘‘need to recall or reconstruct [emphasis added] relevant instances of this behavior from memory.’’ Even questions that intuitively would be expected to be answered through direct recall, such as questions about past behaviors, generally require that responses be constructed (e.g., Blair and Burton, 1987; Menon, 1997). Note, however, that no argument is being put forth here as to the specific psychological processes underlying response construction. Such an argument is beyond the purpose and scope of this article (and any attempt at exposition would greatly exceed the current page constraint). Although it is argued that responses to questions are constructed, construction is perhaps best construed as existing as a continuous phenomenon that can take several different forms and occur at various stages of the question-and-answer process. Semantics aside, the extent to
350
R.A. Peterson / Journal of Business Research 58 (2005) 348–353
which response construction occurs is a matter of degree and conditional on several factors, including the phenomenon or construct being investigated and the accessibility and diagnosticity of information about it residing in memory. Succinctly stated, response construction can vary from complete response fabrication to a minor coalescing of stored information or tweaking the words contained in an answer. Moreover, response construction can be conscious or unconscious, and many memory principles can be invoked when attempting to explain or predict it. Consider a question for which no information is available in memory on which to form an answer. By definition, any response must be constructed. Thus, answers to a question asking about a fictitious issue or entity, such as the ‘‘Agricultural Trade Act of 1983’’ (Bishop et al., 1986), the ‘‘Texas state law regulating television news broadcasting’’ (Peterson, 2000, Fig. 3.1), or bogus (nonexistent) advertising slogans (Glassman and Ford, 1988) can only result from constructive processes. Consider further a question for which relevant information is not readily available in memory or, if available, needs to be aggregated or coalesced prior to an answer being formulated. An example might be a question that inquires about a consumer’s satisfaction with an automobile purchased 3 years ago, since satisfaction with the automobile is not likely to be directly or immediately accessible. Another example might be a question about intention to purchase a new refrigerator or expectations regarding the attributes of a new refrigerator. Unless the consumers questioned are experiencing problems with their current refrigerator, contemplating remodeling their kitchen, planning to purchase a refrigerator as a gift, or the like, they probably have not thought about purchasing a new refrigerator prior to being asked about doing so. Again, constructive processes must logically occur before such questions can be answered. Although the evidence supporting the notion of response construction is primarily indirect and somewhat circumstantial, it is compelling. Rejecting the notion of response construction requires the assumption that all information necessary to answer any question is capable of being (and, indeed, has been) encoded into memory, stored in memory, and recalled from memory without error. There is a substantial body of research in psychology, however, that refutes this assumption. Further, the assumption implies, in part, that the answer to any question is unaffected by, or independent of, the manner and context in which the question is asked. With few exceptions, though, the existing research evidence indicates that the manner and context in which a question is asked influences its answer.
identifiable process (cf. Fischoff, 1991). Consequently, there is a vast literature that can be brought to bear on the topic of response construction. Almost any research finding that demonstrates how a question answer or a judgment can be influenced or manipulated can be interpreted as evidence of response construction. Only if a question answer or a judgment is not subject to being influenced or manipulated can an argument logically be made for direct response retrieval. Consider just the following examples of research providing support for the notion of response construction. (Numerous other examples exist; the present ones were selected simply because they were readily available.) 3.1. Methodological research There is a plethora of methodological research on designing and administering questionnaires that suggest that the form, manner, and context in which a question is asked to a large extent determines the answer to it. Changing a single word (e.g., Gannon and Ostrom, 1996; Loftus and Zanni, 1975) or phrase (e.g., Levine, 1997) in a question has been shown to produce very different responses. Indeed, the accuracy of answers to even straightforward, factual (objective) questions, such as age (Peterson, 1984) and household income (Peterson and Kerin, 1980), has been shown to vary as a function of question format. Perhaps the most convincing evidence of response construction from the methodological literature relates to question order. It has been consistently documented that, under certain conditions, ‘‘people construct answers to survey questions on the basis of their responses to earlier items’’ (Simmons et al., 1993, p. 316). This finding has variously been termed an ‘‘assimilation – contrast effect’’ (e.g., Manis et al., 1988), a ‘‘part – whole’’ effect (e.g., Mason et al., 1994), and a ‘‘carryover – backfire’’ effect (e.g., Bickart, 1993). In general, to the extent that a response to a question is accessible and diagnostic, it may be used to construct responses to subsequent questions (e.g., Feldman and Lynch, 1988). Because the methodological literature is so diverse and voluminous, no attempt is made to summarize it. One needs to merely peruse such methodologically oriented journals as Public Opinion Quarterly or such compilations as Autobiographical Memory and the Validity of Retrospective Reports (Schwarz and Sudman, 1994) to be exposed to methodological research on source effects, question-order effects, mode-of-administration effects, and type-of-question effects, among others. All of these effects can be interpreted as inducing as well as influencing responses. Thus, they support the notion of response construction. Schwarz (1999) presents a concise yet enlightening review of recent research on these various effects.
3. Support for response construction 3.2. Context effects research Response construction is a ubiquitous cognitive phenomenon evidenced throughout the behavioral literature, although it has not traditionally been recognized as an
In a classic series of studies demonstrating group conformity, Asch (1940, 1956) asked people to judge the
R.A. Peterson / Journal of Business Research 58 (2005) 348–353
relative lengths of lines in a display. Although the lines were discernibly different in length, Asch was able to manipulate the responses of target individuals in a group by having research confederates in the group report judgments that were obviously in error. If the responses of the target individuals were not constructed but represented independent judgments, they should not have been subject to manipulation. Asch’s work is important because it represents one of the first empirical demonstrations of the influence of context on answers to research questions. In general, the psychophysics and social psychology literatures posit that the context within which a response is generated can markedly influence that response. As Stapel and Winkielman (1998, p. 634) recently pointed out: There is no such thing as a context-free judgment.. . .The context in which a target stimulus is embedded provides a frame of reference for interpretation and judgment. Hence, the same target can be associated with different responses depending on the context in which it is judged. Researchers (e.g., Birnbaum, 1982; Parducci, 1965; Volkman, 1951) have consistently documented that ratings of the size, weight, or shape of a particular stimulus object in a stimulus set are partially a function of the sizes, weights, or shapes of other stimulus objects in the set. Virtually identical effects have been observed for social judgments as well. In a widely cited study, Pepitone and DiNubile (1976) showed that a homicide was rated as a less severe crime if the rating was preceded by the evaluation of another homicide than if the rating was preceded by the evaluation of an assault. Similarly, Kendrick and Gutierres (1980) demonstrated that ratings of female attractiveness were influenced by prior exposure to extremely attractive females, and Biernat et al. (1997) found that explicit self-ratings influenced subsequent ratings of others. Analogous to Asch’s findings, the above findings lend themselves to a response construction interpretation. If responses (ratings) are not constructed, they should not be subject to influence through a context manipulation. 3.3. False memory research A general theme of current memory research is the fallibility of memory. There is a voluminous yet oftentimes contentious literature on ‘‘constructive memory,’’ ‘‘recovered memory,’’ ‘‘false memory,’’ ‘‘reconstructive memory,’’ and ‘‘confabricated memory.’’ One need only peruse tomes such as Recovered Memories and False Memories (Conway, 1997) or The Recovered Memory/False Memory Debate (Pezdek and Banks, 1996) to become aware of the issues being addressed in research on human memory. Schacter (1999) summarized much of the recent research on memory fallibility in an article titled ‘‘The Seven Sins of Memory.’’ Of interest here is his review of suggestibility
351
(the fifth ‘‘sin’’), the idea that false memories can be created or memories ‘‘implanted’’ as a result of a questioning process. Research on suggestibility supports the proposition that responses are constructed in memory, not retrieved from memory. Several studies are presented to illustrate this distinction. Loftus (1993), in the first of several studies, reported the results of asking an individual to remember the time he had been lost in a shopping mall at age five. Despite the fact that the individual had never actually been lost in a shopping mall, the individual produced a detailed description of the event. Kassin and Kiechel (1996) had subjects perform a computer-based reaction time test. Subjects were told not to press the ‘‘ALT’’ key because doing so would cause the computer to crash. Although no subjects actually pressed the ALT key, a ‘‘witness’’ reported that they did, and ultimately almost 70% of the subjects falsely confessed. Hyman and Billings (1998) conducted a study similar to the one reported by Loftus in which a significant percentage of the undergraduate college students surveyed produced a false memory of a childhood event. What was especially interesting in this study was that the false memories became more vivid and detailed the second time they were elicited. Related research (e.g., Levine, 1997; Markus, 1986) has documented that memories of prior events or even emotions tend to display a consistency bias in that ‘‘recollections tend to exaggerate the consistency between [people’s] past and present attitudes, beliefs, and feelings’’ (Schacter, 1999, p. 193). At the present time, memory research suggests that certain false memories can be created in many people in particular situations. (It is important to note the emphasized adjectives. The creation of false memories appears to be conditional on a variety of factors, including gender, age, nature and intensity of the topic, and so forth.) This research contributes a theoretical foundation for guiding future research on response construction as well as convincingly documents that answers to questions can be completely fabricated, whether consciously or unconsciously. 3.4. Information processing research Research conducted on the construction of preferences, choices, judgments, beliefs, or the like (often referred to as constructive information processing research) offers useful insights into response construction. This is because, in one sense, the construction of preferences, choices, or the like is directly reflected by responses to questions about them. For example, in referencing the work of Slovic et al. (1990), Payne et al. (1992, p. 89) stated that ‘‘preferences for and beliefs about objects or events of any complexity are often constructed—not merely revealed—in the generation of a response to a judgment or choice task.’’ Therefore, if preferences, choices, judgments, or beliefs are constructed at the moment of elicitation (Slovic, 1995), then corresponding or consequent answers must also be constructed to the eliciting questions. (Note that exceptions might include answers to
352
R.A. Peterson / Journal of Business Research 58 (2005) 348–353
questions about established, noncomplex habitual behaviors or long-standing, elementary preferences.) If so, the organizing framework proposed by Bettman et al. (1998) for investigating constructive processes may prove useful when investigating the construction of responses.
4. Discussion and conclusion The objective of this article is really quite modest. The article simply represents an attempt to focus attention on one of the reasons researchers have been relatively unsuccessful in precisely explaining or consistently predicting various consumer behavior phenomena. In particular, the article posits that, with rare exceptions, answers to questions asked in consumer behavior research studies (and, indeed, in behavioral studies generally) tend to be constructed rather than revealed or directly retrieved from memory. If this article does nothing more than alert producers and interpreters of consumer behavior research to the existence and implications of response construction, it will be considered a successful endeavor. Response construction is analogous to other cognitive phenomena in that it is not directly observable. There is, however, substantial indirect evidence that most question responses are indeed constructed. Consequently, the notion of response construction has major implications for consumer behavior research. An obvious implication is that researchers should not generally expect to account for, say, a majority of the variance in the measures of consumer behavior investigated. (Although an argument might be made that research accounting for a large proportion of the variance in measures of consumer behavior is either trivial or the findings obvious and therefore uninteresting, such an argument ignores the reality of response construction as a determinant of a study outcome.) Thus, for example, assessments of the reliability of a multi-item scale should take into account the constructive nature of responses to individual scale items. In light of the existence of response construction, very high internal consistency scale reliability estimates (those close to 1.0) should be viewed with suspicion, not awe, and what is an ‘‘acceptable’’ reliability coefficient should probably be rethought. This is particularly so for the constructs frequently studied in consumer behavior research—attitudes, emotions, attributions, and so forth. Even as theories of consumer behavior and research methodologies become more sophisticated and ever more capable of finer distinctions, as long as self-reports or responses to questions serve as the primary source of information about consumer behavior, a lacuna in knowledge will remain. Because responses are constructed, they are susceptible to a myriad of contaminating influences (many of which are likely yet undiscovered). What is needed is systematic research, conceptual as well as empirical, that will at least establish the general boundary con-
ditions of response construction. In this regard, Hilton (1995) forcefully argues that a ‘‘constructionist perspective’’ specifically requires an explicit recognition of the role of context when obtaining verbal reports. Moreover, because question responses are susceptible to contaminating influences, increased attention should be given to these influences when eliciting responses. Systematic research along the lines of that conducted by Blair and Burton (1987) and Menon (1997) on asking behavioral frequency questions is needed across a variety of consumer behavior phenomena. Incorporating experimental treatments into a research investigation to evaluate the possible effects of questionnaire structure, mode of data collection, and question form on responses should become standard practice. For instance, employing different questionnaire versions as treatments should be routine in basic research. One benefit of such a practice is that it permits both a practical and a statistical assessment of the stability of the phenomenon or construct being investigated. Fortunately, if recent consumer behavior and marketing research publications are an indication of an incipient trend, there is a renewal of interest in topics that can be subsumed under the rubric of response construction. A second implication of response construction is that researchers need to incorporate measures other than selfreports of consumer behavior phenomena into their investigations. Recently, Winer (1999) has argued in favor of an increased use of scanner data in consumer behavior research to corroborate findings based on self-reports. Other researchers have espoused the use of nonverbal data when studying consumer behavior. Although it is not clear exactly what nonverbal data should be collected and analyzed, attempts should be made to substitute, supplement, and complement self-report data in future investigations. For example, Peterson et al. (1995) studied the effects of nonverbal voice characteristics, such as rate of speaking, average pause duration, and fundamental frequency contour, as complements to self-reports. Response construction, as both a reflection of and a consequence of cognitive information processing, is worthy of investigation in and of itself, not as merely a contributor to measurement unreliability or as a methodological artifact. The notion of response construction should minimally serve as the basis of a framework that will facilitate the organization, integration, and application of methodological as well as substantive findings from numerous and disparate disciplines. Until that happens, the explanatory and predictive power of consumer behavior research will remain low.
References Asch SE. Studies in the principles of judgments and attitudes: II. Determination of judgments by group and ego standards. J Soc Psychol 1940;12(November):433 – 65. Asch SE. Studies of independence and conformity: a minority of one against a unanimous majority. Psychol Monogr 1956;70.
R.A. Peterson / Journal of Business Research 58 (2005) 348–353 Bettman JR, Luce MF, Payne JW. Constructive consumer choice processes. J Consum Res 1998;25(December):187 – 217. Bickart BA. Carryover and backfire effects in marketing research. J Mark Res 1993;30(February):52 – 62. Biernat M, Manis M, Kobrynowicz D. Simultaneous assimilation and contrast effects in judgments of self and others. J Pers Soc Psychol 1997;73(August):254 – 69. Birnbaum MH. Controversies in psychological measurement. In: Wegener B, editor. Social attitudes and psychophysical measurement. Hillsdale (NJ): Erlbaum; 1982. p. 401 – 85. Bishop GF, Tuchfarber AJ, Oldendick RW. Opinions on fictitious issues: the pressure to answer survey questions. Public Opin Q 1986;50(Summer): 240 – 50. Blair E, Burton S. Cognitive processes used by survey respondents to answer behavioral frequency questions. J Consum Res 1987;14(September): 280 – 8. Bogart L. No opinion, don’t know, and maybe no answer. Public Opin Q 1967;31(Fall):331 – 45. Conway MA, editor. Recovered memories and false memories. Oxford (UK): Oxford Univ Press; 1997. Cote JA, Buckley MR. Estimating trait, method, and error variance: generalizing across 70 construct validation scales. J Mark Res 1987; 24(August):118 – 315. Ericsson KA, Simon HA. Verbal reports as data. Psychol Rev 1980; 87(May):215 – 51. Feldman JM, Lynch Jr JG. Self-generated validity and other effects of measurement on belief, attitude, intention, and behavior. J Appl Psychol 1988;73(August):421 – 35. Fischoff B. Value elicitation. Am Psychol 1991;46(August):835 – 47. Gannon KM, Ostrom TM. How meaning is given to rating scales: the effects of response language on category activation. J Exp Soc Psychol 1996;32(July):337 – 60. Glassman M, Ford JB. An empirical investigation of bogus recall. J Acad Mark Sci 1988;16(Fall):38 – 41. Hilton DJ. Constructive processes in judgment and decision-making: implications for consumer behavior. In: Jolibert A, Peterson RA, Strazzieri A, editors. Marketing communications and consumer behavior. Aix-enProvence (France): Universite d’Aix-Marseille; 1995. p. 292 – 311. Hyman Jr IE, Billings FJ. Individual differences and the creation of false childhood memories. Memory 1998;6(January):1 – 20. Jacoby J. Consumer research: a state of the art review. J Mark 1978; 42(April):87 – 96. Kassin SM, Kiechel KL. The social psychology of false confessions: compliance, internalization, and confabulation. Psychol Sci 1996;7(May): 125 – 8. Kendrick DT, Gutierres SE. Contrast effects and judgments of physical attractiveness: when beauty becomes a social problem. J Pers Soc Psychol 1980;38(January):131 – 40. Lavine H, Huff JW, Wagner SH, Sweeney D. The moderating influence of attitude strength on the susceptibility to context effects in attitude surveys. J Pers Soc Psychol 1998;75(August):359 – 73. Levine LJ. Reconstructing memory for emotions. J Exp Psychol Gen 1997; 126(June):165 – 77. Loftus EF. The reality of repressed memories. Am Psychol 1993;48(May): 518 – 37. Loftus EF, Zanni G. Eyewitness testimony: the influence of the wording of a question. Bull Psychon Soc 1975;5(January):86 – 8. Manis M, Nelson TE, Shedler J. Stereotypes and social judgment: extremity, assimilation, and contrast. J Pers Soc Psychol 1988;55(July):28 – 36. Markus GB. Stability and change in political attitudes: observed, recalled, and ‘‘explained’’. Polit Behav 1986;8(1):21 – 44. Mason R, Carlson JE, Tourangeau R. Contrast effects and subtraction in part – whole questions. Public Opin Q 1994;58(Winter):569 – 78. Menon G. Are the parts better than the whole? The effects of decompositional questions on judgments of frequent behaviors. J Mark Res 1997; 34(August):335 – 46.
353
Nisbett R, Wilson T. Telling more than we can know: verbal reports on mental processes. Psychol Rev 1977;84(May):231 – 59. Parducci A. Category judgment: a range – frequency model. Psychol Rev 1965;72(November):407 – 18. Payne JW, Bettman JR, Johnson EJ. Behavioral decision research: a constructive processing perspective. In: Rosenzweig MR, Porter LW, editors. Annual review of psychology, vol. 43. Palo Alto (CA): Annual Reviews; 1992. p. 87 – 131. Pepitone A, DiNubile M. Contrast effects in judgments of crime severity and the punishment of criminal violators. J Pers Soc Psychol 1976; 33(April):448 – 59. Peterson RA. Asking the age question: a research note. Public Opin Q 1984;48(Spring):379 – 83. Peterson RA. Constructing effective questionnaires. Beverly Hills (CA): Sage Publications; 2000. Peterson RA, Jolibert AJP. A meta-analysis of country-of-origin effects. J Int Bus Stud 1995;26(4):883 – 900. Peterson RA, Kerin RA. Household income data reports in mail surveys. J Bus Res 1980;8(September):301 – 13. Peterson RA, Kerin RA. The quality of self-report data: review and synthesis. In: Enis BM, Roering KJ, editors. Review of marketing. Chicago (IL): American Marketing Association; 1981. p. 5 – 20. Peterson RA, Albaum G, Beltramini RF. A meta-analysis of effect sizes in consumer behavior experiments. J Consum Res 1985;12(June):97 – 103. Peterson RA, Cannito MP, Brown SP. An exploratory investigation of voice characteristics and selling effectiveness. J Pers Sell Sales Manage 1995;15(Winter):1 – 15. Pezdek K, Banks WP, editors. The recovered memory/false memory debate. San Diego (CA): Academic Press; 1996. Schacter DL. The seven sins of memory. Am Psychol 1999;54(March): 182 – 203. Schwarz N. Self-reports: how the questions shape the answers. Am Psychol 1999;54(February):93 – 105. Schwarz N, Sudman S, editors. Autobiographical memory and the validity of retrospective reports. New York: Springer-Verlag; 1994. Schwarz N, Groves RM, Schuman H. Survey methods. In: Gilbert DT, Fiske ST, Lindzey G, editors. The handbook of social psychology, vol. 1. New York: McGraw-Hill; 1998. p. 143 – 79. Simmons CJ, Bickart BA, Lynch Jr JG. Capturing and creating public opinion in survey research. J Consum Res 1993;20(September): 316 – 29. Slovic P. The construction of preference. Am Psychol 1995;50(May): 364 – 71. Slovic P, Griffin D, Tversky A. Compatibility effects in judgment and choice. In: Hogarth RM, editor. Insights in decision making: a tribute to Hillel J Einhorn. Chicago (IL): University of Chicago Press; 1990. p. 5 – 27. Stapel D, Winkielman P. Assimilation and contrast as a function of contexttarget similarity, distinctness, and dimensional relevance. Pers Soc Psychol Bull 1998;24(June):634 – 46. Tourangeau R, Rasinski KA. Cognitive processes underlying context effects in attitude measurement. Psychol Bull 1988;103(May):299 – 314. Volkman J. Scales of judgment and their implications for social psychology. In: Rohrer JH, Sherif M, editors. Social psychology at the crossroads. New York: Harper & Bros.; 1951. p. 273 – 94. Wilson TD, Hodges SD. Attitudes as temporary constructions. In: Martin LL, Tesser A, editors. The construction of social judgments. Hillsdale (NJ): Erlbaum; 1992. p. 37 – 65. Wilson EJ, Sherrell DL. Source effects in communication and persuasion research: a meta-analysis of effect size. J Acad Mark Sci 1993; 21(Spring):101 – 12. Winer RS. Experimentation in the 21st century: the importance of external validity. J Acad Mark Sci 1999;27(Summer):349 – 58. Zaller J, Feldman S. A simple theory of the survey response: answering questions versus revealing preferences. Am J Polit Sci 1992; 36(August):579 – 616.
Journal of Business Research 58 (2005) 354 – 360
Advertising repetition and quality perception Sridhar Moorthy*, Scott A. Hawkins Joseph L. Rotman School of Management, University of Toronto, 105 St. George Street, Toronto, Ontario, Canada M5S 3E6
Abstract Nelson [J. Polit. Econ. 78 (1970) 311; J. Polit. Econ. 81 (1974) 729; Nelson P. The economic value of advertising. In: Brozen Y, editor. Advertising and society. New York: New York Univ. Press, 1974. pp. 43 – 66] has argued that advertising spending is a signal of product quality for experience goods because consumers can rationally infer that high-quality products would advertise more than low-quality products. In this paper, we compare Nelson’s view of advertising with marketing views of advertising using ad repetition as a surrogate for ad spending. Our results show limited support for Nelson’s theory but substantial support for ad repetition influencing perceived quality through attitude toward the ad. D 2003 Elsevier Inc. All rights reserved. Keywords: Advertising; Spending; Quality
1. Introduction In a seminal paper nearly 30 years ago, Nelson (1970) proposed a distinction between two types of goods, search goods and experience goods, and offered a new theory of advertising based on that distinction. Search goods were defined as products whose quality consumers can verify before purchasing (e.g., clothing, furniture, and jewelry). Experience goods were defined as products whose quality the consumer cannot determine until after buying and experiencing the product (e.g., foods, books, and detergents). Nelson argued that advertising claims for experience goods are uninformative because consumers cannot verify such claims before purchasing the product. Advertising spending, however, is informative because consumers can rationally infer that products that advertise more are of higher quality than products that advertise less. By contrast, for search goods, advertising claims are informative, and no further information is needed from, or provided by, the spending level. Nelson’s view of advertising for experience goods is radically different from ‘‘the marketing view.’’ Whereas Nelson minimizes the importance of the ad itself—for example, it is not even necessary for the firm to advertise as long as it has other ways of ‘‘burning money’’—the
* Corresponding author. E-mail address: [email protected] (S. Moorthy). 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00108-5
marketing literature emphasizes it (Batra et al., 1996). In marketing, it is taken as a given that ad design and ad claims have significant effects on product perceptions— regardless of the nature of the product. In fact, the work of Hoch and Ha (1986) suggests that, if anything, advertising claims might be even more effective in shaping perceptions for experience goods because the experience itself is likely to be ambiguous. Marketing textbooks are replete with examples of how the same physical product (e.g., 7-Up) has been ‘‘positioned’’ in different ways through different ad campaigns. In this paper, we test the Nelson theory against the ‘‘marketing view.’’ We use the term ‘‘marketing view’’ to refer to three theories: learning, ad attitude, and mere exposure. Each of these theories points to a different type of advertising effect, and multiple effects may coexist in an ad campaign. So it is not a matter of picking one of these theories and ruling out the rest. Rather, because these theories as a group have much in common and contrast sharply with the Nelson view, we examine whether the data support the marketing view or the Nelson view. There are potentially many ways to test Nelson’s theory, depending on the myriad ways in which advertising moneys can be spent—repetition (frequency), reach, ‘‘large’’ versus ‘‘small’’ ads, color versus black-and-white ads, TV versus print, etc.—and whether the researcher uses observational or experimental methods. This paper uses experimental methods and advertising repetition as the spending variable.
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
Much of the existing literature uses observational data. The papers by Rotfeld and Rotzell (1976), Kotowitz and Mathewson (1986), Tellis and Fornell (1988), Davis et al. (1991), and Caves and Greene (1996) examine the correlation between some measure of objective quality (e.g., Consumer Reports rankings), or managers’ perceptions of their customer’s perceptions of quality (PIMS data), and advertising spending across brands. In general, these papers find limited support for Nelson’s theory. For example, Caves and Greene’s (1996) comprehensive study of 196 product categories concludes: ‘‘These results suggest that quality-signaling is not the function of most advertising of consumer goods.’’ In contrast to these generally negative results are the generally positive results in the experimental literature. Unlike the observational literature where the focus is on the advertising spending-objective quality correlation, here the focus is on the advertising spending-perceived quality correlation. Kirmani and Wright (1989) cue advertising spending via size, vehicle (reach), production elements, and frequency in some experiments, and in others advertising spending information is simply given to subjects. Their strongest results in support of Nelson are obtained with size, vehicle, production elements, and spending levels. With frequency, their results contradict Nelson: ad campaigns described as high frequency were found to correlate negatively with perceived quality. A common feature of the Kirmani and Wright (1989) experiments is that subjects are not given the opportunity to infer advertising spending from a real advertising campaign—the common situation in the real world. Instead, campaign elements are artificially highlighted and in some cases advertising spending levels are given in dollars. The highlighting of advertising expense information in these experiments raises the possibility of demand effects driving the reported advertising spending-perceived correlation. It also raises the question of whether consumers can spontaneously pickup advertising spending information under naturalistic conditions. If they do not, then Nelson’s theory—even if it is internally valid—would have limited applicability. Kirmani (1990) partially corrects for these deficiencies by exposing subjects to fictional ads of different sizes. Kirmani (1997) goes further, exposing subjects to real ads at different frequencies. Kirmani (1997) finds an inverted U-shaped relationship between ad spending and quality perceptions. The upward sloping part of the curve is consistent with Nelson’s theory; the downward-sloping part is not. In this paper, we manipulate advertising spending via advertising repetition and measure its effect on consumers’ perceptions of quality. A key feature of our experiment is that it is explicitly designed to compare Nelson’s theory to the marketing view. The comparison is important because with ad repetition serving as a surrogate for ad spending, both views can predict the same overall relationship between perceived quality and advertising spending. Never-
355
theless, the mechanisms underlying the two approaches and their implications for advertising practice are completely different. Our experiment exploits these differences to discriminate between Nelson’s theory and the marketing view. In the process, it differs from Kirmani and Wright (1989) and Kirmani (1990) in several important ways. First, we use real ads. Not only does this increase the external validity of our results, but it also increases their internal validity. To adequately test Nelson’s theory, consumers must be assured that money was actually spent on advertising—otherwise, they would not be able to rule out the possibility of a lowquality product masquerading as a high-quality product. With fictitious products and fictitious advertising, this is hard to do (for example, in Kirmani, 1990, the ads were in a black-and-white magazine, and the cover page of the magazine was a set of instructions that asked the subjects to ‘‘look through the magazine as they naturally would if they were reading it at home.’’ Real magazines are generally in color and do not have reading instructions on their cover pages). Second, we experimentally manipulate whether respondents have to spontaneously pick up advertising frequency by actual exposure to an advertising campaign or frequency information is provided to them. By contrast, Kirmani and Wright (1989) only provided the summary information, and Kirmani (1990, 1997) only varied the physical exposure. For Nelson’s theory to have any external validity, of course, consumers ought to be able to pick up advertising spending information spontaneously while an advertising campaign unfolds. But for Nelson’s theory to have internal validity, it ought not to make a difference how the spending information is given to subjects. By contrast, the theories comprising the marketing view are explicitly theories about physical exposure to advertising. Finally, we examine four product categories, two of which are classifiable as search goods and the other two as experience goods. Nelson makes different predictions depending on whether the product is a search good or an experience good; but for the marketing view, it makes no difference.
2. Background 2.1. Nelson’s theory The essence of Nelson’s (1970, 1974a,b) theory is the idea that for experience goods, consumers should rationally infer that only high-quality products would spend much in advertising. This is because only high-quality brands can count on obtaining a significant number of repeat purchases. Low-quality brands pretending to be high quality will be ‘‘discovered’’ to be poor values after the first purchase, will not generate repeat purchases, and so cannot justify matching the high-quality firm’s advertising expenditures. For
356
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
search goods, on the other hand, advertising copy will be directly informative, and advertising spending has no signaling role to play [Nelson’s theory can be seen as giving a logical basis for why consumers should use a brand’s presence in the media as a cue to its popularity or acceptability with other consumers; Sutherland and Galloway, 1981. Note, however, that popularity and quality may not be correlated (see discussion below)]. Nelson’s theory is really a theory about the relationship between the true, objectively measured, quality of a product and advertising spending. Quality as perceived by consumers is nevertheless a key construct in the theory. Nelson postulates that consumers start out being uncertain about the quality of experience goods, but after observing the advertising spending level—and going through the logic outlined above—their perception matches the reality. Despite its compelling logic, several questions may be raised about the applicability of Nelson’s theory in the real world. First, consumers may not be able to observe or infer advertising spending levels. Advertising budgets are not printed on ads so it is difficult for the average consumer to tell whether brand X is spending more or less than brand Y. Moreover, advertising invariably involves nonpublic hidden costs—e.g., the cost of hiring advertising agencies, and the time devoted to developing and testing advertisements. Hertzendorf (1993) shows that the efficacy of advertising spending as a signal of quality goes down considerably if consumers are not able to observe advertising expenditures perfectly. Second, consumers may not be able to observe quality even after using the product. The difficulty arises not only because consumers are not ‘‘expert’’ at evaluating performance (Kempf and Smith, 1998), but also because they typically consume one brand at a time under ‘‘noisy’’ conditions—unlike, say, a professional testing laboratory, which can compare several brands simultaneously under controlled conditions. Hoch and Ha (1986) argue that when product trial is ambiguous, advertising claims shape consumers’ perceptions; Horstmann and MacDonald (1994) show theoretically that if consumption experience is not fully informative, then the signaling role of advertising spending is diluted. Third, advertising spending in the real world may be tied to market share, which is a function of value, not quality per se. For example, the brands rated high in quality by Consumer Reports are often not the products with the highest market share nor even Consumer Reports’ own ‘‘best buys.’’ Fourth, why should firms use advertising spending to signal product quality when other easier-to-read signaling instruments such as price (e.g., Bagwell and Riordan, 1991) and money-back guarantees (e.g., Moorthy and Srinivasan, 1995) are available? Finally, Nelson’s theory assumes that advertising claims on experiential attributes are uninformative. Yet most advertising for experience goods has experiential claims—e.g.,
most cereal advertising claims that the cereal tastes ‘‘great’’—and marketing specialists believe that such claims have a positive effect on their brands on repetition (Hawkins and Hoch, 1992). 2.2. The marketing view In the marketing view of advertising, there are at least three ways by which advertising can shape consumers’ perceptions of product quality: (1) by providing information about product attributes (learning theory: Hovland et al., 1953; Lavidge and Steiner, 1961; McGuire, 1978), (2) by increasing the consumer’s familiarity with the brand (mere exposure theory: Wilson, 1979; Zajonc, 1980; Sawyer, 1981), and (3) by shaping the consumer’s attitude toward the ad (attitude-toward-the-ad-theory: Mitchell and Olson, 1981; MacKenzie et al., 1986; Brown and Stayman, 1992). These alternative ways are not meant to be mutually exclusive. It is quite possible, for instance, that both mere exposure effects and learning go on simultaneously. As alternatives to Nelson’s theory, learning theory, mere exposure effects theory, and attitude-toward-the-ad-theory have much in common. First, like Nelson’s theory, they are consistent with a positive relationship between advertising repetition and perceived quality. Second, unlike Nelson’s theory, they are based on actual physical exposure of consumers to advertising repetition. In contrast, in Nelson’s theory, it does not matter how consumers get the advertising spending information; what is relevant is the spending signal, not how it is acquired. Third, because they are based on physical exposure to advertising, these alternative theories are amenable to ‘‘wear-out’’ phenomena. If the same ad is repeated too often, consumers may get irritated and bored, and this may inspire a degradation of perceived quality (Pechmann and Stewart, 1989). In contrast, wear-out would be hard to accommodate in Nelson’s theory: every additional exposure augments the spending signal and is therefore useful information for a motivated consumer. Finally, none of these alternative theories makes a distinction between search and experience goods. Table 1 summarizes these contrasting predictions.
Table 1 Nelson’s theory versus the marketing view
1. Relationship between ad repetition and perceived quality 2. Does it matter whether the respondent is actually exposed to ads repeatedly or is merely told the repetition frequency?
Nelson
Marketing view
Increasing for experience goods, no relationship for search goods No
Increasing for both search and experience goods. Inverted U-shaped relationship possible. Yes; the relationship between repetition and perceived quality applies only with actual repetition.
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
3. Experiment The basic idea behind our experiment is that we present ad spending information in two different ways, actual repetition of ads and ‘‘told’’ repetition—where the consumers are merely told about repetition frequency in a credible manner—and see whether it makes a difference to how consumers perceive product quality. Further, by including both search goods and experience goods in our study, we examine whether it matters if the product is a search good or an experience good. To make our advertising manipulations the sole source of information and familiarity for the products, we use Italian products advertised in an Italian magazine on non-Italian-speaking Canadian subjects. The cover story was that a number of Italian companies were considering entry into Canada and the subjects’ task was to examine each ad carefully in order to determine the likelihood of the product’s success in Canada. The study involved a 2 (product category type: search vs. experience) 2 (product categories) 3 (number of ad repetitions: one, three, or five) 2 (type of repetition: ‘‘told’’ vs. actual) factorial mixed-model design. One hundred and seventy-nine undergraduate students formed the subject pool. Product category type and (nested within it) the two product categories of each type were within-subjects factors, while the type and level of repetition were betweensubjects factors. We chose overcoats (brand allegro) and cookware (brand Æturnum) as the two search goods, and yogurt (brand Yomo) and nasal spray (brand deltinarolo) as the two experience goods. These classifications are easy to defend. Clothing and food products are often cited as the quintessential search and experience goods, respectively (Nelson, 1970; Davis et al., 1991). Cookware is a search good because its design, materials, and construction are observable in the store. Nasal spray is an experience good because the active ingredients are in an opaque bottle, and the consumer cannot evaluate things like efficacy and ease of use without actually using the product. A manipulation check supported the classification. When asked the question, ‘‘How easy is it to verify advertising claims about quality before purchasing the following products?’’ our subjects responded with mean ratings of 3.4 for nasal spray, 2.9 for yogurt, 2.6 for cookware, and 2.0 for overcoats on a four-point scale ranging from 1 = very easy to 4 = very difficult. These ratings show significant difference between category types when analyzed in a 3 (number of repetitions) 2 (type of repetition) 2 (type of category) 2 (categories) mixed-model MANOVA: F(1,173) = 165.46, P < .01. Subjects in the actual repetition condition saw five consecutive issues of the magazine Gente. Subjects in the told repetition condition saw only the last magazine issue, but on top of each ad was a label noting the frequency of appearance of the ad. Thus, for example, subjects in the three actual repetitions condition saw the test ads three times in five magazine issues, whereas subjects in the three told
357
repetitions condition saw the test ads once but the label told them that the ads appeared in ‘‘three out of five issues.’’ Each magazine issue had eight full-page ads from Gente and the actual cover page of the issue. Some issues had four test ads (for yogurt, cookware, overcoats, and nasal spray) and four filler ads; others had eight filler ads. There was no editorial material in these magazine mock-ups. This was consistent with the stated purpose of our study and our desire not to ‘‘take up too much of their time.’’ By choosing ads from a real magazine and including its cover pages in our magazine mock-ups, we intended to (1) corroborate our cover story, and (2) establish that the advertising was real. As discussed earlier, for Nelson’s signaling theory to work, consumers must be reasonably certain that actual advertising moneys were spent by the manufacturer of the product. Our use of Italian print ads on English-speaking subjects also was designed to improve the chances of Nelson’s theory applying because the subjects could not—even if they wanted to—get any information from the ad copy. But the ads were not totally uninformative. There was visual information in the ads— the pictures of the products and (in the case of the overcoat) the pictures of the models wearing the overcoat as well. These pictures give some information about the search attributes of the products but not about the experience attributes of the products. From a learning theory perspective, therefore, there was not much to learn for these ads.
4. Results 4.1. Manipulation checks In order to assess the effectiveness of the repetition manipulation in the two instruction conditions, the perceived number of repetitions provided by each subject was submitted to a 3 (number of repetitions) 2 (type of repetition) mixed-model MANOVA. There was a significant increase in the number of perceived repetitions as the number of ad repetitions increased [ F(2,173) = 143.8, P < .001]. Type of repetition did not interact with repetition [ F(2,173) = 2.69, P>.05]. Simple effects tests indicated that the perceived number of repetitions increased with the number of ad repetitions in both the actual [ F(2,174) = 52.3, P < .001] and told [ F(2,174) = 95.8, P < .001] repetition conditions. 4.2. Perceived quality Perceived quality was measured by asking questions of the form, ‘‘How would you rate the overall quality of the Æturnum brand of cookware?’’ and taking responses on nine-point scales anchored by ‘‘very poor quality’’ and ‘‘very good quality.’’ The cell means and standard deviations are shown in Table 2.
358
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
Table 2 Perceived quality: cell means and standard deviations
Cookware
Type of repetition
No. of repetitions
Mean
S.D.
N
Actual
1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5 1 3 5
4.79 5.86 6.31 6.26 6.58 6.00 6.34 5.86 7.52 6.61 6.74 6.57 4.52 5.48 5.55 5.23 5.94 4.97 5.59 6.52 6.69 6.23 6.42 6.03
3.19 1.27 1.77 1.57 1.52 1.62 2.14 2.12 1.68 1.75 2.07 1.78 1.92 1.94 1.40 1.91 1.71 1.33 1.50 1.09 1.04 1.48 1.57 1.10
29 29 29 31 31 30 29 29 29 31 31 30 29 29 29 31 31 30 29 29 29 31 31 30
Told
Overcoat
Actual
Told
Nasal spray
Actual
Told
Yogurt
Actual
Told
The effect of number of repetitions on product quality ratings was examined using a 2 (category type) 2 (category) 3 (number of repetitions) 2 (type of repetition) mixed-model MANOVA with categories nested within category type. The first two variables were within-subject factors, and the last two variables were between-subject factors. Product quality ratings show a significant increase with the number of repetitions [ F(2,173) = 4.83, P < .01]. However, Fig. 1 indicates that this effect is qualified by a significant interaction of number of repetitions and type of repetition [ F(2,173) = 8.0, P < .01]. Polynomial contrasts show a linear trend over number of repetitions for actual repetition (t = 4.56, P < .001) and a quadratic trend over number of repetitions for told repetition (t = 1.96, P < .053). Thus, while perceived quality consistently increased as the
number of actual ad repetitions increased, the quality ratings increased and decreased in an inverted U relationship as the number of told repetitions increased. Nelson’s theory would predict no difference in perceived quality’s response to actual and told repetition. There was no significant category_type by repetition interaction on quality ratings [ F(2,173) = 2.12, P>.12]. That is, the effect of repetition on quality ratings did not depend on the nature of the product category, contradicting Nelson’s theory. Polynomial contrasts suggest that the culprit may be a significant linear trend over number of repetitions for search goods (t = 2.2, P < .03)—not predicted by Nelson’s theory—that is insufficiently different from a significant linear (t = 2.0, P < .05) and quadratic trend (t = 2.75, P < .01)—increasing and then decreasing—for experience goods. At the category level (Table 3), actual repetition has a significant effect in all categories; told repetition is not significant in any of the four categories. 4.3. Attitude toward the ad (Aad) Ad attitude was measured by asking questions of the form, ‘‘What was your overall opinion of the advertisement for the Æturnum brand of cookware?’’ and taking responses on ninepoint scales anchored by ‘‘very good’’ and ‘‘very bad.’’ A MANOVA analysis of ad attitudes shows a significant main effect for repetition [ F(2,173) = 4.1, P < .02] and an insignificant interaction effect with repetition type [ F(2,173) = 2.1, P>.12]. The simple effects show that Aad responds linearly to actual repetition [ F(2,173) = 4.7, P < .02], but there is no response to told repetition [ F(2,173) = 1.5, P>.22], which is to be expected since subjects saw the ads only once in all the told repetition conditions. To see whether Aad might be mediating the relationship between repetition and perceived quality, we subjected each category’s perceived quality ratings to ANOVAs, with actual or told repetition as the only between-subject factor, with and without the corresponding Aad as a covariate (a MANOVA covariance analysis does not make sense in this context since, for example, attitude toward the yogurt ad cannot be a covariate for the perceived quality of nasal spray). These analyses with corresponding effect sizes (g2) are shown in Tables 3 and 4. Aad is highly significant in all categories with large effect sizes, and it does not mediate the relationship between told repetition and perceived quality. Table 3 Perceived quality of each category by actual and told repetition Cookware
Overcoats 2
F ratio g ( P) Actual repetition Told repetition Fig. 1. Perceived quality and repetition: actual versus told repetition.
3.54 * (.033) 1.05 (.355)
Nasal spray 2
F ratio g ( P)
.078 5.30 * (.007) .023 0.07 (.930)
* Significant at .05 level.
F ratio ( P)
.112 3.09 * (.051) .002 2.77 (.068)
2
Yogurt
g
F ratio g2 ( P)
.069
6.79 * (.002) 0.58 (.561)
.059
.139 .013
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
359
Table 4 Perceived quality of each category by repetition and repetition type with Aad as covariate Cookware Repetition
Actual repetition Told repetition
Overcoats Aad
Repetition
Nasal spray Aad
Repetition
Yogurt Aad
Repetition
Aad
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
F ratio ( P)
g2
0.74 (.482) 1.25 (.292)
.017
262.19 * (.000) 48.13 * (.000)
.760
7.34 * (.001) 0.018 (.982)
.150
172.41 * (.000) 102.67 * (.000)
.675
0.60 (.550) 2.10 (.128)
.014
57.59 * (.000) 87.83 * (.000)
.410
5.00 * (.009) 3.67 * (.029)
.108
76.77 * (.000) 139.96 * (.000)
.480
.028
.354
.000
.538
.046
.500
.077
.614
* Significant at .05 level.
There is some evidence, however, that it mediates between actual repetition and perceived quality: in two out of four categories, repetition loses significance when Aad is introduced, and in three out of four categories its effect size goes down.
5. Discussion Our results generally do not support Nelson’s theory. Referring to Prediction 2 in Table 1, we find that it makes a difference how advertising spending information is conveyed to subjects. When advertising spending information is embedded in the physical repetition of ads, perceived quality increases linearly with repetition; but when advertising spending information is embedded in statements of ad repetition, perceived quality has an inverted U relationship to repetition, with no linear trend. These differences are even starker at the individual category level (Table 3). Whereas actual repetition has a positive linear effect in all categories, told repetition has no effect in any of the categories. The results with respect to Prediction 1 are also not favorable to Nelson. Nelson would predict an effect for repetition only for experience goods. We find, however, that category type does not matter. Our results are, however, quite congenial to the marketing view. We noted earlier that the marketing view of advertising comprises at least three effects—learning, familiarity, and attitude toward the ad—each of whom can explain a positive relationship between perceived quality and actual repetition. Out of these, our experiment was not particularly well-suited to test for learning because we deliberately minimized the information in the ads in order to give Nelson’s theory its best shot. The other two effects cannot be ruled out, and in particular there is evidence of attitude toward the ad mediating the relationship between actual repetition and perceived quality in three out of the four categories. In fact, even in the category (overcoats) where attitude toward the ad does not seem to mediate, it is almost five times as effective as repetition in explaining the variance in quality. The two different ways of communicating the spending signal were obviously not equivalent. Advertising frequency information—and by extension advertising spending infor-
mation—was more salient and easier to acquire in the told repetition condition than in the actual repetition condition. Nelson’s theory, then, should have its best shot at succeeding in the told repetition condition. Yet even in this condition, what we get is an inverted U relationship between repetition and perceived quality when pooling across categories, and no relationship at all at the individual category level. The upward-sloping part of the inverted U curve is consistent with Nelson’s theory; the downward-sloping part contradicts it. Kirmani and Wright (1989) and Kirmani (1990) obtained similar results—under similar conditions (we discussed earlier that the manipulations of advertising in Kirmani and Wright, 1989 and Kirmani, 1990, made the advertising spending signal more salient than they would be in the real world. In this sense, they were closer to our told repetition condition than to our actual repetition condition). Kirmani (1990) interpreted the downward-sloping part as a ‘‘desperation undermine.’’ Instead of reasoning that ‘‘higher advertising spending signals higher quality because the firm is confident of repeat purchase’’—which is what Nelson would like them to do—subjects seem to think that ‘‘higher advertising spending signals lower quality because the firm must be desperate to spend so much on advertising.’’ Desperation reasoning seems more likely when excessive spending is more salient.
6. Conclusion In the conventional view of advertising, as described in marketing textbooks, the effectiveness of advertising is a function of its content (the message), execution (how the ad conveys the message), and frequency (how often a consumer sees the ad) (Kotler, 1997, chap. 20; Batra et al., 1996, chap. 5). The advertising spending level enters the picture only in service of these three factors. If a company does not spend enough on advertising, then its content might be off-base, or the ads might be poorly executed, or the frequency might be inadequate. Nelson’s (1970, 1974a,b) theory offers a radically different view. It argues that how advertising works depend on whether the product is a search good or an experience good and that the marketing view applies only to search goods. For experience goods, the only thing that matters is the advertising
360
S. Moorthy, S.A. Hawkins / Journal of Business Research 58 (2005) 354–360
spending level—or, for that matter, any "wasteful" spending—not content, execution, or frequency. This paper offers evidence in support of the marketing view. Our experiments show that it matters how advertising spending information is communicated to subjects. Subjects respond more to advertising repetition—in the sense of changing their perceived quality judgments—when they are actually exposed to advertising than when the advertising frequency data are provided to them as an abstract number. On the other hand, it does not seem to make much difference whether the product being advertised is a search good or an experience good. Our results suggest that the ad itself is a critical variable mediating the response to repetition. Subjects in our study seemed to like an ad more after repeated exposure to it and seemed to transfer this liking to the product being advertised, raising its quality rating. Our results are also consistent with ‘‘mere exposure’’ effects: increased ad exposure increasing familiarity with the brand and hence perceived quality. Learning effects were largely precluded by our experimental design, but other research suggests that it is an important way by which real-world advertising works (Batra et al., 1996, p. 151). Marketing professionals ought to be reassured by our findings. As it is practiced in the real world, advertising does not seem to differentiate between search goods and experience goods. Ad claims cover both search and experience attributes, and often even credence attributes. Advertising agencies have existed for a long time, and every year firms channel billions of advertising dollars through them. These practices would be hard to explain if advertising did nothing more than ‘‘burn money.’’
Acknowledgements This research was supported by grants from the Social Sciences and Humanities Research Council of Canada. The authors thank Amna Kirmani, John Lynch, and Akshay Rao for valuable discussions at various stages of the project.
References Bagwell K, Riordan M. High and declining prices signal product quality. Am Econ Rev 1991;81:224 – 39. Batra R, Myers JG, Aaker DA. Advertising management. Englewood Cliffs (NJ): Prentice-Hall; 1996. Brown SP, Stayman DM. Antecedents and consequences of attitude toward the ad: a meta-analysis. J Consum Res 1992;19(June):34 – 51. Caves RE, Greene DP. Brands’ quality levels, prices, and advertising outlays: empirical evidence on signals and information costs. Int J Ind Organ 1996;14:29 – 52.
Davis E, Kay J, Star J. Is advertising rational? Bus Strateg Rev 1991; (Autumn):1 – 23. Hawkins SA, Hoch SJ. Low-involvement learning: memory without evaluation. J Consum Res 1992;19(September):212 – 25. Hertzendorf MH. I’m not a high-quality firm—but I play one on TV. R J Econ 1993;24(Summer):236 – 47. Hoch SJ, Ha Y-W. Consumer learning: advertising and the ambiguity of product experience. J Consum Res 1986;13(September):221 – 33. Horstmann IJ, MacDonald GM. When is advertising a signal of product quality? J Econ Manage Strategy 1994;3(Fall):561 – 84. Hovland CI, Janis IL, Kelly HH. Communication and persuasion. New Haven (CT): Yale Univ. Press, 1953. Kempf DS, Smith RE. Consumer processing of product trial and the influence of prior advertising: a structural modeling approach. J Mark Res 1998;35(August):325 – 38. Kirmani A. The effect of perceived advertising costs on brand perceptions. J Consum Res 1990;17(September):160 – 71. Kirmani A. Advertising repetition as a signal of quality: if it’s advertised so much, something must be wrong. J Advert 1997;(Fall):160 – 71. Kirmani A, Wright P. Money talks: perceived advertising expense and expected product quality. J Consum Res 1989;16(December):344 – 53. Kotler P. Marketing management: analysis, planning, implementation, and control. Englewood Cliffs (NJ): Prentice-Hall; 1997. Kotowitz Y, Mathewson F. Advertising and consumer learning. In: Ippolito PM, Scheffman DT, editors. Empirical approaches to consumer protection economics: proceedings of a conference sponsored by the Bureau of Economics Federal Trade Commission, April 26 – 27, 1984. Washington (DC): Federal Trade Commission; 1986. p. 109 – 34. Lavidge RJ, Steiner GA. A model for predictive measurements of advertising effectiveness. J Mark 1961;25(October):59 – 62. MacKenzie SB, Lutz RJ, Belch GE. The role of attitude toward the ad as mediator of advertising effectiveness: a test of competing explanations. J Mark Res 1986;23:130 – 43. McGuire WJ. An information processing model of advertising effectiveness. In: Davis HL, Silk AJ, editors. Behavioral and management science in marketing. New York: Ronald Press; 1978. p. 156 – 80. Mitchell A, Olson J. Are product attribute beliefs the only mediator of advertising effects on brand attitudes? J Mark Res 1981;18:318 – 32. Moorthy S, Srinivasan K. Signaling quality with a money-bank guarantee: the role of transaction costs. Mark Sci 1995;442 – 66. Nelson P. Information and consumer behavior. J Polit Econ 1970;78: 311 – 29. Nelson P. Advertising as information. J Polit Econ 1974a;81:729 – 54. Nelson P. The economic value of advertising. In: Brozen Y, editor. Advertising and society. New York: New York Univ. Press; 1974b. p. 43 – 66. Pechmann C, Stewart DW. Advertising repetition: a critical review of wearin and wearout. In: Leigh JH, Martin Jr CR, editors. Current issues and research in advertising 1988. Ann Arbor (MI): University of Michigan Press; 1989. p. 285 – 330. Rotfeld HJ, Rotzell KB. Advertising and product quality: are heavily advertised products better? J Consum Aff 1976;10(Summer):33 – 47. Sawyer AG. Repetition, cognitive responses, and persuasion. In: Petty RE, Ostrom TM, Brock TC, editors. Cognitive responses in persuasion. Hillsdale (NJ): Erlbaum; 1981. p. 237 – 62. Sutherland M, Galloway J. Role of advertising: persuasion or agenda setting? J Advert Res 1981;21(October):25 – 9. Tellis GJ, Fornell C. The relationship between advertising and product quality over the product life cycle: a contingency theory. J Mark Res 1988;15(February):64 – 71. Wilson WR. Feeling more than we can know: exposure effects without learning. J Pers Soc Psychol 1979;37(June):811 – 21. Zajonc RB. Feeling and thinking: preferences need no inferences. Am Psychol 1980;35:151 – 75.
Journal of Business Research 58 (2005) 361 – 368
Enhancing or disrupting guilt: the role of ad credibility and perceived manipulative intent June Cottea,*, Robin A. Coulterb, Melissa Moorec a
University of Western Ontario, London, ON, Canada b University of Connecticut, Storrs, CT, USA c Mississippi State University, Mississippi State, MS, USA
Abstract Viewing the consumer as an active, skeptical reader of the persuasion attempt is an emerging perspective in advertising research. This perspective suggests that a consumer’s recognition of an emotional ‘‘tactic’’ in an ad can have a significant impact on an ad’s intended effect. Adopting this approach, we examine whether consumers’ evaluations of an ad’s credibility can enhance, and perceptions of manipulative intent can disrupt, the emotional response intended by the advertiser. We also investigate the effects of these two variables on attitude toward the ad and corporate attributions, including attitude toward the sponsor of the ad. We examine a commonly employed emotional tactic—the guilt appeal—and report the results of an experimental study. Our results suggest that credible guilt advertisements that are not overtly manipulative induce guilt feelings and positive attitudes. However, when consumers infer manipulative intent by the marketer, consumers do not feel guilty, but do have negative attitudes toward the sponsor of the advertisement and the advertisement. D 2003 Elsevier Inc. All rights reserved. Keywords: Advertising; Credibility; Manipulation; Guilt
1. Introduction Marketing communications specialists use a variety of appeals to accomplish their objectives. For example, informational appeals are often used to introduce a new product or to provide an update on brand enhancements, and emotional appeals are developed to make the viewer laugh or feel guilty. Sometimes advertisers are successful in achieving their intended objectives with a particular advertisement and sometimes they are not. Indeed, research has documented that there are no guarantees that the viewing audience actually feels the intended emotion associated with the appeal (Englis, 1990; Stout et al., 1990). For example, advertisers using a guilt appeal expect the audience to feel guilty and have some feeling of failing at their own ideals or ethical principles (Ruth and Faber, 1988; Wolman, 1973), but viewers do not always feel guilty (Coulter and Pinto, 1995). This article discusses two factors—consumers’ evaluations of the advertisement’s credibility and the advertiser’s
* Corresponding author. Ivey School of Business, University of Western Ontario, London, ON, Canada N6A 3K7. E-mail address: [email protected] (J. Cotte). 0148-2963/03/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00102-4
manipulative intent—that we believe can affect whether or not the advertiser accomplishes his or her objectives. We consider these two variables by examining the congruency between visual representation (the advertisement) and consumers’ responses (Scott, 1994) in the context of guilt appeals. We posit that one explanation for a lack of congruency between advertisers’ intentions and the consumers’ reactions is that consumers are active recipients of the advertising attempt. Specifically, consumers evaluate ad credibility and advertisers’ motivations. Further, we believe that when consumers perceive the advertiser as ill-intentioned, the intended emotion associated with the appeal, in our case guilt, may be attenuated such that the consumer does not feel guilty. In contrast, when an ad is perceived as credible and not manipulative, our expectation is that the consumers will experience the guilt intended by the advertiser. We begin by briefly discussing the use of guilt appeals in advertising. Then, we introduce two consumer evaluations— ad credibility and inferences of manipulative intent (IMI)— which we believe can enhance and interfere with marketers’ intentions, respectively. We then describe the experimental research designed to test our hypotheses and discuss our results. We conclude with a discussion of the theoretical and managerial implications of our investigation.
362
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
2. Guilt advertising appeals Guilt appeals—appeals where an advertiser attempts to make consumers feel guilty to influence their behavior—are commonplace in advertising. In fact, the use of this type of appeal is growing (Huhmann and Brotherton, 1997; Samalin and Hogarty, 1994). The psychology literature describes several forms of guilt, including anticipatory, existential, and reactive guilt (Izard, 1977; Rawlings, 1970). Further, Huhmann and Brotherton (1997) have documented the types of advertisers that attempt to generate these forms of guilt. Below, we define each type of guilt and provide a fictitious ad scenario to demonstrate the use of each form of guilt appeal (all ad scenarios taken from Coulter et al., 1997). Guilt that results from an individual contemplating a potential violation of one’s own standards is referred to as anticipatory guilt (Rawlings, 1970). Huhmann and Brotherton (1997) document that companies that manufacture and promote consumer nondurable goods and health care products, health care services, and charities often use this type of guilt appeal. The ad scenario below describes an appeal that attempts to create anticipatory guilt; the ad offers consumers the ability to avoid disappointing their children by purchasing a cell phone. Ad Scenario 1: It’s Saturday morning. The long-awaited (and many times postponed) trip to the beach is finally here. You and the kids are excited about the day’s activities. Then the phone rings. . .It’s your boss. He wants you to contact a potential multimillion dollar client today. Your youngest daughter, mortified by what might be another ‘‘Well, we can go to the beach next week’’ bleats ‘‘Daddy, when can I become one of your clients?’’. . .If only you had a cellular phone. You could avoid disappointing your kids and talk to the potential client en route to the beach. The second type of guilt is reactive guilt, i.e., a guilt response to having violated one’s standards of acceptable behavior (Rawlings, 1970). Huhmann and Brotherton (1997) found that manufacturers and advertisers of consumer goods (both durable and nondurable) and health care products and services focus on reactive guilt. The second scenario illustrates an ad attempting to remind readers of their own past transgressions in an attempt to sell a phone reminder service. Ad Scenario 2: It is 6:00 p.m., and you’re swamped with work, but you promise your wife. . .‘‘I’ll be packing up soon. I just have to read one or two more reports.’’ Predictably, you become engulfed in your work and the next thing you know it’s 9:00 p.m.! Well, she’ll understand. . .As you walk through the dining room, you stumble upon what once was a magnificent dinner for two. But the food is cold and the candles have burned down. Staring you in the face is a card saying, ‘‘Happy Anniversary, Darling.’’ You can’t believe it. How could you let such an important date slip by? If only you had
signed up for that ‘‘Special Event’’ phone calling service that would have reminded you of important dates. Finally, existential guilt is experienced as a consequence of a discrepancy between one’s well-being and of others (Izard, 1977), and Huhmann and Brotherton (1997) find that charity ads very often use this type of appeal. The final ad scenario depicts an ad attempting to evoke existential guilt by contrasting the condition of a starving child with the (presumably healthier) condition of the ad’s reader to solicit charitable donations. Ad Scenario 3: An emaciated child is perched on a log in the barren waste of the desert. He can barely move, as he has no muscle, to support his frail body. His eyes tell a tale of a boy who, if only someone had helped to provide him nutrition, could have blossomed into a bright young man. In the background, a vulture looks on. With just a small contribution, you can help stop such famine and the tragedy faced by these innocent youths. You can make a lifesaving difference. Common to these three forms of guilt appeals is the expectation that seeing ads that employ these tactics will cause viewers to feel guilt and take some action (for example, become volunteers or purchase a product or service). 2.1. Cognitive evaluations of advertisements We argue that understanding more about the effects of ad credibility and perceptions of the advertiser’s motivations will help to explain why some guilt ads work, i.e., have the intended effect of making the viewer feel guilty, whereas other guilt ads do not work. In the advertising and consumer behavior literature, ad credibility has been defined as the ‘‘extent to which the consumer perceives claims made about the brand in the ad to be truthful and believable’’ (MacKenzie and Lutz, 1989, p. 51). Thus, ad credibility focuses on the advertisement and the consumer’s evaluation of the truth and believability of the content of the advertisement (i.e., the visual and verbal content in the ad). We believe readers will interpret the ad itself (rather than only the source) to determine if the claims it makes are true. For example, concerning the ad depicted in Ad Scenario 3, one reader may believe the claim, ‘‘with just a small contribution, you can help stop such famine,’’ whereas another reader may not believe the claim in the ad. IMI have been defined as ‘‘consumer inferences that the advertiser is attempting to persuade by inappropriate, unfair, or manipulative means’’ (Campbell, 1995, p. 228). Hence, consumers’ IMI concern their assessments of the advertiser’s motivations as well as the extent to which the ad is perceived as fair. Thus, the reader who believes the ad (it is credible) and the reader who does not believe the ad (it is not credible) may have an independent assessment of whether the ad is manipulative or not. For example, a reader considering the ad in Scenario 3 might find the ad credible,
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
but feel the ad was manipulative because of its use of a very poignant picture, rather than simply words. To summarize, we believe that ad credibility and perceptions of manipulative intent (by the advertiser) are distinct constructs that will negatively covary. That is, we would expect that the greater the perceived credibility of the ad, the less likely the consumer would perceive the advertiser being ill intended. A consumer can read an ad and think that the claims in the ad are true and credible (for example, one may agree that many children around the world are starving), yet concurrently perceive the advertiser is attempting to manipulate them (i.e., they are trying to make me feel guilty so that I will donate to the charity). Hence, we hypothesize: H1: A negative relationship exists between perceived ad credibility and perceptions of manipulative intent. 2.2. The effects of ad credibility and perceived manipulative intent on feeling guilty Research on persuasive communications has indicated that consumers have a variety of emotional responses to advertisements (Englis, 1990; Stout et al., 1990), and that those responses are not always the intended responses. Research investigating emotional reactions elicited by the use of guilt appeals has identified the felt emotions of guilt, annoyance, and unhappiness as yielding the greatest impact on subsequent attitudes and purchase intention (Coulter and Pinto, 1995; Englis, 1990; Pinto and Priest, 1991). Positive emotions (e.g., feeling happy) yield a positive influence on attitudes toward the ad (Aad) (Holbrook and Batra, 1987). Alternatively, negative emotions (e.g., feeling annoyed) can result in a negative Aad (Burke and Edell, 1989; Edell and Burke, 1987), yet evoke a positive influence on behavior (Bagozzi and Moore, 1994; Ray and Wilkie, 1970; Sternthal and Craig, 1974). Other negative emotions, such as fear (Shelton and Rogers, 1981) and sadness (Cialdini and Kenrick, 1976), can also have a strong influence on creating a positive attitude toward helping. As we have noted previously, we believe that whether or not a guilt appeal will have its intended effect is, in part, dependent upon the viewers’ assessment of the advertiser’s motivations and ad credibility. We discuss each, in turn. 2.2.1. The effects of perceived manipulative intent Research by Eagly et al. (1978) and Wood and Eagly (1981) suggests that when a viewer perceives manipulative intent by the advertiser, they are likely to ‘‘resist’’ the message. We believe that this resistance has an impact on the viewers’ emotional responses (Batra and Ray, 1986). Research on guilt appeals provides some support for our contention (Coulter and Pinto, 1995). Specifically, Coulter and Pinto (1995) examined working mothers’ responses to low-, medium-, and high-intensity guilt ads and found that for high-intensity guilt appeals, the mothers did not feel guilty, but were rather angry. Englis (1990) also found high
363
correlations between guilt appeals and the unintended emotions of anger, disdain, and disgust. We believe that this reaction is a function of the subjects evaluating the ad and perceiving it as inappropriate and unfair. Specifically, with regard to guilt appeals, we posit that: H2a: As consumers perceive more unfairness, inappropriateness, or manipulative intent, they are less likely to feel guilty. H2b: As consumers perceive more unfairness, inappropriateness, or manipulative intent, they are more likely to feel angry. 2.2.2. The effects of ad credibility Cognitive response theory and empirical research suggest that when persuasive communications are perceived as more credible or include strong arguments for the product or topic, the cognitive responses and Aad are more favorable (Petty and Cacioppo, 1986). We believe that ad credibility will also affect emotional responses. We suggest that if consumers believe an ad (find it credible) and perceive no manipulative intent, then consumers’ emotional responses will be more congruent with the intent of the advertiser. In our context, this means that the viewer seeing a credible guilt appeal will likely feel guilty. Thus, we posit that: H3: The greater the perceived ad credibility, the more likely the consumers feel guilty. 2.3. The effects of ad credibility and perceived manipulative intent on Aad Research has theorized about and empirically examined the effects of ad credibility and perceived manipulative intent on measures of ad effectiveness, including Aad. This research suggests that ad credibility is likely to have a positive effect on consumer reactions to the ad (Kavanoor et al., 1997; MacKenzie and Lutz, 1989) and overall ad effectiveness (Goldberg and Hartwick, 1990). Furthermore, IMI have a negative effect on attitude toward the ad (Campbell, 1995). We believe that actually feeling guilty may influence Aad positively or negatively, depending on perceived credibility and level of manipulation. Consistent with this perspective, we posit: H4: A positive relationship exists between perceived ad credibility and Aad. H5: A negative relationship exists between perceptions of manipulative intent and Aad. 2.4. The effects of ad credibility and perceived manipulative intent on the attitude toward the sponsor of the ad (Asponsor) Researchers have also theoretically and empirically examined the effects of ad credibility and perceived manipulative intent on Asponsor and attributions about the sponsor
364
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
(Campbell, 1995; Friestad and Wright, 1994). In the extreme, perceptions of manipulative intent may create cognitions of mistrust of the company, which can lead to a consumer backlash against the advertiser (Basil et al., 1998; Osterhus, 1997). Thus, we expect: H6: A positive relationship exists between perceived ad credibility and corporate attributions and Asponsor. H7: A negative relationship exists between perceptions of manipulative intent and corporate attributions and Asponsor.
what they thought advertisers were attempting to accomplish with a series of ads and that each student would get one ad to assess. After examining the ad, students completed a short questionnaire. Embedded among other questions regarding intended emotions (e.g., make me happy, feel afraid) was a five-point Likert-type question, ‘‘[The sponsor] intended for me to feel guilty when viewing this advertisement’’ (1 = strongly disagree; 5 = strongly agree). The two ads did not differ significantly from one another in their perceived level of intended guilt [x¯EDF = 3.67, x¯STC = 4.00, t(35) = 0.85, P > .40].
2.5. Ad selection and pretesting Consistent with recent advertising research, we chose to use real ads in our study (see e.g., Kover, 1995). We made a concerted effort to identify ads for which our subject pool (undergraduate students) perceived that the advertiser’s intent was to create guilt in the audience. As a first step in identifying print advertisements to use in our study, we asked 80 undergraduate students to collect print ads for which they perceived that the advertiser’s intent was to instill guilt in them. From this collection, the authors identified four ads that appeared with the greatest frequency and then talked with the students to understand more about their reactions to the advertisements. The ads were for (1) MCI, (2) Environmental Defense Fund (EDF), (3) Save the Children (STC), and (4) Wrigley’s Gum (WG). Our next decision concerned which ads to use as stimuli. We wanted to examine only one type of guilt—so that our results would not be a function of type of guilt—and so the three authors individually evaluated the types of guilt evoked in the four ads. Social marketing causes often use guilt appeals, and the type of guilt appeal they most frequently use is existential guilt (Huhmann and Brotherton, 1997). We concurred that both the EDF and STC ads intended to evoke existential guilt, whereas the MCI ad intended to evoke anticipatory guilt and the WG ad intended to evoke reactive guilt. Thus, based on the debriefings and our assessments of the types of guilt intended in the ads, we selected the EDF and STC ads for our experiment. Due to the prevalence of existential appeals by charitable organizations, readers of these ads should be familiar with guilt appeals in social marketing ads. Although as we discuss below, prior familiarity with the specific ad shown did not affect our results. Past research has indicated that the intensity of the guilt appeal has an effect on emotional reactions and attitude to the ad, as well as behavior (Coulter and Pinto, 1995). To ensure that the EDF and STC ad had similar levels of perceived intended guilt, we conducted a pilot study with 37 undergraduate students (none of whom participated in the previous data collection). The EDF ad was originally in color, and research has shown that color gets more attention and creates longer viewing (e.g., Chute, 1980; Lamberski and Dwyer, 1981). We wanted to minimize this potential confound and so we used black and white versions of both ads. We told the students we were interested in understanding more about
3. Methodology Sixty-three undergraduate students, none of whom participated in our previous data collections, participated in our study. We randomly gave black and white copies of the EDF (n = 32) and STC ads (n = 31) to the students and instructed them that we were interested in their initial reactions to the ad. Having seen the ad, the students completed a questionnaire. The first question assessed subjects’ prior exposure to the ad. On the following pages, the subjects completed a questionnaire including 17 five-point Likert-scaled felt emotional responses to the ad (Edell and Burke, 1987; Izard, 1977; Plutchik, 1980). Several guilt items were embedded among many other emotion items to ensure that subjects did not focus on guilt. Additionally, the subjects completed five-point Likert ad evaluation items measuring Aad and the sponsor, or organization (Asponsor) in the ad (MacKenzie and Lutz, 1989). Next, the subjects indicated the level (on a five-point Likert scale) of five emotions (including guilt) they believed the advertiser was intending to create, as well as three items measuring attributions about the ad’s sponsor (Coulter and Pinto, 1995). Last, they evaluated the advertiser’s manipulative intentions (Campbell, 1995). All measures of felt emotions were taken before any items asking about inferences of advertisers’ intentions. 3.1. Measurement The five-point scale items used to measure each variable and the corresponding reliability coefficients are included in Table 1. The scales were either entirely or partially adopted from previous research. Specifically, the items measuring ad credibility are a combination of items used by Eaton (1988) and MacKenzie and Lutz (1989). The IMI scale is adopted from Campbell (1995). A factor analysis (including a scree plot) of the IMI indicates that the scale is unidimensional, with one factor accounting for 68% of the variance extracted (eigenvalue of 4.1), and other factors having eigenvalues of 0.5 or less. Feeling guilty is comprised of four items found in the research conducted by Bozinoff and Ghingold (1983) and Pinto and Priest (1991). The measures for attitude toward the ad (Aad) and sponsor (Asponsor) come from MacKenzie
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368 Table 1 Scales and reliabilities Scale items Ad credibilitya Believable Truthful Realistic Inferences of manipulative intenta The way this ad tries to persuade people seems acceptable to me. The advertiser tried to manipulate the audience in ways I do not like. I was annoyed by this ad because the advertiser seemed to be trying to inappropriately manage or control the consumer audience. I didn’t mind this ad; the advertiser tried to be persuasive without being excessively manipulative. The ad was fair in what was said and shown. I think that this advertisement is unfair/fair. Feeling guiltya Guilty Irresponsible Accountable Ashamed Attitude towards the ad Good/Bad Favorable/Unfavorable Positive/Negative Attitude towards the sponsor Good/bad Favorable/unfavorable Positive/negative
Reliability measure (a) .87
.89
.80
.96
.97
a Items measuring this construct were on a five-point Likert scale, 1 being strongly disagree and 5 being strongly agree.
and Lutz (1989). Items used to assess corporate attributions were taken from Coulter and Pinto (1995) and appear in Table 2. 3.2. Manipulation check To determine whether subjects perceived the advertiser was intending to make them feel guilty, we embedded the same five-point Likert-type question, ‘‘[The company] intended for me to feel [emotion] when viewing this advertisement’’ (1 = strongly disagree; 5 = strongly agree), that was used in the pilot study. Subjects perceived no significant difference in level of intended guilt across the two ads [x¯EDF = 3.78, x¯STC = 4.13, t(61) = 1.28, P >.21]. 3.3. Results Our analyses indicate that all our hypothesized relationships were significant in the predicted direction. Across the two ads, we found that ad credibility and IMI are negatively correlated (r = .51, P < .001, supporting H1). Additionally,
365
there is a significant negative correlation between IMI and feeling guilty (r = .42, P < .01). This finding supports H2a; as readers infer more of a manipulative intent on the part of the marketer, they are less likely to feel guilty. In addition, as readers perceive more manipulation, they are more likely to become angry (r =.34, P < .01), supporting H2b. Furthermore, ad credibility is positively related to feeling guilty (r =.48, P < .001, supporting H3). Thus, if readers view an ad as credible, they are more likely to feel guilty. As expected, ad credibility is positively related to Aad (r =.52, P < .001, supporting H4), whereas perceived manipulative intent is negatively related to Aad (r = .62, P < .001, supporting H5). Supporting H6, ad credibility is positively related to corporate attributions and attitude toward the sponsor (see Table 2). Additionally, we found that the greater the perceived manipulative intent, the more negative the attributions made about the sponsor of the ad, and the more negative the overall attitude to the Asponsor (supporting H7, see Table 2). The two ads were perceived by readers as equally credible [x¯EDF = 4.14, x¯STC = 4.06, t(61) = 0.30, P >.76]. However, the ads did differ on perceived manipulative intent, with readers perceiving significantly higher IMI in the children’s relief charity ad [x¯EDF = 1.78, x¯STC = 2.72, t(61) = 3.45, P < .001]. Therefore, we also analyzed each ad individually. Reactions to the same ad varied across subjects, although the variation was not very large. Because these variances are not large, using correlations to explain the variance accounted for in our tests is relatively conservative; a significant correlation depends in part in how much variance there is to explain in the variable. Our analyses of the individual ads again illustrate the significant negative relationship between ad credibility and IMI ( .62 for the EDF ad and .48 for the STC ad; see Table 3). Analyses of ad credibility, IMI, and feeling guilty demonstrate that when the ad is not perceived as manipulative (the EDF ad), ad credibility is positively and significantly related to feeling guilty, A ad , and corporate attributions. In addition, the relationship between IMI and the other variables is nonsignificant (see Table 3). However, when the advertiser is perceived as manipulative (the STC ad), credibility is not significantly related to feeling guilty.
Table 2 Credibility, inferences of manipulative intent, and corporate attributions Ad credibility Asponsor Corporate attributions [The sponsor] is primarily concerned with making money. I have a good feeling about [the sponsor]. [The sponsor] has consumers’ best interests at heart. n = 63. * P < .01. *** P < .001.
IMI
.45***
.49***
.41 *
.55***
.39 * .49***
.56*** .38 *
366
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
Table 3 Credibility, inferences of manipulative intent, feeling guilty
IMI Feeling guilty Aad Asponsor Corporate attributions [The sponsor] is primarily concerned with making money. I have a good feeling about [the sponsor]. [The sponsor] has consumers’ best interests at heart.
EDF ad
STC ad
Ad IMI credibility
Ad IMI credibility
.62*** .64*** .67*** .34
.30 .31 .21
.48*** .29 .44** .60***
.44** .72*** .66***
.47*
.29
.49*
.52*
.52*
.27
.47*
.41**
.43**
.22
.39**
.71***
EDF, n = 32; STC, n = 31. * P < .01. ** P < .05. *** P < .001.
Thus, if consumers believe that an ad is credible, yet manipulative, they are less likely to react as the advertiser intended, i.e., feel guilty. Moreover, consumers will have negative ad and sponsor evaluations, and make negative attributions about the corporation. At the same time, credibility remains positively and significantly related to Aad, Asponsor, and corporate attributions. Feeling guilty is positively related to Aad for both ads (EDF: r =.46, P < .001; STC: r =.54, P < .01). Since a potential confound might have been prior familiarity with the ad used in the study, it is important to point out that there was no difference in familiarity (reported number of times the ad was seen) between the two ads used and familiarity with the ad was not significantly related to either feeling guilty or IMI for either ad.
4. Discussion This article discusses two factors—consumers’ evaluations of the advertisement’s credibility and consumers’ perceptions of the advertiser’s manipulative intent—that can, respectively, enhance or disrupt the advertiser’s intended objectives. Because consumers are active readers of advertising, they may or may not respond as the advertiser expected, that is, they may or may not feel guilty as a consequence of seeing a guilt appeal. Specifically, when consumers perceive that a guilt appeal ad is credible, they are more likely to feel guilty. However, this is only the case when consumers infer the advertiser’s intent is not manipulative. If the consumer does perceive manipulative intent, he or she is less likely to feel guilty. Thus, a consumer’s evaluation of an ad’s credibility and the advertisers’ motivations can either enhance or disrupt message response. Our findings also suggest that consumers’ evaluations of ad credibility and advertiser motivations extend beyond
message response and impact subsequent attitudes toward the ad, attitudes toward the sponsor, and corporate attributions. Specifically, if consumers perceive an ad as credible, they are more likely to hold a positive attitude toward the advertisement and sponsor. Conversely, if consumers perceive manipulative intent on the part of the marketer, they are more likely to hold a negative attitude toward the advertisement and sponsor as well as negative corporate attributions. As such, advertisers need to walk a fine line between getting a message across and being perceived as overtly manipulative. Advertisers need to use caution when selecting guilt-inducing tactics due to the negative effects of perceived manipulation and unfairness on attitudes towards the company. Our research sheds some explanatory light on a controversy in the research dealing with negative emotional appeals (like fear or guilt). In their recent meta-analysis, Brown et al. (1998) point out that there are ‘‘strong and robust’’ results stating that negative appeals lead to negative ad, brand, and sponsor attitudes. However, they point out conflicting research that demonstrates positive effects of negative appeals. The present research demonstrates that if a negative appeal ad (in our case guilt) is credible and is not perceived as manipulative, the ad will lead to positive corporate attributions and attitudes. However, if the ad is seen as manipulative, despite the fact that it may be credible, then negative appeals can lead to negative corporate attributions and attitudes. One possible explanation for why one ad may be seen as manipulative and one may not (given equal credibility and levels of intended guilt) may be the type of guilt intended. As we reviewed earlier, there are three main types of guilt: reactive, anticipatory, and existential (Huhmann and Brotherton, 1997). Our study focused exclusively on two ads intended to elicit existential guilt. Future research might examine whether there is a link between the type of guilt intended and IMI: If there is indeed a link, advertisers would have some guidance on how to avoid these types of inferences by portraying different types of guilt in their ads. Although we did not assess individual differences in this research, it is possible that some people are more prone to feeling guilty, or feeling emotions in general (Basil et al., 1998; Moore and Harris, 1996). Future research should examine the role dispositional (trait) guilt plays in evaluating the advertiser’s intentions, and the ad itself (see, for example, Basil et al., 1998). Finally, although we did assess levels of prior familiarity with the ad (and found no difference), the persuasion knowledge literature suggests that consumers do not necessarily need to be familiar with the ad itself, but can call on their prior knowledge (Friestad and Wright, 1994) to make their evaluations. This knowledge may be of the tactics of this particular marketer in previous ads (for example, that STC ‘‘always’’ includes pictures of starving children) or knowledge of the typical marketing tactics in a product category (for example, charities ‘‘always’’ try to make you
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
feel guilty in their ads). Future research could determine how this type of agent and topic knowledge might interact with ad credibility and IMI (see Coulter et al., 1999). Some product categories (like social marketing campaigns) have used guilt appeals for decades. Persuasion knowledge of these guilt tactics in these categories may be higher than in some other areas, like nondurable consumer products. If the use of a guilt appeal is novel for a product category, then perhaps a lack of persuasion knowledge might lead to less perceptions of marketer manipulation. 4.1. Implications for marketers Looking across our results, we suggest that the advertiser striving for an effective use of a guilt appeal should focus on presenting viewers with messages that resonate with consumers’ experiences. More speculatively, marketers may also want to consider offering strategies for reducing guilt, such that consumers believe the claims, feel guilty, and act (by purchasing or following through with some other behavior) to alleviate the guilt, all without feeling overtly manipulated. However, an important caveat is that the more guilt tactics have been used in an industry, the higher will be the persuasion knowledge of most readers of the ad and the less likely it is that the tactic will go unrecognized. This research offers some empirical insights on guilt appeals, a growing strategic choice of advertisers (Murphy, 1994; Samalin and Hogarty, 1994). Guilt appeals that are perceived as containing well known manipulative tactics do not ‘‘work’’; there will not be a congruency between the representation in the ad and the consumer’s response (Scott, 1994). In this era of increasingly market-savvy consumers, advertisers need to avoid well-known guilt ‘‘tricks’’ and rather develop appeals that provide credible information in a nonmanipulative way.
References Bagozzi RP, Moore DJ. Public service advertisements: emotions and empathy guide prosocial behavior. J Mark 1994 (Jan);58(1):56 – 70. Basil DZ, Ridgway N, Nakamoto K, Basil M. The effects of guilt and empathy on charitable donations. Working paper presented at 1998 Society for Consumer Psychology Conference; Austin (TX); 1998. Batra R, Ray ML. Affective responses mediating acceptance of advertising. J Consum Res 1986;13(September):234 – 48. Bozinoff L, Ghingold M. Evaluating guilt arousing marketing communications. J Bus Res 1983;11(June):243 – 55. Brown SP, Homer PM, Inman JJ. A meta-analysis of relationships between ad-evoked feelings and advertising responses. J Mark Res 1998; 35(February):114 – 26. Burke MC, Edell JA. The impact of feelings on ad-based affect and cognition. J Mark Res 1989;26(February):69 – 83. Campbell MC. When attention-getting advertising tactics elicit consumer inferences of manipulative intent: the importance of balancing benefits and investments. J Consum Psychol 1995;4(3):225 – 54. Chute AG. Effect of color and monochrome versions of a film on incidental and task-relevant learning. Educ Commun Technol 1980;28(1):10 – 8. Cialdini RB, Kenrick DT. Altrusims as Hedonism: a social development
367
perspective on the relationship of negative mood state and helping. J Pers Soc Psychol 1976;34:907 – 14. Coulter RH, Pinto MB. Guilt appeals in advertising: what are their effects. J Appl Psychol 1995;80(6):697 – 705. Coulter RH, Cotte J, Moore M. Guilt appeals in advertising: are you feeling guilty? In: LeClair DT, Hartline M, editors. 1997 Winter Marketing Educators’ Conference Proceedings. Chicago (IL): American Marketing Association; 1997. p. 109 – 15. Coulter RH, Cotte J, Moore M. Believe it or not: persuasion, manipulation and credibility of guilt appeals. In: Arnould EJ, Scott LH, editors. Advances in consumer research, vol. 26. Provo (UT): Association for Consumer Research; 1999. p. 288 – 94. Eagly AH, Wood W, Chaiken S. Causal inferences about communicators and their effect on opinion change. J Pers Soc Psychol 1978 (April); 36(4):424 – 35. Eaton H. The effect of recall, importance and believability on attitude change: a study of television commercials. Thesis, University of Connecticut, Department of Communication Sciences; 1988. Edell JA, Burke MC. The power of feelings in understanding advertising effects. J Consum Res 1987;14(December):421 – 33. Englis BG. Consumer emotional reactions to television advertising and their effects on message recall. In: Agres SJ, Edell JA, Dubitsky TM, editors. Emotion in advertising: theoretical and practical explorations. New York: Quorum Books; 1990. p. 231 – 53. Friestad M, Wright P. The persuasion knowledge model: how people cope with persuasion attempts. J Consum Res 1994;21(June):1 – 31. Goldberg ME, Hartwick J. The effects of advertiser reputation and extremity of advertising claim on advertising effectiveness. J Consum Res 1990;17(September):172 – 9. Holbrook MB, Batra R. Assessing the role of emotions as mediators of consumer responses to advertising. J Consum Res 1987; 14(December):404 – 20. Huhmann BA, Brotherton TP. A content analysis of guilt appeals in popular magazine advertisements. J Advert 1997;26(Summer):35 – 45. Izard CE. Human emotions. New York: Plenum; 1977. Kavanoor S, Grewal D, Blodgett J. Ads promoting OTC medications: the effect of ad format and credibility on beliefs, attitudes, and purchase intentions. J Bus Res 1997;40(November):219 – 27. Kover AJ. Copywriters’ implicit theories of communication: an exploration. J Consum Res 1995; 21(March):596 – 611. Lamberski RJ, Dwyer FM. The instructional effect of color in intermediate and delayed reaction. J Instr Psychol 1981;8(4):122 – 31. MacKenzie SB, Lutz RJ. An empirical examination of the structural antecedents of attitude toward the ad in an advertising pretesting context. J Mark 1989;53(April):48 – 65. Moore DJ, Harris WD. Affect intensity and the consumer’s attitude toward high impact emotional advertising appeals. J Advert 1996;25(Summer): 37 – 50. Murphy AP. Feeling guilty. Parents 1994;69(September):8. Osterhus TL. Pro-social consumer influence strategies: when and how do they work. J Mark 1997;61(October):16 – 29. Petty RE, Cacioppo JT. Communication and persuasion: central and peripheral routes to attitude change. New York: Springer-Verlag; 1986. Pinto MB, Priest SS. Guilt appeals in advertising: an exploratory study. Psychol Rep 1991;69(October):375 – 85. Plutchik R. Emotion: a psychoevolutionary synthesis. New York: Harper & Row; 1980. Rawlings EI. Reactive guilt and anticipatory guilt in altruistic behavior. In: Macaulay JR, Berkowitz L, editors. Altruism and helping behavior. New York: Academic Press; 1970. p. 163 – 77. Ray ML, Wilkie W. Fear: the potential of an appeal neglected by marketing. J Mark 1970;34(1):54 – 62. Ruth JA, Faber RJ. Guilt: an overlooked advertising appeal. In: Leckenby JD, editor. The American Academy of Advertising proceedings. Austin (TX): American Academy of Advertising; 1988. p. 83 – 9. Samalin N, Hogarty DB. Guilt busters. Parents 1994; 69(September): 133 – 7.
368
J. Cotte et al. / Journal of Business Research 58 (2005) 361–368
Scott LM. The bridge from text to mind: adapting reader-response theory to consumer research. J Consum Res 1994;21(December):461 – 80. Shelton ML, Rogers RW. Fear-arousing and empathy-arousing appeals to help: the pathos of persuasion. J Appl Soc Psychol 1981 (Jul – Aug);11(4):366 – 78. Sternthal B, Craig CS. Fear appeals: revisited and revised. J Consum Res 1974;1(3):22 – 34. Stout PA, Homer PM, Liu SS. Does what we see influence how we feel? Felt emotions versus depicted emotions in television commer-
cials. In: Agres SJ, Edell JA, Dubitsky TM, editors. Emotion in advertising theoretical and practical explorations. New York: Quorum Books; 1990. p. 195 – 210. Wolman BB. Dictionary of behavioral science. New York: Van Nostrand Reinhold; 1973. Wood W, Eagly AH. Stages in the analysis of persuasive messages: the role of causal attributions and message comprehension. J Pers Soc Psychol 1981 (Feb);40(2):246 – 59.
Journal of Business Research 58 (2005) 369 – 376
Purchase occasion influence on the role of music in advertising Mark I. Alperta,*, Judy I. Alpertb, Elliot N. Maltzc a
Department of Marketing, The University of Texas at Austin, Austin, TX 78712, USA b Moonglow Jazz Group, USA c Willamette University, Salem, OR, USA
Abstract The role of background music in audience responses to commercials (and other marketing elements) has received increasing attention in recent years. This article extends the discussion of music’s influence in two ways: (1) by using music theory to analyze and investigate the effects of music’s structural profiles on consumers’ moods and emotions and (2) by examining the relationship between music’s evoked moods that are congruent versus incongruent with the purchase occasion and the resulting effect on purchase intentions. The study reported provides empirical support for the notion that when music is used to evoke emotions congruent with the symbolic meaning of product purchase, the likelihood of purchasing is enhanced. D 2003 Elsevier Inc. All rights reserved. Keywords: Music; Mood; Emotions; Situational influence; Advertising; Consumer behavior
1. Introduction How does the perceived purchase situation affect the impact of music in advertising on consumers’ moods, attitudes, and behaviors? This issue is motivated by the increased interest in emotional advertising and the complex roles that music assumes within it. For many marketing communication settings, the amount of objective information-processing activity is minimal, and there is substantial evidence that affective information processing may be instrumental in forming (or reinforcing) preferences and choices. Music has been used in stores, offices, and as a background in advertisements and has been reported to influence listeners’ emotions and behaviors. Music is a very useful tool for persuasion and exploring just how and why this is so is an important area for research. This article discusses, integrates, and builds upon the work of Gorn (1982) and others (e.g., Bruner, 1990) who provide theoretical and empirical insight into the ways in which music may influence consumer responses. To that end, we present an investigation into the effect of congruity and incongruity between affect or ‘‘mood’’ evoked by music imbedded in an advertisement and affect consistent with the
* Corresponding author. E-mail address: [email protected] (M.I. Alpert). 0148-2963/03/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00101-2
symbolic meaning of the purchase itself. Implications for marketing theory and strategy will be discussed. 1.1. Theoretical base and literature review The role of music in marketing and consumer behavior research has been addressed in education, psychology, communication, and other fields to determine its effects on behavior, mood, and preferences. As a result of this body of work, we know that in some instances, music appears to increase communication effectiveness in the context of advertisements. In other circumstances, music may decrease effectiveness for reasons that are not selfevident (e.g., ‘‘When is ‘popular’ music an inappropriate background?’’) Discussing how, when, and why music works seems to be appropriate to understanding the role of music in communications. In an effort to provide possible explanations, this article discusses the structural elements of music in the surrounding context of an advertisement and its interaction with the consumer. Although knowledge of formal musical analysis can assist in drawing inferences regarding how listeners may be affected by particular musical passages, it is also necessary to consider the context in which the musical and advertising ‘‘communication’’ takes place. Accordingly, we shall focus on musical structure and its interactions with important moderators such as ‘‘fit’’ between musical and
370
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
nonmusical meanings in the advertisement and purchase situation. Musical structure consists of elements such as sound, harmony, melody, and rhythm. Key factors in how these musical elements impact on the ad and the product are (1) the consumer, through different levels of involvement and cognitive or affective processing, (2) the consumer’s subjective perception of the appropriateness of the music as it relates to the central idea of the ad (‘‘fit’’ as defined by MacInnis and Park, 1991), and (3) the organization of musical elements. There has been interest in examining how musical elements influence affect and processing (Alpert and Alpert, 1990; Bruner, 1990). In addition, knowledge of cultural and social conditioning in forming musical taste as well as products can help in this prediction (Farnsworth, 1969; Holbrook and Schindler, 1989). Given a target market’s demographics, we can predict, with some accuracy, its musical and product preferences and tastes. Thus, we may know with some degree of certainty how they might perceive the appropriateness of certain musical selections with the overall message of an ad, although level of involvement and processing may vary across individuals and situations (Petty et al., 1983). Here, we investigate two key factors: music’s role in determining the emotional message of an advertisement and the impact of congruence between this message and the intended emotional meaning that may be conveyed through the purchase of the advertised product.
salient cues are processed to yield attitudes in a manner affected by mood. The likelihood that a host of behaviors may be performed appears to be enhanced by positive moods (Gardner, 1985). For negative moods, there have been mixed results, and several studies have shown that negative affect serves as a motivator to mood improvement through the performance of positive, prosaic acts. This finding agrees with the notion that where possible, people feeling badly will try to feel better. Negative moods’ effects on behavior may be more complex than the effects of positive moods (Cialdini and Kenrick, 1976). For example, helping may be enhanced by some negative mood states such as sadness (Baumann et al., 1981) and not by others such as frustration. This may be due to some evidence that negative mood states are not as homogeneous as positive ones (Isen, 1984) and that behaviors seen to reverse unpleasant mood states (e.g., helping) may overcome tendencies to enact mood-congruent behavior (e.g., withdrawal). In view of the fact that music is a common element in commercials and one that has a long history of mood inducement in a variety of contexts, the next section will focus on how music has been used as an independent variable to affect moods, as well as other dependent variables of interest to marketers. For brevity, this section will highlight key studies. Details on these and other studies are in Alpert and Alpert (1990), Bruner (1990), and Kellaris and Cox (1993).
1.2. Music also affects important mood states
1.3. Music effects
Music not only enhances recall for a product or an ad through an evoked image, but it may evoke moods, feelings, emotions, and behaviors. Consumer behavior theorists have conceptualized how consumers’ attitudes, affective states, and behaviors have been impacted by moods under central and peripheral processing, as well as affect and behavior conditioning. The association between mood states and affective responses, judgments, and behavior can be seen as both direct and indirect. A direct affective reaction may be viewed as a conditioned response when there are direct linkages in associations in memory between mood states and affective reactions (Griffitt and Guay, 1969) and between mood states and behavior (see Gardner, 1985, for additional references). Indirect associations between feeling states and affective responses and/or behavior include the influence of information processing or cognitive activity. Mood may affect evaluations by evoking moodcongruent thoughts and affect the performance of the behavior by increasing the accessibility of positive associations to the behavior (Clark and Isen, 1982; Goldberg and Gorn, 1987). To the extent that associations are direct and involve little conscious information processing, mood’s effects may be seen as via the peripheral route. Indirect associations may operate via the central route when other
Gorn (1982) suggested that peripheral influences such as background music used in commercials may become associated with the advertised product (in memory, even if not consciously) and influence product choice through classical conditioning. Mere exposure did not lead to liking, which apparently depended on whether the target product, a pen, was presented with liked versus disliked music. The second experiment by Gorn (1982) provided support for his hypothesis that when subjects were not in a decision-making mode, the commercial’s impact appeared to be more influential in its appeal when presented with musical background as opposed to product information. He concluded that through classical conditioning, the product becomes associated with the positive feelings of liked music. A replication of the Gorn study (1982) by Kellaris and Cox (1989) failed to reproduce the positive effect of liked versus disliked music after controlling for musical structural elements and possible demand effects. Their results question the effect of single exposures that merely vary background music’s appeal. They called for research on the influence of music’s structural elements on cognitive and affective responses (such as mood) toward the ad and the product. The key basic research relating musical elements to emotional responses was reported by Hevner (1935), who presented subjects with identical pieces, controlling for all
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
371
elements but major and minor modes. She concluded that all of the historically affirmed characteristics of the two modes, perceived as happy and sad, respectively, were confirmed in her study. In later research, she also reported associations between musical elements, such as fast tempo, loud dynamics, lively and varied rhythm, and high register with perceptions of the music as happy, merry, graceful, and playful. Musical elements such as slower tempo, quiet dynamics, unvaried rhythm, and low register were reported to be sad, dreamy, and sentimental (Hevner, 1935, 1936). She noted that although mode is never the sole factor that determines the way music is perceived, it is the most stable, generally understood, and influential of any of the elements in expressing the affective mood of music. Additional studies of musical structural elements’ effects on emotional responses are summarized in Bruner (1990) and North et al. (1999). Additional musical effects on mood and consumer behavior are found in studies such as Dube et al. (1995), who varied the musical background of a videosimulated bank and produced effects on pleasure and arousal and corresponding effects on desires to affiliate with bank employees. Higher affiliation was associated with musically induced pleasure and arousal. Alpert and Alpert (1990) replicated and extended Hevner’s findings, concluding that equally liked but unfamiliar music produced emotional responses predictable from analysis of its structural profile of musical elements. These mood states were associated with influence on purchase intention towards greeting cards viewed with the varying background music. Rather than generalize main effects from their study (sad music was ‘‘better’’), they suggested that future research may be productively directed at the interactions among music type, card type, and situation. They speculated that happy music may ‘‘help’’ happy cards when purchased for joyous occasions; sad music may help sad cards more than happy ones for sad situations such as funerals. The present study replicates and extends this work and tests their hypotheses about the role of musical structure on moods and consumer behavior under situational variation. For comparison, we also use unfamiliar music and simulated greeting card advertisements, but we vary the purpose of the greeting card by varying the occasion for which it is to be bought.
or purchase behavior of products presented along with the music.
2. Hypotheses
H2: When evoked mood is congruent with the mood of the purchase occasion, buying intention is higher than when the buyer and occasion moods are inconsistent.
It appears from the music literature that major and minor modes are important influences on listeners’ feelings and moods, and along with other musical elements, create a gestalt for happy and sad as perceived by listeners. Moreover, equally well-liked but unfamiliar music (to remove some effects specific to remembered music) with different musical structures may induce different moods. These in turn may predispose listeners toward different attitudes and/
H1: All else equal, music whose structural profile is ‘‘happy’’ influences listener moods to become more positive than music analyzed a priori as ‘‘sad.’’ From the extensive literature on the effects of mood states on information processing and interpersonal activities, a number of related inferences may also be derived. Most relevant for this particular study are those that pertain to the congruity between mood and thoughts and actions that may be affected by mood. As noted in a review by Bower (1991), ‘‘people who are temporarily (or characteristically) happy or unhappy tend to select interpersonal activities as well as social situations which will maintain their mood’’ (p. 31). He also noted that moods influence the processing of affectively congruent information as well impact-biased recall. In the context of the present study, this implies that persons feeling good (bad) will prefer choosing a greeting card for a happy (sad) occasion. This behavior would also be consistent with findings reported by Park and Young (1986) regarding the need for consistency between an advertisement’s musical ‘‘meaning’’ and the intended advertising message. Later work by MacInnis and Park (1991) provided additional support for the positive effect of musical and nonmusical consistency of message and resulting impact on advertising effectiveness. In the present study, the consistency would be reinforced through purchasing a greeting card one believes to represent feelings the sender wishes to convey to the card’s recipient. To the extent that a card is sent to symbolize a happy or sad occasion and one’s (music-induced) mood while viewing the card may be affectively conditioned and associated with the advertised product, purchase probability may be enhanced by congruity between communication effects and consumers’ feelings about greeting card ‘‘messages.’’ A number of studies have defined congruency in differing ways. Their findings are mixed and show congruency effects may be moderated by a number of factors. However, when congruent or moderately congruent information is presented in an ad, information processing generally is enhanced. Taken together, these studies suggest that:
3. Methodology The experimental subjects were 75 student volunteers from principles of marketing and promotional policies classes in a major southwestern university. Although the
372
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
test products were ones purchased by students, use of convenience samples of students limits the ability to generalize to some extent. Their use in this study was deemed appropriate for theory application and testing to investigate the basic process of musical and purchase occasion influence on emotions, attitudes, and purchase intentions (Calder et al., 1981). Extensions using other samples would be justified if initial results were promising. Volunteers received credit toward their course grade and were randomly assigned to treatment groups and times. The design was a 2 (music) by 2 (occasion) between-subjects factorial contained within a parallel study of source credibility and expertise effects. Subjects were told that the study concerned consumer responses to advertisements, and that they would be exposed to a series of slides depicting partial advertisement materials to be tested for possible inclusion in finished advertisements. Some would contain music and others would not. Because of time constraints, they would be asked questions about their responses to only a few of the ads. Confidentiality was assured. Attention was directed to the booklets on the tables in front of them, and they were told to complete the first page: names and classes (for credit) and wait for instructions. Then they were exposed to a series of 10 black-and-white slides, each of which was viewed for 45 seconds and preceded by an identifying announcement (e.g., ‘‘This is an ad for a financial institution, . . . greeting card . . . etc.’’). Black and white was used to avoid extraneous color effects on moods (and other responses) that might confound the measurements (Bellizzi et al., 1983). 3.1. Musical influence manipulation This treatment involved varying the background music for the target ad, 30-second excerpts from Bach preludes in Volume I of the Well-Tempered Clavier. All groups heard the same (pretested as) neutral music with the first simulated greeting card ad. Two of the four groups heard sad music with the second simulated card (the target ad) and the other two heard happy music. The sad music was from a prelude (XXII in B-flat minor) whose musical elements had been structurally analyzed as likely to produce feelings of sadness. The happy music was from a prelude (III in C#-major) that had a contrasting structure believed to produce feelings of happiness. Pretests had indicated the pieces were perceived as unfamiliar, equally liked but happy and sad, respectively. These excerpts were used as background music to the same, (pretested as) neutral slide of a mountain range and verbal message, ‘‘Friendship is a gift,’’ operationalizing the background music construct. 3.2. Occasion manipulation Also randomized within groups, half of those hearing happy or sad music with the second card were asked about their purchase likelihoods for happy occasions (‘‘for a friend
who was having a birthday’’) and half for sad occasions (‘‘for a friend who was sick in the hospital’’). Pretests had shown these scenarios evoked different perceptions of happiness/mood. 3.3. Measures Immediately after exposure to the fourth and sixth ads, subjects were asked to complete measures of feelings and then thoughts evoked by the ads, including their likelihood of purchasing the product in the ad. After the seventh and tenth ads, which simulated pictures for greeting cards, matched with the neutral and happy (or sad) musical excerpts, respectively, they were asked to provide the same measures. Hence, evoked feelings and attitudes were asked about ads with celebrities, experts, music, and no music. Using similar measures after ads with and without music was intended to minimize demand effects (and apparently did so, as indicated below). All measures were obtained using seven-point bipolar scales such as good/bad, happy/sad, excited/calm, and the like. Subjects were first asked to circle the number between the pair of words to indicate the feelings and emotions they had while viewing the advertisement. Next, they were asked to indicate their thoughts and opinions about the advertisements using similar (but randomly ordered) bipolar scales. Then they were asked to indicate their opinions of the product itself using three scales (good/bad, like/dislike, favorable/unfavorable). These were averaged to produce an index (a=.97) of attitude toward the product (Ao). Next, they were asked how likely they would be to choose the product that was advertised (very unlikely/very likely) that became the measure of purchase intentions. As noted in the occasion manipulation description, this question regarding the greeting card target ad was modified by the occasion for which the card was to be purchased (birthday versus hospital). Finally, the last four questions for the target ad asked how the respondent would characterize the occasion for which the greeting was to be sent (pleasant/ unpleasant, likable/unlikable, sad/happy, miserable/glad). Responses to these four scales were averaged to produce an index of the strength of the occasion manipulation (a=.90).
4. Results Although the subjects were predominantly U.S. citizens, it was possible that the three foreign students might respond differently to the advertisements; hence, they were eliminated. Subjects had been asked to write down what they believed to be the purpose(s) of the study. Answers generally described the purpose as ‘‘studying what ads people like and dislike.’’ One respondent who thought that the study had something to do with how music affects emotion was eliminated, leaving 71 usable responses.
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
4.1. Occasion manipulation
Table 2 Means by treatment condition
Tables 1 and 2 summarize major results for the manipulation check and dependent variables. Subjects told to consider purchasing the greeting card for a friend’s birthday scored lower on the (negative) occasion index (X¯ = 3.55) than did those rating the purchase situation of ‘‘for a friend in the hospital’’ (X¯ = 3.97, F = 3.38, P < .05, one-tail). The scaling of the four items in the occasion index indicates that the lower the score, the more the occasion was seen as pleasant/likable/happy/glad. Hence, the occasion manipulation produced adequate results. Given the modest difference between means and the fact that the birthday occasion was seen as near the midpoint of the situation index, this condition may be appropriately viewed as ranging from a fairly sad occasion to a neutral one.
Measure
‘‘Happy music’’
Happy – sad mood
Pleasant – unpleasant situation perceptions Favorable – unfavorable card attitude Purchase intention
373
‘‘Sad music’’
3.38
4.16
Happy occasion (n = 22)
Sad occasion (n = 18)
Happy occasion (n = 14)
Sad occasion (n = 17)
3.31
3.49
3.81
4.73
3.41 2.65
4.23 3.54
3.26
3.50
2.24
3.05 3.17
3.98 3.66
3.35
4.93 3.45
4.2. Impact on mood, attitudes, and intentions A mood index was formed by averaging the seven feelings scales that were highly intercorrelated. This produced an index (a=.92) that may be termed negative mood, as a high score is associated with feeling unpleasant, miserable, bad, gloomy, sad, boring, and weak. A low score is the opposite pattern, with a positive mood. Subjects exposed to the target card ad with a background of the ‘‘happy’’ music scored significantly lower on the (negative) mood index (X¯ = 3.38) than did those (X¯ = 4.16) exposed while listening to the ‘‘sad’’ music ( F = 8.52, P=.003, one-tail, eta2=.11). The lower mean (e.g., ‘‘more positive mood’’) for those exposed to ‘‘happy’’ music is consistent with H1 about the effect of advertising background music on receiver’s mood. Music/mood did not have a main effect on purchase intention ( F=.23, P>.10). It is possible that these effects may be moderated by factors (other than those stressed in the present study) such as degree of central versus peripheral processing and individual variations in need for cognition (Batra and Stayman, 1990). However, purchase occasion’s manipulation did affect purchase intention ( F = 7.72, P < .01, V2=.24), with the greeting card more likely to be purchased for a friend who was sick in the hospital (X¯ = 4.06) than one having a birthday (X¯ = 2.71). Since this
effect occurred when music was ‘‘held constant,’’ this result suggests that the picture and statement may have been viewed as more appealing for the former situation than the latter or that students felt it more important to send cards to ill friends than those with birthdays. Regardless of the relative acceptability of the greeting card for a given situation, the major interest here is on the extent to which music’s emotional meaning and influence on audience mood enhances or inhibits the purchase intentions under situations with varying perceived mood. In fact, there was a significant interaction effect of Music Occasion ( F = 5.68, P=.02, V2=.17). As shown in Fig. 1, in the (happier) birthday situation, subjects who had viewed the card with the happy background music, who had happier evoked moods, were more likely to choose it (X¯ = 3.17) than those who viewed it with sad musical background, whose evoked moods were sadder (X¯ = 2.24). On the other hand, when choosing a card for a friend who was sick in the hospital, a sadder situation, those who viewed the card ad with a background of sad music had a purchase intention that was substantially higher (X¯ = 4.93) than those exposed to the same ad with happy music (X¯ = 3.50).
Table 1 ANOVA results for treatments Dependent variable
Variation source
F value
Significance of F
V2
Occasion ‘‘mood’’
Music Occasion Music Occasion Music Occasion Music Occasion Music Occasion Music Occasion
9.801 3.379 1.776 1.844 4.473 0.046 0.227 7.717 5.682
.003 .070 .187 .179 .038 .830 .635 .007 .020
.29 .08 .03 .04 .16 .00 .00 .24 .17
Card attitude
Purchase intention
Fig. 1. Purchase intention interaction.
374
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
5. Discussion The results of this study support the hypothesis that variations in the formal music structure of background music in commercials may have significant influence over the emotional responses of an audience. Prior research in consumer behavior had shown that varying specific background music selections along dimensions of familiarity and liking could affect responses to ‘‘advertised’’ products. The present article extends the discussion concerning the effect of musical content that may lead to emotional and affective responses among consumers. It does so by determining whether congruity between musical advertisement ‘‘messages’’ and nonmusical variables that are part of the communications context but outside the advertisement itself (e.g., occasion) increases or blocks affect and behavior. Different profiles of musical structural elements of modality, tempo, dynamics, and rhythm may, all things being equal, lead to a perception of happy or sad musical content. In this study, equally liked musical backgrounds that differed in their profile of these structural elements were shown to affect audience moods in directions predictable from analysis of the musical structure, confirming earlier research by Hevner (1935) and Alpert and Alpert (1990). This finding has direct relevance to those interested in the impact on mood from factors such as the structural elements in background music. It was noted earlier that simultaneous variation of the entire profile of elements (major/minor, tempo, rhythm, and volume) precludes inferences from this study regarding their relative influence on moods and other dependent variables of interest. Other research suggests the dominance of major versus minor melodies (influencing moods to be happy versus sad), all else equal. To test the possibility that other factors might confound the study (Kellaris and Kent, 1993), a second factor of affective responses corresponding to ‘‘arousal’’ was extracted and found to vary in a manner consistent with the fast/slow tempi of the musical excerpts in the treatments. However, individual-level regression analyses showed that the feelings dimension ‘‘happy/sad’’ was correlated with buying intentions while feelings of arousal/nonarousal were not. More important, having found that musical structure does make a difference to moods and behavioral intentions towards products ‘‘shaded’’ with music, it may be appropriate to extend the present work with carefully controlled manipulations of specific structural elements of music. To this end, the methodologies employed by Holbrook and Huber (1979) and Kellaris and Kent (1993) may be productively used. Some advocates of classical conditioning might criticize the use of a single exposure to the messages and lack of reinforcement. However, evidence of mood-induced conditioning is demonstrated by the effect on purchase intent in ‘‘appropriate’’ situations. Perhaps, ‘‘affective transfer’’ (of music to mood to purchase occasion ‘‘fit’’) is an alternative explanation to ‘‘learning.’’ Earlier research has shown that
single exposures to background music-induced moods may affect buying intentions in the absence of significant intervening effects on the perceived sadness and even stated liking for a card and may be supportive of peripheral path processing in this setting (Alpert and Alpert, 1990). Given that the advertisements presented no verbal claims and that subjects were not told they would have to make an actual purchase choice (Petty et al., 1983), motivation to process information via the central route may have been diminished. The presence of music that evokes emotions and other ‘‘noninformational’’ aspects of the ad may also stimulate peripheral processing, and there were no central-route arguments involving objective claims presented in these target ads. Embedding the target ads in a series of control and distracter ads may also lower cognitive ad involvement to levels similar to typical ad viewing behavior. Research on the relevance of ‘‘fit’’ between message elements (Kellaris and Cox, 1993; MacInnis and Park, 1991) would suggest that evaluations might be influenced by congruence between key communication elements. We have manipulated these ‘‘outside the ad’’ by framing the purpose of the greeting card choice in terms of what was interpreted as a happy versus sad occasion. The musicevoked moods congruent with the feelings appropriate in such situations were associated with increases in purchase intentions. This interaction between music-induced mood and the purchase situation is itself an important finding. We have not directly investigated the manner by which this congruity/incongruity is perceived and processed by consumers. Future research may be productively employed to probe the extent to which consumers actively process the extent to which the message elements, thoughts, and/or feelings invoked by an informational versus emotional message (verbal or nonverbal) in a specific communication may lead to subsequent attitudes and behaviors. In this study, sad music-induced negative moods were associated with higher likelihood of sending a greeting card to a hospital than for a birthday. This may have come about because selecting a card that ‘‘made one feel sad’’ is inconsistent (or ‘‘jarring’’) with birthday thoughts. Similarly, subjects whose moods were induced to be happy (by ‘‘happy music’’) were more likely to maintain their moods by sending the card for a birthday than for a friend sick in the hospital. The main purpose of the present study has been to extend the utility of music’s structural analysis to the relationships between hypothetical influences on audience mood and the resulting congruency/incongruity with the purchase purpose or situation. In addition, it is hoped that as more is understood about how consumers process congruent and incongruent musical (or other nonverbal) and verbal information, the more it will be possible for creative efforts to be optimized. Having shown potential application of the concepts expressed, it would be useful to extend this work by testing music’s effects on moods and intentions for products for which decisions might be made in the presence of music.
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
Situations similar to the present study’s context are found in television presentations of products to be ordered by phone, or on the Internet, with background music varied to be appropriate to the product or purchase occasion. Here, the influences on mood and intentions might be sufficient to produce buying. In other contexts, repetition of similar advertisements might produce purchase intentions of longer duration, if not toward a specific card (or product), perhaps towards a brand such as Hallmark. Seeing the advertised products in a store could evoke the mood originally stimulated by music in the advertising situation. Although such an outcome is plausible and is supported by the literature (Lutz and Lutz, 1978), further research is needed to test the efficacy of this linking of feelings. As noted earlier, research cited by Gardner (1985) has generally found positive correlations among mood inducers, moods, and a number of dependent variables such as evaluations. However, studies such as Cialdini and Kenrick (1976) found that older children were more generous when self-generated thought made them sad. In a negative emotional ad, there may be more of a payoff. In this case, a sick friend is in the hospital (negative emotion), followed by a resolution (a positive emotion), in the form of a card. The point is made that in a negative emotional ad there is a higher level of activity involvement than with positive ads, where people are in a good mood, have positive thoughts, and want to stay that way with minimum elaboration. It may be likely that both the sad purchase situation and sad music manipulation resulted in more moderate levels of elaboration, leading to higher scores in purchase intentions. In addition, this particular example of a negative emotion and its resolution has high social approval and would increase motivation to purchase a card. As Gardner (1985), Park and Young (1986), and Kellaris and Cox (1993) have stated, a key factor is the congruence between associated feelings and behaviors consistent with that advocated in a message. In this situation, college students may have responded more positively to sad emotional evocations (induced by music) in the context of sending greeting cards to sick friends (with messages like ‘‘friendship is a gift’’). It makes sense intuitively to say that in this study, when happy music was paired with a sad purchase occasion, this was an incongruent situation where unexpected, irrelevant, inappropriate, or perhaps no information was conveyed and where purchase intent was low. The same was true with the other incongruent situations. In the happy music with happy occasion, and sad music with sad occasion, congruency was achieved through both expected and relevant information. The result of increased purchase intent, when there was congruity between happy music/ mood and happy purchase occasion, as well as sad music/ mood and sad purchase occasion is supported by Bower (1981), who points out that people who are happy or unhappy tend to select interpersonal activities as well as social situations that will maintain their mood. Certainly, Hallmark positions (verbally and nonverbally) many of
375
their greeting cards to appeal to these market situations. So did AT&T with their memorable ads supported by the song, ‘‘Feelings’’ (a sad song, in minor mode, that helped people resolve distance by phoning). On the other hand, visits to an amusement park may be more effectively advertised with happy music than with sad. In addition, happy and sad may well be multidimensional constructs. Different gradations within these emotions may require different inducers and may in turn produce different responses and behavior. For example, there may be different kinds of happiness (or sadness), influenced by different factors, and may lead to different responses (relaxation after completion of a difficult task, enthusiastic expressions of joy, and the like). In addition, music has a host of elements that may be influential beyond the musical structure. These include the words, artistic interpretation, and specific memories that may be associated with the selection, type and period of music, and the interaction of all of these with the product and use situation stressed in the advertisement. Additional research may eventually be able to decompose overall effects into elements of all of these components, taking into account the effect of moderator variables, such as the respondents’ countries and cultures, demographics, personality and lifestyle, cognitive and affective involvement in the communication setting, and familiarity with the music. The tasks in pursuing these issues are considerable, but it seems worthwhile to decompose factors such as musical influence into theoretical elements and their combinations. It is encouraging in this process of inquiry to find that predictions from musical theory may be derived that show correspondence in the emotional responses of audiences. To the extent that this phenomenon might be validated in future experiments, it may be possible to provide better explanation of this source of emotional response to commercials, as well as screen potential advertisements for predicted influences. As has been shown here and by other researchers, moderating effects such as the role of involvement, prior experience with the product category and specific brands or types, as well as the specific musical and other message elements, are all part of this picture. Programmatic research is needed to evaluate their independent and interacting effects under controlled conditions. There remain a number of issues to be explored in the links between verbal and nonverbal elements of a communication setting. It is hoped that this exposition and empirical study of congruity between advertisement background musical structure, consumer’s evoked mood, and ‘‘mood’’ of the purchase purpose or situation will stimulate future efforts along these lines.
Acknowledgements We thank Kalpesh Desai for the assistance in data collection, the student subjects at the University of Texas at Austin, and Meryl Gardner and Morris Holbrook for the helpful comments. Responsibility for errors is ours.
376
M.I. Alpert et al. / Journal of Business Research 58 (2005) 369–376
References Alpert JI, Alpert MI. Music influences on mood and purchase intentions. Psychol Mark 1990;7(Summer):109 – 34. Batra R, Stayman DM. The role of mood in advertising effectiveness. J Consum Res 1990;17(September):203 – 22. Baumann DJ, Cialdini RB, Kenrick DT. Altruism as hedonism: helping and self-gratification as equivalent responses. J Pers Soc Psychol 1981;40(6): 1039 – 46. Bellizzi JA, Crowley AE, Hasty RW. The effects of color in store design. J Retail 1983;59(Spring):21 – 45. Bower GH. Mood and memory. Am Psychol 1981;36:129 – 48. Bower GH. Mood congruity of social judgement. In: Fargar JP, editor. Emotion and social judgements. Oxford: Pergamon; 1991. p. 31 – 53. Calder BS, Phillips L, Tybout AM. Designing research for application. J Consum Res 1981;8(September):197 – 207. Cialdini R, Kenrick D. Altruism as hedonism: a social development of negative mood state and helping. J Pers Soc Psychol 1976;34(5):907 – 14. Clark M, Isen A. Toward understanding the relationship between feeling states and social behavior. In: Hastorf A, Isen A, editors. Cognitive social psychology. New York: Elsevier/North-Holland; 1982. p. 73 – 108. Dube L, Chebat J-C, Morin S. The effects of background music on consumers’ desire to affiliate in buyer – seller interactions. Psychol Mark 1995; 12:305 – 19. Farnsworth PR. The social psychology of music. Ames (IA): Iowa State Univ. Press; 1969. Gardner MP. Mood states and consumer behavior: a critical review. J Consum Res 1985;12(December):281 – 300. Goldberg ME, Gorn GJ. Happy and sad TV programs: how they affect reactions to commercials. J Consum Res 1987;14(December):387 – 403. Gorn GJ. The effects of music in advertising on choice behavior: a classical conditioning approach. J Mark 1982;46(Winter):94 – 101. Griffitt W, Guay P. ‘Object’ evaluation and conditioned affect. J Exp Res Pers 1969;4(July):1 – 8.
Hevner K. The affective character of the major and minor modes in music. Am J Psychol 1935;47:103 – 18. Hevner K. Experimental studies in the elements of expression in music. Am J Psychol 1936;48:246 – 68. Holbrook MB, Huber J. Separating perceptual dimensions from affective overtones: an application to consumer aesthetics. J Consum Res 1979; 5(March):272 – 83. Holbrook MB, Schindler RM. Some exploratory findings on the development of musical tastes. J Consum Res 1989;16(June):119 – 24. Isen AM. The influence of positive affect on decision making and cognitive organization. In: Kinnear TC, editor. Advances in consumer research, vol. 11. Provo (UT): Association for Consumer Research; 1984. p. 534 – 7. Kellaris JJ, Cox AD. The effects of background music in advertising: a reassessment. J Consum Res 1989;16(June):118 – 28. Kellaris JJ, Cox D. The effect of background music on ad processing: a contingency explanation. J Mark 1993;57(October):114 – 25. Kellaris JJ, Kent RJ. An exploratory investigation of responses elicited by music varying in tempo, tonality, and texture. J Consum Psychol 1993; 2(4):381 – 401. Lutz K, Lutz RJ. Imagery-eliciting strategies: review and implications for research. In: Hunt K, editor. Advances in consumer research, vol. 5. Chicago (IL): Association for Consumer Research; 1978. p. 611 – 20. MacInnis DJ, Park CW. The differential role of characteristics of music on high- and low-involvement consumers’ processing of ads. J Consum Res 1991;18(September):161 – 73. North AC, Hargreaves DJ, McKendrick J. The influence of in-store music on wine selections. J Appl Psychol 1999;84(February):271 – 6. Park CW, Young MS. Consumer response to television commercials: the impact of involvement and background music on brand attitude formation. J Mark Res 1986;23(February):11 – 24. Petty RE, Cacioppo JT, Schumann D. Central and peripheral routes to advertising effectiveness: the moderating role of involvement. J Consum Res 1983;10(September):135 – 46.
Journal of Business Research 58 (2005) 377 – 386
Separate and joint effects of medium type on consumer responses: a comparison of television, print, and the Internet Majorie Dijkstra*, Heidi E.J.J.M. Buijtels, W. Fred van Raaij Department of Marketing Management, Rotterdam School of Management, Erasmus University Rotterdam, P.O. Box 1738, 3000 DR Rotterdam, The Netherlands
Abstract This study explores the effects on consumer responses of single- and multiple-media campaigns consisting of television, print, and the Internet. Multiple media in a campaign are expected to have synergy effects. We examine whether a complementarity effect is present in multiple-media campaigns related to media differences in evoking cognitive, affective, and conative responses. Media contribute differentially to the route to persuasion and may complement each other in a marketing-communication campaign. The results show that TVonly campaigns are superior in evoking cognitive responses. This superiority is probably due to the larger number of senses stimulated as well as the forced exposure associated with television as a delivery medium. Affective and conative responses do not significantly differ between the single-medium campaigns. Product involvement influences brand affect and purchase intention. The analysis of covariance reveals a complementarity effect in multiple-media campaigns compared to the Internet-only campaign. However, compared to the TV-only campaign, multiple-media campaigns are less effective in evoking cognitive responses. For most responses, print-only campaigns are as effective as multiple-media campaigns. D 2003 Published by Elsevier Inc. Keywords: Single-medium campaign; Multiple-media campaign; Affective responses; Cognitive responses; Conative responses
1. Introduction New media, such as the Internet, and better-targeted magazines, radio, and TV channels create new opportunities for marketing communication. Given the available media options, a strong need arises to coordinate the use of media in marketing-communication campaigns to reach the desired communication objectives. To effectively communicate a message by use of multiple media, advertisers, media planners, and advertising agencies need to understand the strengths and weaknesses of each medium and, more importantly, need to understand differences in the way consumers process information ‘‘presented’’ by different media (Buchholz and Smith, 1991; Vakratsas and Ambler, 1999). Research provides evidence of differences between media in processing and consumer responses. Several studies assessed the media effectiveness as well as the joint effects of mass media, such as broadcast (radio and TV) and print media (Buchholz and Smith, 1991, Edell and Keller, 1989,
* Corresponding author. E-mail address: [email protected] (M. Dijkstra). 0148-2963/$ – see front matter D 2003 Published by Elsevier Inc. doi:10.1016/S0148-2963(03)00105-X
1998; Jacoby et al., 1983). In contrast, little research has been done to assess the effectiveness of the Internet compared to traditional media (Sundar et al., 1998). To our knowledge, no research examined how the Internet can be integrated in marketing-communication campaigns and how synergy can be obtained from using media combinations, including the Internet. Given the differences in information processing and consumer responses among media, and based on the idea that each medium possesses modalities and strengths that may be instrumental in accomplishing specific objectives, synergy effects may be expected in multiple-media campaigns. Synergy occurs when different media with their specific strengths complement each other in a campaign or when the strength of one medium compensates for the weakness of another medium. Our research goal is to explore the effects of single- and multiple-media campaigns on consumer responses using television, print media, and the Internet. These responses are cognitive, affective, and conative responses and are related to communication objectives such as creating brand knowledge, brand affect, and purchase intention. We examine whether media complement each other in the route to persuasion because of their differential effects of
378
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
evoking cognitive, affective, and conative responses. We discuss the literature and test research hypotheses in an experimental study, comparing consumer responses to single- and multiple-media messages. Furthermore, we discuss the implications for media planning and marketing-communication management.
2. Theoretical background and hypotheses Several dimensions are instrumental in the processing of information in media. Two of these dimensions are modality of the medium and control over the medium (i.e., pacing).
by handling one sensory mode at a time and switching back and forth between modes in a scanning procedure to piece the message together. Edell and Keller (1989) noted that while television does provide more information to the viewer, this information might require more effort to process. Furthermore, distracters (e.g., discrepant verbal and visual components) may interfere and inhibit cognitive elaboration and critical thinking (Edell and Keller, 1989). According to Wright (1980), the interference or ‘‘negative’’ synergy of concurrent sensory modes depends on the nature of the sensory modes themselves (i.e., their postponability, their interest value, and the absolute information load). 2.2. Control over the medium
2.1. Modality of the medium Modality refers to the mode of presentation (i.e., text, audio, picture, or video) that corresponds to human senses used for processing the presented material. Media differ in the content and number of sensory modes stimulated. Each sensory mode may potentially affect processing by directly evoking cognitive and affective reactions or indirectly influencing the processing of and reactions to other sensory modes (Edell, 1988). Television reaches ear and eye by moving visuals, words, and sounds. The combination of moving visuals and audio gives television other powers than static media such as print with only text and visuals. The Internet may combine the moving visuals of television with the detailed information as given in print media. Two contrasting views exist on the capacity of consumers to process information from different sensory modes. According to Jacoby et al. (1983), the commonsense notion is that the larger the number of sensory modes that are engaged in the communication process, such as with television, the greater the likelihood of effective communication. No or little interference between modes is assumed. According to this view, television should be superior on memory and cognitive responses. Besides, television advertisements contain more visual elements that are easier to process and that support the verbal information delivered in the message. Paivio’s (1971) dual-coding hypothesis suggests that although pictures are coded with more variation than words, pictorial elements are easier to retrieve from memory. Research has shown that consumers have a rather impressive ability to pay attention to what they need to attend to (‘‘focussing’’) and at the same time monitor other less well-attended sources of information at some minimum level of meaning (‘‘scanning’’) (Deutsch and Deutsch, 1963; Triesman, 1964). Furthermore, Kisielius and Sternthal (1984) found that multiple sensory modes facilitate learning, i.e., recognizing the meaning of information in one mode may facilitate the interpretation of meaning in another mode. On the other hand, Broadbent (1958) found that the capacity of the human perceptual system is limited and that it can only process sensory information selectively. In a complex and cluttered media environment, individuals react
Control over the speed and sequence of information transfer (i.e., pacing) is another discriminating media factor. This may be controlled by the sender (external pacing) or by the receiver (internal pacing) (Pieters and van Raaij, 1992). As a consequence of pacing, delivery and retrieval media may be distinguished (Van Raaij, 1998). Television is an example of a delivery medium with external pacing. The advertiser controls the speed of information transfer and the order of the information items in a TV ad. Because of its transient character, television is poor on documentation, that is, it does not facilitate consumers to keep the presented information, such as prices, telephone numbers, and Internet addresses. On the other hand, it is expected that television, as a delivery medium that uses multiple sensory modes, has cognitive impact even on low or uninvolved consumers (Buchholz and Smith, 1991). In this respect, under lowinvolvement condition, consumers may engage in superficial processing and may produce cognitive responses, consistent with Krugman’s (1965) ‘‘learning without involvement.’’ In contrast to television, print media and the Internet allow consumers to process the information at their own pace and sequence (internal pacing). This enhances the opportunity to process the information (Jacoby et al., 1983) if motivation is high. Thus, with high involvement, more elaborate information processing may occur. In addition, retrieval media, such as print media and the Internet, enable consumers to retain the information and to read messages at their own convenience. This facilitates a better cognitive processing of the message and might lead to more cognitive responses. On the other hand, retrieval media have the disadvantage that they have limited opportunity to influence less involved or passive consumers (Buchholz and Smith, 1991). Consumers that are not interested in the message can easily skip the message. But if consumers are interested, they may give more attention to the message and more cognitive processing may take place. In our view, the Internet and print media are now largely similar considering the senses stimulated and the speed and control of information transfer, and therefore, we expect no differences in cognitive response. However, Sundar et al. (1998) found better memory performance for the print
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
medium than for the Internet. The difference in delivery mechanism (i.e., paper versus computer) is a possible explanation for this finding. They stated that print ads allow the readers to process the ad in its entirety, whereas, Internet ads must be scrolled through to be processed. In conclusion, it is not evident which medium is most effective in evoking cognitive responses. Paivio’s (1971) dual-coding hypothesis predicts that television lead to more cognitive responses. On the other hand, the nature of print and the Internet as retrieval media with internal pacing suggest that these media are superior on cognitive responses if involvement is sufficient. For this reason, no hypothesis is formulated with regard to the cognitive responses. The data are explored to assess which medium is most effective in evoking cognitive responses. Compared to print media and the Internet, the speed of information transfer in television is relatively high and not under control of the viewer. As a consequence, the cognitive processing and rehearsal of messages transmitted by television will be low. Under these circumstances, it is expected that consumers restrict themselves to global evaluative responses in terms of good/bad or attractive/unattractive (Pieters and van Raaij, 1992). Furthermore, with its communication power, television may transfer moods, feelings, and images that facilitate affective responses. Therefore, it is expected that consumers exposed to television will give more affective responses than exposed to print media or the Internet. H1a: Participants exposed to TV only give more affective responses than participants exposed to print or Internet only. Compared with print and television, the Internet has an additional property. It enables consumers to request more information and to buy the advertised brand or product directly and when it is convenient to them. The Internet enables integration of advertising and purchase. Therefore, the Internet may evoke more conative responses because it facilitates purchase without time delay or other constraints. In other words, we expect that the Internet will facilitate direct ordering and purchase after persuasion and thus, may stimulate impulse buying. H1b: Participants exposed to Internet only give more conative responses than participants exposed to TV-only or print-only. Assuming that media differ to the extent in which they evoke responses because of their differences in communication power, we expect that media complement each other in the route to persuasion. Media are complementary when the strengths of different media are combined in a multiplemedia campaign and when a weakness of one medium is compensated by the strength of another medium. Based on the complementarity effect, we expect that campaigns consisting of three different media build on the strengths of each medium in the campaign and evoke more responses than campaigns with only one medium.
379
H2: Media campaigns consisting of three media will evoke more cognitive, affective, and conative responses than media campaigns with only one medium. Note, however, that in this hypothesis, we do not argue that the more media used in an integrated marketingcommunication campaign, the better the results will be. On the contrary, using too many media may result in a fragmentation of the communication budget and thus, result in low attention of the consumer and less effect of the campaign. Integrated communication does not mean that many media should be employed but rather, an optimal number of media with different and complementary characteristics and with consistent messages.
3. Research methodology An experimental approach was used to study the effects of multiple- and single-medium campaigns on consumer responses under controlled conditions. One hundred and forty-six students of Erasmus University Rotterdam participated, of which 85 were male and 61 were female. As a reward for their participation, they received a $5 lottery ticket. 3.1. Procedure A complete experimental design takes all combinations and sequences of the three media (television, print, and the Internet) into account. In this study, an incomplete design was used including 3 one-medium campaigns (television only, print only, and Internet only) and six different sequences of three-media campaigns. This resulted in nine conditions with about 15 participants in each condition. Each participant was exposed three times to news items of the Rotterdam Broadcast and three times to the two target ads and two filler ads. In all conditions, participants were exposed to the target ads three times. The exposure (external pacing) to television was 8 minutes to the news and 2 minutes to the ads. The opportunity to see (internal pacing) the print ads and the Internet ads (and banners) was 10 minutes. During these 10 minutes, participants could read the news and the ads. The participants were told that the news items were sponsored and that there was an arrangement with the sponsors that all participants could make use of the promotions offered in the ads. After exposure to the news items, participants filled out a questionnaire to assess their responses to the ads. 3.2. Stimuli The target ads were for a book and a brand of wine. Ads for unfamiliar brands were obtained and developed in cooperation with an advertising agency and an audiovisual company. Using unfamiliar brands eliminates the differences between the participants due to prior knowledge or
380
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
attitudes associated with existing brands and ads (Edell and Keller, 1998). The filler ads were not related to the target ads. The ads were pretested on appreciation. Three news broadcasts from a local news station were recorded and edited. Based on the informational content of the news broadcasts, the newsletter and news websites were constructed. The informational content of the three news items differed within a condition to get or to keep the attention of the participants. Exposing participants three times in a row to the same news program might diminish their attention to the news while probably increasing their attention to the ads, thereby, distorting the results of the experiment. The book and the wine ad were selected because these products are interesting for the students participating in the experiment and are low-risk purchases for them. The book ad contains a picture of the cover of the book (Milan Kundera’s The Unbearable Lightness of Being), pictures of the leading figures, a voice-over telling what the book is about and that the book is available at a reduced price, and music to create an attractive ambiance. The wine ad consists of a picture of the wine bottle with its label (Oude Kaap, a South African wine), pictures of wine fields, a voice-over giving information about the wine and the price, and music. The book title and the wine label are the ‘‘brands’’ in these ads. The print and Internet ads were consistent with the TV ads. The pictures in the TV ad were used in the print and Internet ads. Furthermore, the spoken words were used as written copy in the ads. Participants in the print- and television-only conditions received a coupon to buy the product at a reduced price. The news websites were in an Intranet environment and consisted of news pages and advertising pages. On each news page, banner ads served as a link to the ad pages. The ad pages contained additional information about the brands and included a response device to order the product at a reduced price. The banners consisted of a picture from the ad, the brand name, and price. The place of the four banners on the news pages differed randomly. The logfile was used to trace how participants moved through the website, how long they stayed at a certain page or ad, and whether they ordered the product. 3.3. Dependent variables The dependent variables of this study are cognitive, affective, and conative responses. Cognitive responses are knowledge items and inferences about the exposed messages. Affective responses are evaluations and emotions related to the exposed messages. Conative responses are behavioral responses to advertising such as purchase and purchase intention. 3.3.1. Cognitive responses The three measures used to assess the cognitive responses are the number of brand claims correctly recalled, the number of ad elements correctly recalled, and brand
claim recognition. Participants were asked to write down everything they remember about the advertised brand (its features, benefits, and uses) and the ad itself to assess the number of brand claims and the number of ad elements correctly recalled. In addition, brand claim recognition measures were collected with seven-point scale items from ‘‘very certain’’ to ‘‘very uncertain.’’ The items contained claims that were or were not mentioned in the ads. 3.3.2. Affective responses Participants were asked to list all thoughts, reactions, and ideas that went through their minds while watching or reading the advertisements. The total number of evaluative thoughts (i.e., the sum of negative and positive thoughts from the thought listing procedure) captured affective responses. In addition, the affective reactions toward the ads were assessed with 4 seven-point semantic differential scales: clear/confusing, interesting/uninteresting, appealing/ unappealing, and likable/dislikable. The affective reactions toward the advertised products were measured with 4 sevenpoint semantic differential scales: high/low quality, likable/ dislikable, good/bad, and pleasant/unpleasant (Edell and Keller, 1998). 3.3.3. Conative responses All participants were asked for their intention to buy the advertised products on a seven-point scale (extremely unlikely to extremely likely). The number of emails received via the Internet and the price coupons turned in captured purchase. The coupons were numbered and could be related to the questionnaires. 3.4. Covariates Three covariates that may affect ad response were also measured. These covariates were product knowledge, product involvement, and general attitude toward advertising (Lutz, 1985; Edell and Keller, 1989). Participants with prior knowledge of the product category and with more favorable attitudes toward advertising in general have more positive attitudes (Edell and Keller, 1989). Participants were asked to indicate their knowledge about the product category on a seven-point scale from ‘‘not at all knowledgeable’’ to ‘‘very knowledgeable.’’ Product involvement may influence the processing intensity and may lead to stronger positive or negative. The two facets of consumer’s product involvement, perceived product importance, and interest (McQuarrie and Munson, 1992) are measured with 2 seven-point scales (very important/very unimportant and very interested/very uninterested). A two-item scale, rather than a lengthy 20item Personal Involvement Inventory attitudes (Zaichkowsky, 1985), is sufficient for measuring involvement with concrete and singular products such as a book and wine. With too many items, we incur the risk of conditioning and satiation of participants. Conditioning and satiation reduce the validity of the measurement. Thus, a multi-item measure
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
381
may increase the reliability of the measurement but decreases the validity of the measurement (Rossiter and Kayande´, 1999). The ‘‘general attitude toward advertising’’ covariate was formed by averaging participants’ responses to three items measuring, on seven-point scales, how informative, enjoyable, and useful they judge advertising in general.
covariates product involvement and product knowledge. First, the single-medium conditions are compared on cognitive, affective, and conative responses to test H1a and H1b. Subsequently, we compare the single-medium conditions with multiple-media conditions on consumer responses to test H2.
3.5. Coding procedure
4.1. Single-medium campaigns: television, print, and the Internet
The thoughts listing and recall measures were coded by two graduate students. A third person solved all disagreements. The thoughts were coded on two aspects (i.e., whether the thoughts were product- or ad-related and whether the thoughts were positive, negative, or neutral). The positive and negative thoughts were classified as affective thoughts. The agreement measure Cohen’s kappa was .82. They also coded the open-questions ‘‘recall of brand claims’’ and ‘‘ad elements.’’ The Cohen’s kappa for book claims recall was .88, for wine claim recall .91, for book ad recall .58, and for wine ad recall .75. 3.6. Reliability To test the internal consistency of the multiple-item scales, Cronbach’s alphas were computed for book affect (=.92), wine affect (=.89), book ad affect (=.76), wine ad affect (=.77), book involvement (=.90), wine involvement (=.88), and general attitude toward advertising (=.66). The low alpha for general attitude toward advertising shows that this concept is not unidimensional. For this reason, this variable is excluded from further analysis. The values of the other alphas indicate a good internal consistency.
4. Analysis and results Analysis of covariance was used to correct for any linear relationship between the dependent variables and the
4.1.1. Cognitive responses Three measures are used to assess cognitive responses: brand claim recall, ad recall, and brand claim recognition. The means and significance tests for the media conditions on the cognitive responses are presented in Table 1. We did not specify which medium would be more effective in generating cognitive responses. The analysis of covariance of the number of correctly recalled brand claims shows that participants in the TV-only condition recalled significantly more brand claims than participants in the Internet-only condition (book, P < .001; wine P < .001). Although the TV-only condition has a higher mean for brand claim recall, the difference between the TV- and print-only conditions is marginally significant for the book (P=.090) and significant for the wine (P < .001). No significant differences in means between the print- and the Internet-only condition are found. Regarding ad recall, participants in the TV-only condition produced more recall of ad elements than in the Internet-only condition (book: P < .001; wine: P < .001). For the book ad, print-only participants had a higher ad recall than Internet-only participants ( P < .001), but the difference between the television- and print-only participants was not significant. For the wine ad, ad recall of the television-only condition was also significantly higher than the print-only condition. No significant difference between the Internet- and print-only conditions was found.
Table 1 Means for cognitive response measures Media exposure1
Internet only Print only Television only I–P–T I–T–P P–I–T P–T–I T–I–P T–P–I T–T–P abc. . .n 1 2
Book
Wine
Brand claim recall
Ad recall
Brand claim recognition
Brand claim recall2
Ad recall2
Brand claim recognition2
1.80abcdef 3.40 5.00a 4.60b 4.31c 4.93d 3.43 4.27e 3.33 4.75f
2.40abcdefghi 6.47a 8.08bjklmn 5.73cj 5.69dk 6.13e 4.71flo 5.20gm 4.73hnp 7.25iop
4.51abcd 4.29efghij 5.08 5.16e 5.62af 5.64bg 4.90 5.38ch 5.09i 5.56dj
2.38ai 2.20bj 5.51abcdefgh 3.28ck 2.88dl 3.47e 3.66f 2.35gm 3.39h 4.89ijklm
4.12abc 4.23dk 9.31adefghij 6.06eo 5.01bfl 6.06gp 5.70h 4.00imop 5.77j 7.53cklmn
5.04 5.00 5.59abc 5.07 5.13 4.91a 4.90b 4.67cd 5.09 5.52d
Means with the same letter within a column differ significantly from each other. I = Internet; P = print; T = television. Brand claim recall, ad recall, and brand claim recognition for wine is adjusted for wine involvement.
382
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
The analysis of covariance of brand claim recognition indicates no differences between the print- and Internet-only conditions for both products. Marginal differences were found between print- and TV-only conditions (book: P=.055; wine: P=.085).
4.1.3. Conative responses Analysis of covariance of purchase intention shows a significant main effect of product involvement ( Fbook = 4.94, P=.028; Fwine = 52.07, P < .001) (Table 2). On the other hand, no significant main effect of medium type is obtained ( Fbook=.62, P=.795; Fwine=.88, P=.549). Similar to our findings for brand affect, product involvement seems to determine purchase intention.
4.1.2. Affective responses To assess the affective responses, the following three measures were used: total evaluative thoughts, affect towards the brand, and affect towards the ad. Table 2 contains the means on these measures and significance tests for all media conditions. The analysis of total evaluative thoughts shows that in the Internet-only condition, as compared with the TV- and print-only conditions, less evaluative thoughts were given. In contrast to H1a, TV-only participants did not have significantly more affective thoughts than the Internet- and print-only participants. Between the single-medium conditions, no significant differences were found for brand affect and ad affect. Contrary to the literature (e.g., Chaudhuri, 1996), analysis of covariance of brand affect indicated no significant main effect of medium type ( Fbook=.830, P=.590; Fwine = 1.38, P=.204) and no more affective reactions for the TV-only participants. However, the analysis showed a significant main effect of the covariate ‘‘product involvement’’ ( Fbook = 8.73, P=.004; Fwine = 21.64, P < .001), consistent with theory (Zaichkowsky, 1985). In our experiment, product involvement seems to determine affect towards the product. The higher the product involvement, the more extreme, positive or negative, the brand affect. Brand affect seems to be a function of product involvement rather than medium type. The analysis of ad affect indicates a marginal significant main effect of product involvement ( Fbook = 3.42, P=.072; Fwine = 3.52, P=.063). Product involvement influences ad affect. Thus, product involvement seems to affect both ad affect and brand affect.
4.2. Single- versus multiple-media campaigns In H2 it was stated that multiple-media campaigns are more effective than single-medium campaigns, both with the same number of message exposures. The results of the experiment will be presented and the most important differences between single- and multiple-media conditions will be discussed. 4.2.1. Cognitive responses For both products, the multiple-media conditions, as compared to the Internet-only condition, have higher means on brand claim recall and ad recall, however, most of these differences are not significant for wine (Table 1). Similarly, no significant differences for brand claim and ad recall were found between the print-only condition and the multiplemedia conditions. The analysis of covariance of the cognitive responses for wine indicates a dominance for television-only campaigns. Participants in the television-only exposure recalled significantly more brand claims and ad elements and had better brand claim recognition than the participants in the three-media exposures. Likewise for the book, the findings indicate a dominance for televisiononly condition although not significant for brand claim recall and recognition. For both products, television-only participants recalled significantly more ad elements than the multiple-media participants.
Table 2 Means for affective and conative response measures Media exposure1
Internet only Print only Television only I–P–T I–T–P P–I–T P–T–I T–I–P T–P–I T–T–P abc. . .n 1 2
Book
Wine
Total evaluative thoughts
Brand affect2
Ad affect2
Purchase intention2
Total evaluative thoughts
Brand affect2
Ad affect2
Purchase intention2
1.47abc 2.40 2.15 3.00a 2.38 3.07b 2.00 2.20 2.87c 2.12
3.54 3.31 2.81a 3.57 4.00b 2.71bc 3.89 3.39 3.52 4.06ac
4.07 3.86 4.58 4.41 4.29 4.13 4.12 4.27 4.34 4.56
1.79 1.48 1.82 1.99 2.15 1.35 1.52 1.80 2.17 1.85
1.54abc 2.80a 2.62 2.20 2.00 2.93b 3.07cd 2.54 1.93d 2.37
5.46abc 4.98d 4.74 4.03adefg 5.02e 5.23fh 4.80 5.12g 4.37b 4.32ch
4.67 4.65 5.05a 4.78 4.13b 4.73 4.94 5.17bc 4.50 4.14ac
3.45 3.90 3.44 2.93a 4.13 3.73 4.34abc 3.89 2.99b 2.74c
Means with the same letter within a column differ significantly from each other. I = Internet; P = print; T = television. The means of the dependent variables are adjusted for product involvement.
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
4.2.2. Affective and conative responses For both products, the analysis of covariance did not reveal any significant differences on affective and conative responses between the single- and the multiple-media conditions (Table 2). Hence, H2 is not confirmed. However, compared to some of the multiple-media conditions, the number of evaluative thoughts were significantly lower in the Internet-only condition.
5. Discussion A laboratory experiment was conducted examining the effects on consumer responses of single- and multiple-media campaigns consisting of television, print, and the Internet. 5.1. Cognitive responses The results show that, compared with print and the Internet, television is superior in evoking cognitive responses. This is consistent with the findings of Kisielius and Sternthal (1984) and the view that larger number of sensory modes lead to more effective communication (Jacoby et al., 1983). However, this does not agree with the general notion that television is a low-involvement medium evoking less cognitive responses. Compared with print media and the Internet, television, due to the combination of visual and audio modes, evokes more attention and has thus more impact. As a consequence, probably even low-involved participants processed the brand information of the television ads. This is consistent with Krugman’s (1965) ‘‘learning without involvement’’ of television ads. Furthermore, in contrast to participants in the print- and Internet-only conditions, participants in the TV-only condition were forced to watch the news items and, thus, the ads. With print and the Internet, the processing of information is more under personal control (internal pacing) and thus requires more active and involved participants. Low-involved participants can easily skip the ads and read the news items only. The finding that print-only participants reported more cognitive responses than the Internet-only participants may be explained by differences in the confrontation situation. Both media are retrieval media, though, compared to the Internet, print has some ‘‘delivery’’ characteristics. By turning over the pages of the news magazine, participants were forced to notice the book ad and this might have caused the higher cognitive responses for print-only than for Internet-only. Dre`ze and Hussherr (1999) showed empirically that banner advertising has a significant impact on consumers. In our experiment, participants were forced to notice the banner ads on the news pages and therefore, we might expect them to recall the information on these banners (i.e., the brand name, a picture from the ad, and the price). However, Dre`ze and Hussherr (1999) also found that Internet users tend to avoid banner ads and that only 50%
383
of banners are attended to. Another explanation for the finding that print media evoke more cognitive responses than the Internet, is the different delivery mechanism. As Sundar et al. (1998) indicated, a print ad could be processed in its entirety, while the Internet ad has to be scrolled. As a consequence, the processing of Internet ads requires even more involved and active consumers than the processing of print ads. We expect that multiple-media campaigns are more effective than single-medium campaigns (H2). For cognitive responses, this is only true comparing multiple-media, and Internet-only campaigns. However, for multiple-media campaigns, compared to the print- and TV-only campaigns, the findings do not support this hypothesis. The higher means for the TV-only compared to multiple-media conditions indicate the presence of a ‘‘replay effect’’ due to forced exposure. Participants in the TV-only condition were forced to see the ad three times, while the participants in the multiple-media condition were only forced to see the ad once (TV) and could decide for themselves to have a second or third look at the ad (with print and the Internet). In subsequent experiments, we tested the replay effect in two ways. First, in a condition in which participants saw two television ads and one print ad (T– T –P), T –T – P participants reported less cognitive responses than television-only participants but more cognitive responses than participants in three-media conditions (Table 1). Second, we conducted a small experiment with 17 MBA students to test the effect of the confrontation situation. We showed a subgroup of students the TV commercial three times and another subgroup the television, print, and the Internet ads (T – P– I), both times without the news items. If the confrontation situation is kept constant, there are no significant differences in cognitive responses. In the threemedia condition, the means of brand claim recall (M = 6.50) and ad recall (M = 9.00) were higher than in the televisiononly condition (Mbrand recall = 5.60; Mad recall = 6.20) although not significantly. The cell means did not differ on brand claim recognition. These findings indicate that, if the confrontation situation is kept constant, multiple-media campaigns may be as effective as television-only campaigns in communicating brand claims. The superiority of television is likely to depend on the forced-exposure character of this medium and not on its inherent medium power. 5.2. Affective responses For the product wine, participants of the Internet-only condition, as compared to television- and print-only conditions, reported significantly less evaluative thoughts. For the product book, no significant differences were found, but the findings point into the same direction. This finding may be a result of a novelty effect, since online advertising is relatively new and participants may associate the Internet as a channel for news and information (Sundar et al., 1998) and
384
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
therefore, may have paid more attention to the news than to the ads. Another plausible explanation is that the difference in evaluative responses is due to the delivery mechanism (Sundar et al., 1998). No significant medium effect was obtained for brand and ad affect. In contrast to other studies (e.g., Chaudhuri, 1996), our results show no superiority of television in evoking affective responses. Therefore, H1a is not confirmed. For both products, the analysis of covariance did not report any significant differences in affective responses between single- and three-media conditions, with the exception of the Internet-only condition on evaluative thoughts. H2 is thus not supported. An explanation is the strong product-involvement effect on ad and brand affect, overruling the weaker medium effect on ad and brand affect. 5.3. Conative responses According to H1b, we expected participants in the Internet-only condition, as compared to the televisionand print-only conditions, to have more conative responses because products can be purchased directly via the Internet. We did not expect that the Internet as a medium is more effective to persuade consumers (measured by purchase intention), but that if people have been persuaded, the Internet facilitates purchase. In the experiment, 3 participants ordered the book and 13 participants ordered the wine. The frequencies of ordering the products (via the Internet or a coupon) showed that the majority ordered the products by use of the Internet. These figures indicate that the Internet facilitates purchase.
6. Implications Media planners mainly focus on quantitative media planning, that is, the selection of media in a campaign based on the minimization of the costs of these media to reach the target group with the required message frequency. With the advent of new and better-focussed media, media planners should consider, besides this efficiency perspective, an effectiveness perspective. Qualitative media planning takes an effectiveness viewpoint, and is concerned with the fit between the message and the media. The ‘‘proper’’ media, media combinations, and sequences should be selected for the campaign, in terms of how consumers process the message and respond to it. The results of this study indicate that multiple-media campaigns can be as effective in communication as TV- and print-only campaigns. This is a positive finding in the sense that multiple-media campaigns can be as effective as TVand print-only campaigns at a lower cost. Multiple-media campaigns may have lower costs of contact, due to a larger reach and the use of media with lower placement costs, as well as more effectiveness of contact with the target group.
This is only valid for large communication campaigns, because the initial costs of developing advertisements, commercials, and websites must be recovered. Note however that using too many media in a marketing-communication campaign might result in low attention and thus a lower effectiveness of the campaign. Although Internet-only campaigns seem to be hardly effective, the findings for the multiple-media campaigns indicate that the Internet may fulfill a complementary function after traditional media created awareness and interest for the advertised product. The main disadvantage of the Internet as a medium is that it is difficult to get target consumers to the website. As a consequence, the Internet needs traditional media for reaching the target groups, creating attention, and communicating the Internet address (URL). The Internet is thus not a substitute for traditional media such as print and television, but the Internet complements the media plan. Traditional media are needed to complement for the weakness of the Internet to attract attention, whereas the Internet complements for the weakness of traditional media to facilitate purchase. Finally, media planners need to take into account consumer’s product involvement. Consumer involvement is crucial for the differential effects of retrieval (print and Internet) and delivery (television) media. Retrieval media will be effective under conditions of high rather than low product involvement, because retrieval media have limited opportunity to influence uninvolved and passive consumers. Contrary to retrieval media, delivery media are better suited for influencing uninvolved consumers. Delivery media can attract the attention of uninvolved consumers and, as a consequence, may induce superficial processing (Buchholz and Smith, 1991) and low-involvement learning (Krugman, 1965). Note that delivery media may also be effective under high-involvement conditions.
7. Limitations The limitations of this study are similar to other laboratory studies investigating different media: the forced-exposure situation and the short time interval between ad exposure and answering the questionnaire. As in other advertising studies, the measurement of affect through questionnaires has its limitations. In this study, participants gave their cognitive responses before their affective responses written down in a questionnaire. The time interval between their affective experiences and their responses provided ample opportunity to think about these experiences and to rationalize them. The ‘‘cognitive’’ and self-reporting way of measuring affective responses with a questionnaire may also lead to report a ‘‘reasoned affect’’ rather than a primary, spontaneous affect. This may explain the small or nonsignificant effects of medium type on brand and ad affect. In further studies, simultaneous affect measurement (with ‘‘liking’’ or ‘‘disliking’’ buttons or with
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
physiological measurement) may provide a more valid indicator of the affect people experience while being confronted with a message in different media. This is especially relevant for measuring affect towards the ad. Another limitation concerns the difference in confrontation situation. In our experiment, participants were forced to watch the TV ads, print ads, and (Internet) banner ads. However, participants were able to decide themselves whether to click-through to the advertising pages behind the banners on the Internet. This affected the results and led to less cognitive responses for the Internet-only condition. Research showed that banner advertising has a significant impact on consumers in terms of brand awareness and advertising recall even in the absence of click-throughs (Dre`ze and Zufryden, 1999). Participants in our experiment, who did not click through, were exposed to less information than the print- and TV-only participants. Participants exposed to the banners only could only recall the brand name, the picture (also shown in the TV and print ads), and the price of the product. Although, the confrontation to ads was held constant across media, the confrontation to the informational content was not. In future experiments, the confrontation to the informational content of ads should be held constant across media to compare the processing differences of media in a more valid way. Finally, although 146 students participated, the total number of participants per cell was still relatively small. This decreases the chance to detect small main effects and interactions.
385
campaigns and communication effectiveness. The number of media in an integrated campaign might also be a function of reach, message complexity, duration of a campaign, and consumer involvement. Although in this study multiple-media campaigns do not significantly differ from each other in evoking the different types of responses, some sequences of media had higher means (e.g., the sequence P– I– T). The effectiveness of different sequences in a multiple-media campaign needs further investigation because the complementary effect may be due to sequential integration of media in a campaign. We expect that the effectiveness of a certain sequence of media depends on the type of product, message complexity, communication objectives, and, most importantly, consumer information processing. More research is necessary to gain a better understanding of the unique characteristics of the Internet as an advertising medium, as well as its role in multiple-media campaigns. Currently, the Internet fulfills a complementary role to traditional media. However, changes in technology and consumers’ perceptions of and attitudes toward the Internet may alter the function of the Internet as a communication medium. In the future, the Internet may even become a viable alternative to traditional media. Finally, consumer responses to multiple-media campaigns have to be investigated in a field experiment to obtain higher external validity. In such a setting, media exposure will be more similar to ‘‘real-life’’ media use by consumers. In real life, the advertiser has less control over the frequency and sequence of media exposures. Synergy effects may thus be more difficult to reach.
8. Directions for future research The effects on consumer responses were studied of singleand multiple-media campaigns consisting of print, television, and the Internet. With the advent of new media, investigation of the contributions of media to advertising effects will become more relevant for media planners and advertisers. Especially, the strengths and weaknesses of media in a campaign, and thus their complementarity, need further investigation. Moreover, investigating consumer responses to other media (e.g., radio and outdoor media) and other combinations of media is necessary to get more insights in the effects of multiple-media campaigns. Future research should examine under what conditions positive or negative synergy effects are obtained from multiple-media campaigns. Several factors might mediate the impact of multiple-media campaigns, such as the complexity of the message and consumer’s involvement with the product. For example, a complex message communicated by multiple media might confuse the consumer and therefore, might lead to lower communication effectiveness. An interesting research question related to this is: ‘‘What will be the most effective number of media to be used in a multiple-media campaign?’’ If too many media are included, fragmentation will occur. We expect a curvilinear relationship between the number of media in multiple-media
Acknowledgements We thank Wayne Hoyer, Kaj Morel, and Alain Strazzieri for their valuable comments, Martine van Iperen for designing the websites, Patrick Ubags for preparing the videos, and StadsRadio Rotterdam for recording the voices.
References Broadbent DE. Perception and communication. New York: Pergamon; 1958. Buchholz LM, Smith RE. The role of consumer involvement in determining cognitive response to broadcast advertising. J Advert 1991;20(Spring): 4 – 17. Chaudhuri A. The effects of media, product and message factors on ad persuasiveness: the role of affect and cognition. J Mark Commun 1996;2(December):201 – 18. Deutsch JA, Deutsch D. Attention: some theoretical considerations. Psychol Rev 1963;70:80 – 90. Dre`ze X, Hussherr F-X. Internet advertising: is anybody watching? Working paper, University of Southern California, Los Angeles (CA); 1999. Dre`ze X, Zufryden F. Internet advertising: the medium is the difference. Working paper, University of Southern California, Los Angeles (CA); 1999.
386
M. Dijkstra et al. / Journal of Business Research 58 (2005) 377–386
Edell JA. Nonverbal effects in ads: a review and synthesis. In: Hecker S, Stewart DW, editors. Nonverbal communication in advertising. New York: Lexington Books; 1988. p. 11 – 27. Edell JA, Keller KL. The information processing of coordinated media campaigns. J Mark Res 1989;26(May):149 – 63. Edell JA, Keller KL. Analyzing media interactions: print reinforcement of television ad campaigns. Working paper, Fuqua School of Business, Duke University, Durham (NC); 1998. Jacoby J, Hoyer WD, Zimmer MR. To read, view, or listen? A cross-media comparison of comprehension. Current issues and research in advertising. Ann Arbor (MI): University of Michigan; 1983. p. 201 – 17. Kisielius J, Sternthal B. Detecting and explaining vividness effects in attitudinal judgments. J Mark Res 1984;21(February):54 – 64. Krugman HE. The impact of television advertising: learning without involvement. Public Opin Q 1965;29:349 – 56. Lutz RJ. Affective and cognitive antecedents of attitude toward the ad: a conceptual framework. In: Alwitt LF, Mitchell AA, editors. Psychological processes and advertising effects: theory, research and applications. Hillsdale (NJ): Erlbaum; 1985. p. 45 – 63. McQuarrie EF, Munson JM. Product involvement inventory: improved usability and validity. Adv Consum Res 1992;19:108 – 15.
Paivio A. Imagery and verbal processes. New York: Holt, Rinehart and Winston; 1971. Pieters RGM, van Raaij WF. Reclamewerking (How Advertising Works). Houten: Stenfert Kroese; 1992. Rossiter JR, Kayande´ U. Construct measurement in marketing. Working paper, Australian Graduate School of Management, Sydney, Australia; 1999. May. Sundar SS, Narayan S, Obregon R, Uppal C. Does web advertising work? Memory for print versus online media. Journal Mass Commun Q 1998;75(Winter):822 – 35. Triesman AM. Verbal cues, language and meaning in selective attention. Am J Psychol 1964;77:206 – 19. Vakratsas D, Ambler T. How advertising works: what do we really know? J Mark 1999;63(January):26 – 43. Van Raaij WF. Interactive communication: consumer power and initiative. J Mark Commun 1998;4(March):1 – 8. Wright PL. Message-evoked thoughts: persuasion research using thought verbalizations. J Consum Res 1980;7(September):151 – 75. Zaichkowsky JL. Measuring the involvement construct. J Consum Res 1985;12(December):341 – 52.
Journal of Business Research 58 (2005) 387 – 396
Interactivity and vividness effects on social presence and involvement with a web-based advertisement David R. Fortina,*, Ruby Roy Dholakiab a
Department of Management, University of Canterbury, Christchurch, New Zealand b University of Rhode Island, Kingston, RI, USA
Abstract This article measures the effects of various levels of interactivity and vividness of a message on attitudes and behavioral intentions within a web-based advertisement. As a conceptual foundation, the study introduces the multistep model of the impact of interactivity on advertising effectiveness. The model is termed multistep because of the hierarchy of direct and indirect effects. A 3 3 interactivity by vividness between-subjects factorial design tests the model. A total of 360 responses were collected through an online web interface. Data were analyzed by means of individual analysis of covariance (ANCOVA) procedures. The multistep model was also tested with path analysis to verify the significance of the interrelationships between constructs in a simultaneous equations procedure. Results indicate moderate effects of interactivity and vividness on social presence and, indirectly, involvement that in turn have strong effects on traditional advertising effectiveness measures. The findings suggest that the effects of interactivity reach a ‘‘plateau’’ at medium and high levels, indicating a diminishing returns effect. Conversely, the impact of vividness appears to be linear with a steady increase across low, medium, and high levels. No interaction effect was found between the two treatments. The study also provides some insights on using the web as a gateway for experimental research and data collection. D 2003 Elsevier Inc. All rights reserved. Keywords: Interactivity; Vividness; Involvement; Advertising; Social presence; Web
1. Introduction For over a century, advertisers have used a variety of media to reach specific target markets. With the advent of each new medium, advertisers created messages that exploited the media’s physical capabilities, such as color, audio, and moving images (video). The introduction of computer-mediated environments (CME) has stimulated advertisers to establish commercial web sites but there are still unanswered questions regarding the effective use of this new medium. Early web sites were mostly text-based, but technological developments now allow for a variety of features that can enhance the richness of the interface. Advertisers have the opportunity to combine the advantages of print and electronic media and allow for control of pace and exposure to customized information as well as more vivid forms of communication (motion, audio and video files, etc.). As a new conduit for advertising, CMEs raise the
* Corresponding author. Tel.: +64-3-364-2987x7026. E-mail address: [email protected] (D.R. Fortin). 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(03)00106-1
question as to how advertisers will use the capabilities of the new media to achieve their communication objectives. To answer this question, we need to understand how the new media are similar and/or different from traditional media and how these differences might influence advertising effectiveness. Interactivity is one of the key characteristics of the new media that is expected to not only transform the way advertising is designed and implemented but also the manner in which it affects consumers’ opinions and attitudes. As technology evolves, we will probably see an increase in the interactive capabilities of new media approximating the level reached by face-to-face (FtF) interactions. Given that most interactive communication attempts historically have failed in the marketplace (Neuman, 1991) and that the effects of interactivity have been inconclusive in academic research (Shaw et al., 1993), this underscores the need for further investigation regarding optimal levels of interactivity. The research objectives of this article are to provide a review of interactivity and evaluate the effects of incremental levels of interactivity on attitudes about an online advertisement.
388
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
2. Literature review
Table 1 Review of literature on interactivity (partial list)
One could argue that the web is simply another medium to convey commercial information along with other media, such as television, radio, print, etc. However, some researchers believe that the interactive nature of the web, i.e., the ability for the user to respond and react, creates a totally new environment that changes the traditional parameters of mass communication within a CME (Hoffman and Novak, 1996). Central to the issue is the interactive nature of a CME. Many researchers have defined interactivity as it relates to the capabilities of a given technology (Rafaeli, 1988; Williams et al., 1988; Newhagen and Rafaeli, 1996). Because of its unique feature as a medium that cannot only receive but also emit transmissions, the web is labeled as a ‘‘many-to-many’’ channel (Hoffman and Novak, 1996). This new approach clearly differs from the traditional model of mass communication where one source transmits to many receptors in a one-to-many fashion (Rust and Oliver, 1994; Hoffman and Novak, 1996; Venkatesh et al., 1995).
Author (year)
Definition
Wiener (1950)
2.1. Defining interactivity
Williams et al. (1988)
Notion of feedback: a method of controlling a system by reinserting into it the results of its past performance. The capability of new communication systems (usually containing a computer as one component) to offer interchangeability of roles between sender and receiver in real or delayed time thus enabling human control over the pace, structure, and content of the communication. The capability of new communication systems (usually containing a computer as one component) to talk back to the user, almost like an individual participating in a conversation. Within a recursive setting, an expression of the extent that in a given series of exchanges, any later transmission is related to the degree to which previous exchanges referred to, even in earlier transmissions. For full interactivity, communication roles need to be interchangeable. A construct that includes control, exchange of roles and mutual discourse. Control refers to the content, timing, and sequence of communication. Exchange of roles implies that sender and receiver can assume both these roles and mutual discourse that each communication feeds on the previous. Six dimensions of interactivity: 1 = complexity of choices, 2 = level of effort, 3 = media responsiveness, 4 = media monitoring capability, 5 = ability to add information, and 6 = capability for interpersonal communication. The extent to which users can modify the form and content of a mediated environment in real-time. Person interactivity is between humans through a medium and machine interactivity is between human and machine to access hypermedia content. A communication that reflects back on itself, feeds on, and responds to the past.
Due to the rapid growth of the web, there is much ‘‘hype’’ in the popular press about new interactive media. Interactivity is now the buzzword of the day, but the term is used loosely and means many things to many people. The lack of a rigorous definition is problematic for researchers who want to study this area in a more systematic fashion. The definition of interactivity by Williams et al. (1988) is a three-dimensional construct. It includes control, exchange of roles, and mutual discourse. Newhagen and Rafaeli (1996) determine five qualities of communication on the Internet: multimedia, hypertextuality, packet switching, synchroneity, and interactivity. Hoffman and Novak (1996) distinguish two levels of interactivity: person interactivity that occurs between humans through a medium and machine interactivity that occurs between humans and machines to access hypermedia content. Various definitions of interactivity are briefly summarized in Table 1. For the purpose of this research, we propose the following definition of interactivity:
Rice and Associates (1984)
Rogers (1986)
Rafaeli (1988)
Heeter (1989)
Steuer (1992)
Hoffman and Novak (1996)
Newhagen and Rafaeli (1996)
2.2. Interactivity continuum the degree to which a communication system can allow one or more end users to communicate alternatively as senders or receivers with one or many other users or communication devices, either in real time (as in video teleconferencing) or on a store-and-forward basis (as with electronic mail), or to seek and gain access to information on an on-demand basis where the content, timing and sequence of the communication is under control of the end user, as opposed to a broadcast basis.
Based on the proposed definition of interactivity, it becomes apparent that different media can display different levels of interactivity and hence the construct cannot be simply categorized as dichotomous. As Heeter (1989) clearly points out, a single medium can exhibit more than one function and within that, some functions might be at different levels of interactivity. Within a highly interactive media, there are also opportunities for varying levels of interactivity. For example, a web site could simply be a page
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
of text without any links, feedback options, or search engine. This site would score very low on the interactivity continuum although it is part of a potentially highly interactive medium. Conversely, some television broadcasts, such as infomercials and shopping channels, allow for immediate response through the use of toll-free telephone numbers or on-the-air live interaction with the show’s hosts; this specific broadcast within the (traditionally noninteractive) television medium would score much higher on the interactivity continuum. Rafaeli (1990) argued that for traditional mass media, once-passive audiences are now more active with letters to the editor, on-the-air talk shows, infomercials, etc. Hence, interactivity is a quality of a communication setting, which can vary within the same medium (Rafaeli, 1990). 2.3. Correlate of interactivity: vividness Vividness relates to the breadth and depth of the message: breadth being the number of sensory dimensions, cues, and senses presented (colors, graphics, etc.), and depth being the quality and resolution of the presentation (bandwidth) as defined by Steuer (1992). It is also referred to as media richness (Daft and Lengel, 1986). Also part of Hoffman and Novak’s (1996) model, vividness is often mistaken for interactivity (Rafaeli, 1988; Steuer, 1992). It differs, however, on the capacity for two-way communication; in fact, certain pieces of communication can be highly vivid but noninteractive (e.g., television, magazine). Similarly, certain forms of communication can be highly interactive but also be low in vividness. Using the case of e-mail, the level of interactivity can fluctuate whether it is part of a oneon-one, Listserv, or newsgroup communication. There appears to be ample evidence to suggest that the interactivity construct is in dire need of a clear conceptualization and many researchers have expressed this view (Rafaeli, 1988; Heeter, 1989; Morris and Ogan, 1996).
389
persuasion where involvement is part of the process, the elaboration likelihood model (ELM) of Petty et al. (1983) is particularly appropriate in this context since interactivity has the potential to affect involvement and arousal.
3. Conceptual framework 3.1. Multistep model of the impact of interactivity on advertising effectiveness Fig. 1 displays the conceptual multistep model of the impact of interactivity on advertising effectiveness. The model is termed multistep because of the hierarchy of direct and indirect effects of interactivity on advertising effectiveness. Along with interactivity, vividness is a quality of a communication setting that is part of the multistep model. Because it is often mistaken for interactivity (Steuer, 1992; Hoffman and Novak, 1996), it is critical to consider its effects to verify the discriminant validity of the interactivity construct. The model considers interactivity and vividness as characteristics of communication settings that can, either directly, or indirectly through social presence and involvement, affect arousal that in turn affects attitudes. Hence, the following hypotheses are presented. 3.2. Hypotheses 3.2.1. Direct effects of interactivity As mentioned previously, interactivity is best defined as a variable quality of a communication setting, not necessar-
2.4. Understanding consumer use of interactive media Some studies seem to suggest that although interactive technologies may be available, the consumer interest in using them has been disappointing (Lee and Lee, 1995; Neuman, 1991). Studies that have manipulated the level of interactivity of a stimulus (Bailey, 1992; Ku, 1992; Shaw et al., 1993; Frazer and McMillan, 1996; Ketanurak, 1996) seem to indicate a weak effect of interactivity on learning and cognition. There is also research evidence showing that enhanced vividness of the medium does not translate into increased persuasion or recall (Taylor and Thompson, 1982; Walther, 1996). The uses and gratifications paradigm appears to be the framework of choice to study CMEs as it asks why people engage in this form of communication and what benefits they derive from it (Eighmey and McCord, 1995; Newhagen and Rafaeli, 1996; Morris and Ogan, 1996). When considering advertising effects on
Fig. 1. Multistep model of the effects of interactivity on advertising effectiveness.
390
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
ily of a media per se. Consequently, within the same medium, it is possible that different executions of an advertisement will exhibit varying degrees of interactivity. Interactivity will have a direct impact on three other variables: social presence, involvement, and arousal. Social presence: Social presence represents the degree to which a medium conveys the perceived presence of communicating participants in the two-way exchange (Short et al., 1976). This is also termed telepresence in Hoffman and Novak’s (1996) model of network navigation. Interactivity is likely to create feelings of social presence for the user through the availability of open channels allowing for two-way communication. Hence, our first hypothesis: H1: There is a positive relationship between the degree of interactivity of an advertisement and the social presence it conveys. Involvement: Because interactivity can yield increased control of the content appearing in the ad and offer the opportunity to communicate with the advertiser and/or other consumers, it is likely to affect an individual’s sense of involvement, not necessarily with the product category (enduring), but mostly with the ad directly (Zaichkowsky, 1994). Thus,
level of involvement (Zaichkowsky, 1994) of the user with the ad. Thus, H4: The relationship between the degree of interactivity of an advertisement and the level of involvement will be mediated by social presence. 3.3. Moderator variables An individual trait variable called need for cognition (NFC), developed by Cacioppo and Petty (1982), is an individual’s intrinsic enjoyment of and motivation to engage in effortful cognitive information processing. This trait is likely to be of research interest in a media, such as the web, where communication is initiated by the end-user. It has also been examined as a moderating variable in the context of the ELM in previous research (Raman, 1996; Zhang, 1996; Haugvedt et al., 1992). Hence, high-NFC individuals are more likely to be positively affected by high interactivity features than are low NFC. Conversely, low-NFC individuals are more likely to rely on external cues, such as vividness of the message, which is consistent with the peripheral route to persuasion.
4. Method H2: There is a positive, direct relationship between the degree of interactivity of an advertisement and the level of involvement with the ad. Arousal: Because interactive features of an advertisement are usually quickly identifiable within the ad, there is a possibility that the mere availability of these features might increase arousal directly, unmediated by involvement. Arousal is a psychobiological trait of human behavior and is referred to here as phasic activation, a short-term reaction of enhanced energy that increases the overall cortical processing of information (Kroeber-Riel, 1979). Arousal can be generated by an increase in motor activity or muscular responses (Zajonc and Markus, 1982; Wells and Petty, 1981) required by the increased interactivity of the ad. But it can also be the result of cognitive or affective reactions to a stimulus (Holbrook and Batra, 1987). This direct link between interactivity and ‘‘flow’’ is also part of Hoffman and Novak’s (1996) model. Hence, H3: A positive, direct relationship exists between the degree of interactivity of an advertisement and the level of arousal (phasic activation) observed during exposure to the ad. Social presence: Because interactivity generates feelings of social presence for the user through open channels allowing for two-way communication, it is likely that the enhanced social presence will impact positively on the
4.1. Research context: advertising on the web The web is chosen as a context for this research because it provides an opportunity to investigate the effects of varying levels of interactivity and vividness on advertising effectiveness. A commercial home page is defined as an interactive advertisement (Raman, 1996, Ducoffe, 1996) that can vary widely in interactivity depending on its design configuration and the specific features it provides (Frazer and McMillan, 1996). For experimental purposes, we now have the capability of manipulating interactivity and vividness levels of both web site- and banner-based forms of advertising. 4.2. Experimental design The research consists of a 3 3 between-subjects factorial design. Three levels of both interactivity and vividness (low, medium, and high) were manipulated as independent variables, generating nine experimental conditions. 4.3. Stimulus material The stimuli for the experiment consisted of a banner along with a web site advertisement for a fictitious product of relevance to web users (a power surge protector called PowerStrip) and of a moderate level of involvement to avoid extreme effects (Raman, 1996). Both ad configurations and
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
data recording forms were created specifically for the experiment by using HTML, the markup language used as a standard of web communication, JavaScript, and CGI scripts. Based on the factorial design, nine versions of the stimulus material representing three levels of interactivity by three levels of vividness were generated. The goal of the study was to manipulate interactivity and vividness while maintaining other factors constant. To this end, the information content of the stimulus material essentially remained constant throughout all experimental conditions and only the interactivity and vividness features were allowed to vary.
Attitude toward the brand Purchase consideration
Covariate measures Need for cognition
Manipulation checks Perceived interactivity
4.4. Procedure 4.4.1. Recruitment of participants A self-recruited convenience sample of web users was used for this experiment. The target population consisted of actual users of the web, aged 18 years old and older. A small incentive in the form of five $50 prizes was offered to stimulate participation. Participants were invited to join the study by clicking through links to the research site that appeared in e-mail Listservs and newsgroups announcements. 4.4.2. Experiment Once they clicked through to the experimental site, participants were randomly directed to one of the nine experimental cells. After examining a banner ad for the product, participants were instructed to click on it to proceed to section two. This section of the experiment was the actual web site for the PowerStrip product. Participants were told to spend as much time as they required going through the test site. The last phase of the experiment asked the participants to respond to our questionnaire, which involved filling out the dependent measures and a short set of background demographics using an easy and friendly ‘‘point-and-click’’ interface. 4.4.3. Instruments
Dependent measures Involvement with the ad Social presence Arousal
Attitude toward the ad
The revised personal involvement inventory measuring involvement with the ad was adapted from Zaichkowsky (1994). The five-item scale developed by Short et al. (1976) was used to measure this construct. The subscale of nine items that measures arousal derived from advertising was pulled from the Standardized Emotional Profile of Holbrook and Batra (1987). An 11-item scale broken down into three components: hedonism (fun to see/not, unpleasant/pleasant, entertaining/not, enjoyable/not), interestingness (important/not, helpful/not, informative/not, useful/not), and utilitarianism (curious/not, boring/not, interesting/not) of the ad (Batra and Ahtola, 1990; Olney et al., 1991).
Perceived vividness
391
A four-item scale measuring attitude toward the brand was adapted from MacKenzie and Lutz (1983) and Raman (1996). A three-item scale adapted from Raman (1996) was used.
Ten items were selected from the revised and shortened version of the 18-item need for cognition scale of Cacioppo et al. (1984).
A six-item scale was created to evaluate the interactivity manipulation based on perceived control of content, mutual discourse, and exchange of roles. Some items were inspired from the media richness scale of Trevino et al. (1987) and the dimensions previously identified in our interactivity definition. Six items were created to measure the strength of the vividness manipulation based on Steuer’s (1992) dimensions of sensory breadth (number of elements) and depth (resolution).
4.4.4. Pretest The purpose of the pretest was threefold. First, to evaluate the performance of the program and the web server in responding to requests for the experiment pages from outside locations. Second, to verify that the experimental procedure was understood well and that instructions and questions were clearly labeled and written. Third, to empirically check the effectiveness of the manipulation for all levels of vividness and interactivity within the stimulus material. Both measures of perceived interactivity and perceived vividness were used as manipulation checks in the pretest so that corrections to the stimulus material could be made before the experiment went into the field. A convenience sample consisting of undergraduate students at a university in the northeast United States was used for the pretest. A total of 76 usable questionnaires were collected through the web server. An e-mail announcement was also sent to all faculty members in the College of Business so they could provide comments on the experimental stimuli and the wording of the questionnaire. 4.4.5. Manipulation checks To examine the effectiveness of both manipulations, two scales for perceived interactivity and perceived vividness were used as manipulation checks. Both scales were normally distributed and performed quite well with ’s of .88 and .90 for the interactivity and vividness scales, respectively. Using the scale means as dependent variables, two separate ANOVAs were conducted to examine the effectiveness of the manipulation. Results indicate that both manipulations were successful [interactivity: F(2,73) = 48.99, P < .001; vividness: F(2,73) = 40.72, P < .001]. The Scheffe’s post hoc test also shows significant mean differences ( P < .05) among all conditions (low, medium, and high) for both scales. In conclu-
392
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
sion, the experimental procedure successfully implemented nine variations of the stimulus material incorporating three levels of both interactivity and vividness. The manipulation checks showed that pretest participants were able to perceive significant differences among all conditions.
5. Analysis and results 5.1. Sample size and composition The survey was left open for a month and a counter on the first page indicated that over 1000 visits were logged in during that period. Of the 397 questionnaires submitted, 385 were usable, which represents a 39% completion rate in traditional survey terminology. To satisfy basic assumptions of statistical techniques and to avoid other problems caused by unequal cell sizes, 25 cases were randomly deleted to obtain an equal cell size design of 40 participants per cell, yielding a total sample size of 360. Prior to analysis, all data were examined for missing values and outlier contamination. No cases were found with extreme or systematic response patterns. Twenty-eight cases had missing values in some variables so a mean replacement procedure was implemented as recommended by Tabachnick and Fidell (1989) when missing values do not exceed 5% of the cases for that variable.
The sociodemographic information about the sample shows a male/female split of 60/40, which is consistent with Internet user demographics (Gupta, 1997). A total of 55% of the sample is in the 25 –44 age group and 46% described themselves as being in the middle third of the household income scale for their country. Most (77%) of the sample came from the United States, with representation from other regions, such as Canada, Australasia, and the United Kingdom. The sample appears better educated than the national average with 47% reporting a graduate or professional degree and 48% working outside the home. A total of 50% of the sample considered themselves experts in Internet usage, 49% access the web between 1 and 10 hour per week and only 27% have started using the Internet after 1996. The main draw to the experimental site came from email Listserv announcements (53%) and 48% accessed the survey from their workplace. Finally, over two-thirds of the sample accessed the site with a Pentium-based PC, 36% with a high-speed connection (e.g., network, T1). Results indicate that both manipulations were successful although the spread of the means across levels is not as large as that found in the pretest. Both manipulation effects are considered large (Cohen, 1977) as evidenced by the 2 statistic for strength of association (2=.25 for interactivity and 2=.44 for vividness). The Scheffe’s post hoc test also shows significant mean differences ( P < .05) among all conditions (low, medium, and high) for both scales.
Table 2 Source table, univariate tests of direct effects Source and dependent variable Need for cognition Involvement Affective subscale Cognitive subscale Social presence Arousal
Type III SS
3.387E 2.402 1.396 8.224 0.830
02 1
df
3 1 1 1 1
Mean square
0.387E 2.402 1.396 8.224 0.830
02
P
2
0.026 1.472 0.862 5.024 1.378
.872 .226 .354 .026 .241
.000 .004 .002 .014 .004
.053 .227 .152 .608 .216
F
Observed power
Interactivity Involvement Affective subscale Cognitive subscale Social presence Arousal
14.082 20.998 9.083 43.475 3.443
2 2 2 2 2
7.041 10.499 4.542 21.738 1.722
5.420 6.434 2.803 13.279 2.857
.005 .002 .062 .000 .059
.030 .036 .016 .071 .016
.844 .902 .550 .998 .558
Vividness Involvement Affective subscale Cognitive subscale Social presence Arousal
13.644 52.380 2.359E 57.508 14.131
2 2 1 2 2
6.822 26.190 0.179E 28.754 7.065
5.252 16.048 0.007 17.565 11.727
.006 .000 .993 .000 .000
.029 .084 .000 .091 .063
.832 1.000 .051 1.000 .994
4 4 4 4 4
1.323 2.390 0.991 1.798 1.261
1.018 1.464 0.612 1.098 2.093
.398 .213 .655 .357 .081
.012 .017 .007 .012 .023
.321 .454 .201 .346 .621
Interactivity by vividness Involvement Affective subscale Cognitive subscale Social presence Arousal
5.290 9.559 3.963 7.191 5.044
02 2
02
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
393
Fig. 2. Profile plots for interactivity at levels of vividness.
5.2. Tests of direct effects on process variables A 3 3 between-subjects analysis of covariance (ANCOVA) is performed for the three dependent process variables (involvement, social presence, and arousal). The two subscales for involvement (affective and cognitive) are also integrated into the analysis. Adjustment is provided by two covariates: NFC and expertise. The two grouping variables (independent) used are interactivity and vividness. The ANCOVA procedure examines main, interaction, and covariate effects at the =.05 level. Standard 2 coefficients are computed to ascertain the individual effect sizes of each independent variable on the dependent variable under examination. The results of the ANCOVA procedure are shown in Table 2. The source table displays the F value, exact P significance, and corresponding strength of association (2) for each effect. The NFC covariate provides a small but significant adjustment for social presence ( P=.026, 2=.01). The interactivity construct, the main variable of interest in our study, has a significant impact on two process variables. The strongest impact is on social presence with a moderate effect (2=.07) followed by a smaller effect on involvement (2=.03), all significant at the P < .05 level. Only the affective component of involvement is significantly affected by interactivity (2=.04) although the cognitive dimension approaches significance ( P =.062). Arousal also appeared marginally affected ( P =.059). The vividness construct exhibits main effects on all three dependent process variables with moderate effects on social presence (2=.09) and arousal (2 =.06) and smaller effects on involvement (2=.03). A further examination of the affective component of involvement reveals a moderate effect of vividness (2=.08) but none for the cognitive component. Based on effect size, it would appear that vividness has greater impact on the affective dimension than interactivity. None of the dependent variables are sensitive to the interaction of interactivity and vividness, only main effects will be examined from this point on. A graphical display of all the significant main effects using interactivity on the horizontal
axis is found in Fig. 2a – c. Visual inspection seems to reveal a direct linear relationship between interactivity and social presence at all levels of vividness. For involvement, there appears to be a ‘‘plateau’’ effect of interactivity for both medium and high levels of vividness where the effect either stabilizes or drops off slightly after a positive effect is found between low and medium levels.
Fig. 3. Results of the multistep model.
394
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
5.3. Tests of the multistep model: path analysis The full multistep model is tested using path analysis to verify the significance of the interrelationships between constructs in a simultaneous equations procedure. Fit is ascertained by the root mean square residual (RMR), the goodness-of-fit index (GFI), and the comparative fit index (CFI), as recommended by Bentler (1991). After optimization, the resulting Model 1 with standardized path coefficients is displayed in Fig. 3. The model exhibits excellent fit with the data. Model 1 (GFI=.972, CFI=.981, RMR=.031) has a 2/df ratio of 2.98, which is within the recommended ratio of 5.00 (2 = 44.74, df = 15, P < .01). In Model 1, the impact of interactivity on social presence is significant ( =.25) as is that of vividness ( =.27); however, both links to involvement are not and the relationship appears to be mediated by social presence ( =.62). In Model 1, involvement ( =.55) and social presence ( =.26) significantly impact arousal as hypothesized in our original model. The most provocative finding resulting from the path analysis is the unmediated and strong impact of involvement on all three measures of advertising effectiveness: attitude toward the ad ( =.62), attitude toward the brand ( =.32), and purchase consideration ( =.28). Our model assumed that arousal would strongly mediate all the antecedent effects but this was not supported.
the literature review. However, few of these studies had manipulated interactivity in an advertising communication context or within a web interface. Our results show a moderate and direct effect of interactivity on the construct of social presence, which in turn has a strong effect on involvement and arousal and ultimately advertising effectiveness measures. 7.2. Linear and nonlinear discoveries
6. Review of hypotheses
All but one significant relationship turned out to be linear in nature. A consistent finding showed ‘‘plateau’’ effects of interactivity on dependent measures between the medium and high levels. Because higher interactivity levels also add to the complexity of the advertisement, it is possible that the observed effect is the result of an ‘‘information overload’’ phenomenon. Previous studies in marketing have shown that the addition of information elements and cues in advertisements do not necessarily translate into better consumer evaluations. A similar effect may be present here, which could suggest an effect of diminishing returns. Our initial contention that arousal is a critical element in the process was not supported, suggesting that the route through involvement might be more typical of an information processing model. Hoffman and Novak’s (1996) model of network navigation proposing that ‘‘flow,’’ which is conceptually stronger than arousal, is a result of both interactivity and vividness is not supported by our findings.
6.1. Direct effects of interactivity
7.3. Effects of covariates
The most significant impact of interactivity is on social presence, the degree to which the stimulus conveyed the perceived presence of participants in the communication exchange. Hence, H1 is supported. Although interactivity does have a significant effect on involvement, the relationship appears to be mediated by social presence; thus H2, which hypothesized a direct link, is not supported by Model 1. The hypothesized direct relationship between interactivity and arousal, resulting from muscular responses or phasic activation is not significant and H3 is rejected.
The NFC covariate had a weak impact generally except for one significant interaction with interactivity in the case of social presence. Those high in NFC displayed a directly proportional relationship between interactivity and social presence while those low in NFC showed a ‘‘plateau’’ effect. The ‘‘overkill’’ phenomenon may not be a problem for those who enjoy cognitive stimulation. Not turned off by increased complexity, those high in NFC actually felt higher social presence levels as interactivity increased. 7.4. Critical role of involvement
6.2. Indirect effects The relationship between interactivity and involvement, as mentioned earlier, appears to be significantly mediated by social presence thus supporting H4.
7. Discussion, conclusions, and implications 7.1. Summary of findings Previous research has consistently demonstrated either weak or nonexistent effects of interactivity as discussed in
Interactivity and vividness both impact significantly on social presence and indirectly, involvement, which turns out to have the most significant effect on all advertising effectiveness measures, greater than arousal, as was initially hypothesized in this study. The pivotal role of involvement discovered here underscores the process by which consumers are likely to interact with the new media, namely a more conservative affective process than was originally thought of, more in line with an information processing perspective and supportive of a uses and gratifications paradigm. If the medium itself is fundamentally more involving than other media from a cognitive point of view, then the incremental
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
level of involvement for a particular advertisement within that medium might draw upon more affective than cognitive dimensions. The interaction effect between NFC and interactivity observed for social presence also supports this position. Those higher in NFC tend to respond more positively to enhanced interactivity than those scoring lower on that dimension. 7.5. Implications for advertisers The results have significant implications for advertisers. First, it appears that new media, such as the web, have the capability of impacting on attitude formation and change and therefore can be interesting and potentially powerful outlets for consumer communication. However, as is often the case with advertising in conventional vehicles, the ‘‘more-is-better’’ approach does not necessarily lead to enhanced communication effectiveness. As complexity of the advertisement increases, so does the possibility of diminishing returns on effectiveness. However, if the interactive features and design elements are properly balanced, the new media have the ability to impact favorably on involvement, which has been traditionally hard to achieve in conventional media. Our findings would suggest that providing enhanced vividness of the message by means of colors, graphics, and animation is more likely to generate a favorable impact than comparable levels of interactivity. The optimal mix would seem to prescribe a moderate level of interactivity (such as navigational links, e-mail forms, etc.) and a high level of vividness. By focusing on the enhancement of social presence elements in the design configuration of the advertisement, firms have the unique opportunity of establishing virtual or perceived relationships with their customers on a one-on-one basis, which is formidably achieved in the context of the web. Advertisers also have the opportunity of exploiting the creative flexibility of the new media by incorporating specific versions of a message targeted to consumers varying in NFC. Although it would be futile to try and measure every consumer coming into a web site on that variable, our findings suggest that at least two versions of interactive messages could be designed for low and high cognitivesavvy groups who are likely to have different expectations in terms of information display and interactive features. An advertiser can strategically design both configurations and offer them as options within a site. 7.6. Directions for future research Our results showed significant effects of interactivity, which underscores the need for further research of this construct with a long trend of nonsignificant findings reported in previous literature. As the web evolves into a universal medium and a wider array of technological features makes it possible to enhance the interactive experience, there is a need to pursue the interactivity
395
research agenda to monitor how consumers use the new media and how they progress through the adoption and learning curve associated with a new technology. Future studies should strive to make the experimental conditions as real and fully functional as possible; a solution to this is to conduct research in partnership with existing firms already engaged in web-related activities. The research technology is such that we can record, measure, and monitor changes in real-time, which reveals a great opportunity for researchers to refine their data collection strategies based on that. More research, however, is needed on the reliability and validity of this channel for research through replication and cross-validation.
Acknowledgements This project was funded in part by the RITIM Institute at the University of Rhode Island.
References Bailey M. The effects of progressive levels of interactivity in an interactive video lesson on achievement, attitude and peer interaction. Unpublished doctoral dissertation, Kansas State University; 1992. Batra R, Ahtola OT. Measuring the hedonic and utilitarian sources of consumer attitudes. Mark Lett 1990;2(2):159 – 70. Bentler P. Comparative fit indices in structural models. Psychol Bull 1991; 107:238 – 46. Cacioppo J, Petty R. The need for cognition. J Pers Soc Psychol 1982;42: 116 – 31 [January]. Cacioppo J, Petty R, Kao CF. The efficient assessment of need for cognition. J Pers Assess 1984;48(3):306 – 7. Cohen J. Statistical power analysis for the behavioral sciences. New York: Academic Press; 1977. Daft R, Lengel R. Organizational information requirements, media richness and structural design. Manage Sci 1986;32(5):554 – 71. Ducoffe R. Advertising value and advertising on the web. J Advert Res 1996;36:21 – 34 [September]. Eighmey J, McCord L. Adding value in the information age: uses and gratifications of the World Wide Web. In: Dholakia RR, Fortin DR, editors. COTIM-95 Proceedings. Kingston (RI): RITIM; 1995. p. 174 – 82. Frazer C, McMillan S. Sophistication on the World Wide Web: evaluating structure, function and commercial goals of web sites. Presented at the Advertising and Consumer Psychology Conference, Bloomfield Hills (MI); 1996 [May]. Gupta S. The Fourth WWW Consumer Survey. Ann Arbor (MI): University of Michigan; 1997. Available at: http://www.umich.edu/~sgupta/hermes/ survey4/. Haugvedt C, Petty R, Cacioppo J. Need for cognition and advertising: understanding the role of personality variables in consumer behavior. J Consum Psychol 1992;1(3):239 – 60. Heeter C. Implications of new interactive technologies for conceptualizing communication. In: Salvaggio J, Bryant J, editors. Media use in the information age. Hillsdale (NJ): Erlbaum; 1989. p. 217 – 35. Hoffman DL, Novak TP. Marketing in hypermedia computer-mediated environments: conceptual foundations. J Mark 1996;60:50 – 68. Holbrook M, Batra R. Assessing the role of emotions as mediators of consumer responses to advertising. J Consum Res 1987;14:404 – 20 [December]. Ketanurak V. An empirical investigation of the degree of interactivity in an
396
D.R. Fortin, R.R. Dholakia / Journal of Business Research 58 (2005) 387–396
interactive multimedia instructional program. Unpublished doctoral dissertation. University of Wisconsin-Milwaukee; 1996. Kroeber-Riel W. Activation research: psychobiological approaches in consumer research. J Consum Res 1979;5:240 – 50 [March]. Ku L. Impacts of interactivity from computer-mediated communication in an organizational setting: a study of electronic mail. Unpublished doctoral dissertation. Michigan State University, East Lansing (MI); 1992. Lee B, Lee R. How and why people watch TV: implications for the future of interactive television. J Advert Res 1995:9 – 18 [November/December]. MacKenzie S, Lutz R. Testing competing theories of advertising effectiveness via structural equations models. Proceedings of the 1993 Winter Educators Conference. Chicago (IL): American Marketing Association; 1983. p. 70 – 5. Morris M, Ogan C. The Internet as mass medium. J Commun 1996;46(1): 39 – 50. Neuman WR. The future of the mass audience. Cambridge (MA): Cambridge Univ. Press; 1991. Newhagen J, Rafaeli S. Why communication researchers should study the Internet: a dialogue. J Commun 1996;46:4 – 13 [Winter]. Olney T, Holbrook M, Batra R. Consumer responses to advertising: the effects of ad content, emotions and attitude toward the ad on viewing time. J Consum Res 1991;17:440 – 53 [March]. Petty RE, Cacioppo JT, Schumann D. Central and peripheral routes to advertising effectiveness: the moderating role of involvement. J Consum Res 1983;10:135 – 46 [September]. Rafaeli S. Interactivity: from new media to communication. In: Hawkins R, Pingree S, Weimann J, editors. Advancing communication science: merging mass and interpersonal processes. Newbury Park (CA): Sage; 1988. p. 110 – 34. Rafaeli S. Interacting with media: para-social interaction and real interaction. In: Ruben B, Lievrouw L, editors. Mediation, information and communication. New Brunswick (NJ): Transaction Publishers; 1990. p. 125 – 81. Raman N. Determinants of Desired Exposure to Interactive Advertising. Unpublished doctoral dissertation. University of Texas at Austin; 1996. Rice R, Associates. The new media: communication, research, and technology. Beverly Hills (CA): Sage; 1984.
Rogers E. Communication technology: the new media in society. New York: Free Press; 1986. Rust RT, Oliver RW. The death of advertising. J Advert 1994;23(4):71 – 8. Shaw T, Arnason K, Belardo S. The effects of computer mediated interactivity on idea generation: an experimental investigation. IEEE Trans Syst Man Cybern 1993;23:737 – 46 [May – June]. Short J, Williams E, Christie B. The social psychology of telecommunications. London: Wiley; 1976. Steuer J. Defining virtual reality: dimensions determining telepresence. J Commun 1992;42(4):73 – 93. Tabachnick B, Fidell L. Using multivariate statistics. New York: HarperCollins; 1989. Taylor S, Thompson S. Stalking the elusive vividness effect. Psychol Rev 1982;89(2):155 – 81. Trevino K, Lengel W, Daft R. Media symbolism, media richness and media choice in organizations: a symbolic interactionist perspective. Communic Res 1987;14(5):553 – 75. Venkatesh A, Dholakia, RR, Dholakia N. New visions of information technology and postmodernism: implications for advertising and marketing communications. RITIM Working Paper No 95-05; 1995. Walther JB. Computer-mediated communication: impersonal, interpersonal and hyperpersonal interaction. Communic Res 1996;23:3 – 43 [February]. Wells G, Petty R. The effects of overt head movement on persuasion: compatibility and incompatibility of responses. Basic Appl Soc Psychol 1981;(1):219 – 30. Wiener N. The human use of human beings: cybernetics and society. New York: Houghton Mifflin; 1950. Williams F, Rice R, Rogers E. Research methods and the new media. New York: Free Press; 1988. Zaichkowsky J. The personal involvement inventory: reduction, revision, and application to advertising. J Advert 1994;23(4):59 – 70. Zajonc R, Markus H. Affective and cognitive factors in preferences. J Consum Res 1982;9:123 – 31 [September]. Zhang Y. Responses to humorous advertising: the moderating effect of need for cognition. J Advert 1996;25:15 – 32 [Spring].
Journal of Business Research 57 (2004) 633 – 634
Editorial
Introduction to the special issue on marketing communications and consumer behavior (selected papers from the 2001 La Londe seminar) 1. Introduction The ‘‘International Research Seminar on Marketing Communications and Consumer Behavior’’ is organized on a biennial basis by the Institut d’Administration des Entreprises d’Aix-en-Provence (IAE Aix), a major French business school (University of Aix-Marseille). It is one of the two conferences emerging from the ‘‘International Research Seminar in Marketing’’ created by IAE Aix in 1974.
2. A small and ‘‘interactive’’ conference The seminar is unlike many other conferences in that it is small (70 participants this year) and is characterized by a friendly and informal atmosphere of exchange among researchers from all over the world (14 countries represented this year). In order to facilitate fruitful exchanges, each presentation lasts 45 min. Participants are scholars with an interest in developing a scientific understanding of consumer behavior in order to better design and implement advertising and other marketing communication tools.
3. An outstanding scientific committee The IAE Aix en Provence and the Conference organizers are very grateful to the respected colleagues who served as members of the Scientific Committee. These persons thoroughly reviewed and constructively commented on the 76 papers submitted. Thirty-six papers were accepted and included in the proceedings. Heartfelt thanks are given to the following scholars: Gerald Albaum (University of New Mexico), Rajeev Batra (University of Michigan at Ann Arbor), Russell W. Belk (University of Utah), Yves Evrard (HEC, France), Wayne D. Hoyer (The University of Texas at Austin), Alain Jolibert (University Pierre Mende`s-France, Grenoble), Michel Laroche (Concordia University, Montreal), Siew Meng Leong (National University of Singapore), Sidney Levy (University of Arizona), Richard J. Lutz (University of Florida), Claude R. Martin (University of Michigan at Ann Arbor), Hans 0148-2963/$ – see front matter D 2002 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00305-3
Mu¨hlbacher (Universita¨t Innsbruck), Robert A. Peterson (The University of Texas at Austin), Rik Pieters (Tilburg University), Christian Pinson (INSEAD), Bernard Pras (University of Paris-Dauphine, ESSEC), Don E. Schultz (Northwestern University), Jan-Benedict Steenkamp (Tilburg University), W. Fred van Raaij (Tilburg University), Arch G. Woodside (Boston College). We requested other respected colleagues to be ad hoc reviewers. We strongly appreciated their fruitful cooperation. These colleagues are: Mark I. Alpert (University of Texas at Austin), Eric Arnould (University of Nebraska), Alain d’Astous (HEC Montre´al), Philippe Aurier (University of Montpellier II), Michelle Bergadaa` (University of Geneva), David M. Boush (Unversity of Oregon), Joe¨l Bre´e (University of Caen), Daniel Caumont (Universite´ de Nancy), Jean-Jack Cegarra (Universite´ de Lyon 3), JeanLouis Chandon (University of Aix-Marseille), Jean-Charles Chebat (HEC Montre´al), Terry Childers (University of Minnesota), Anthony D. Cox (Indiana University), Rene´ Darmon (Groupe ESSEC), Patrick De Pelsmaker (Antwerp University), David Fortin (University of Canterbury), Pierre Gregory (University of Paris I), Chris Janiszewski (University of Florida), Jean-Noe¨l Kapferer (Groupe HEC), Richard Ladwein (Universite´ de Lille), Gilles Laurent (Groupe HEC), Gilles Marion (EM Lyon), Jean Moscarola (University of Savoie), Peter Reed (Monash University), Gilles Roehrich (University Pierre Mende`s-France, Grenoble), Greg Rose (University of Mississippi), Elyette Roux (University of Aix-Marseille), Aviv Shoham (Institute of Technology, Haifa), Eric Vernette (University of Toulouse), Monique Zollinger (University of Tours).
4. From individual differences in consumer behavior to cause-related campaigns through the role of emotions in persuasion The six competitive papers selected for inclusion in this special issue address various problems, from measuring innovativeness to testing the brand – cause fit effect in cause-related marketing communication. John W. Pracejus and G. Douglas Olsen (Canada) present the results of two experimental studies aiming at assessing
634
C. Derbaix et al. / Journal of Business Research 57 (2004) 633–634
‘‘how much’’ (and not only ‘‘if’’) brand – cause fit affects consumer choice, through choice-based conjoint analysis (five other characteristics of the brand being considered). The studies show notably a strong impact of brand – cause fit (5 and 10 times in terms of dollar value trade-off). Using LISREL 8, Elisabeth Cowley (Australia) tests a model where recognition confidence (feeling of accuracy) is influenced by recognition accuracy, knowledge (objective and subjective) and task (encoding instruction), and influences significantly memory-based choice. The results provide strong support to the hypothesized relationships. Recognition confidence is found to be an important factor of choice, while recognition accuracy is not directly predictive of choice. Generally speaking, the paper of Ross Buck, Erika Anderson, Arjun Chaudhury and Ipshita Ray (US) deals with the importance of affect (reptilian, individualistic and prosocial biological emotions) in persuasion. Specifically, the study measures emotions involved in the purchase and use or nonuse of condoms with sexual relationships of varying exclusivity. The results support the idea of an important influence of emotions. They bring to the fore the intricacy of influence of specific reptilian and prosocial emotions. Salvador Ruiz and Maria Sicilia (Spain) examine the relationships between the impact of processing styles (‘‘Need for Cognition’’ and ‘‘Preference for Affect’’) on responses to different types of ad (informational, emotional and both informational and emotional). An experimental study shows notably that ads matching the processing of subjects generate better attitude toward the brand, purchase intention and brand choice. Karen Brunsø, Joachim Scholderer and Klaus G. Grunert (Denmark) study the relationships among values, lifestyle and consumption behavior, using a means – end theory of lifestyle (‘‘intervening system of cognitive structures mediating the relations between personal values and behavior’’). Based on a domain-specific (food) lifestyle instru-
ment, the LOV inventory and a set of food-related behaviors, the model proposed is validated on a sample of 1000 French consumers. Gilles Roehrich (France) offers a state-of-the-art paper on consumer innovativeness. Briefly speaking, he shows that despite a substantial literature, the concept is still not well understood and measured.
5. Final words Finally, as co-chairs and coordinator of this conference, we strongly wish to express our gratefulness to Arch Woodside, Editor-in-Chief of the Journal of Business Research, for instigating in 1995 a special issue devoted to a selection of the La Londe seminar papers. This was a key factor of the success of this biennial meeting of worldwide scholars involved in consumer behavior and marketing communications research.
Christian Derbaix* LABACC Marketing Department Catholic University of Mons (FUCAM) 151 Chaussee de Binche, Mons 7000, Belgium E-mail address: [email protected] Lynn R. Kahle University of Oregon, Eugene, OR, USA Dwight Merunka Alain Strazzieri IAE Aix, University of Aix-Marseille, France
* Corresponding author. Tel.: +32-65-323-325; fax: +32-65-323-426.
Journal of Business Research 57 (2004) 635 – 640
The role of brand/cause fit in the effectiveness of cause-related marketing campaigns John W. Pracejus*, G. Douglas Olsen Department of Marketing, Business Economics and Law, Faculty of Business, University of Alberta, Edmonton, Alberta, Canada T6G 2R6
Abstract Previous studies of cause-related marketing (CRM) have demonstrated that it can impact consumer choice. We replicate and extend these findings using choice-based conjoint. Two studies involving 329 respondents show that fit between brand and charity can impact choice. In terms of trade-offs against price discounts, donation to a high-fit charity can result in 5 – 10 times the value of donation to a low-fit charity. We also find, however, that in both studies, the value of CRM does not justify its cost, at least in terms of short-term sales. Implications for the selection of optimal donation levels for CRM campaigns are discussed. D 2002 Elsevier Inc. All rights reserved.
1. Introduction Cause-related marketing (CRM) is an increasingly common form of promotion. Expenditures on this form of communicating with customers are expected to surpass $828 million in North America in 2002 (IEG, 2001). As defined here, CRM involves the contribution to a cause by a firm which is ‘‘linked to customers’ engaging in revenueproducing transactions with the firm’’ (Varandarajan and Menon, 1988, p. 60). We adopt this definition, and only refer to promotions in which the amount given to a charity by a firm is somehow tied to the purchase behaviour of consumers. CRM has recently received considerable attention in the literature. Drumwright (1996), for example, explored the process organizations go through in the determination of how and when to use advertising with a social dimension. Webb and Mohr (1998) found that most of their sample were aware of CRM and could provide two examples, and that nearly one third said that CRM had some impact on their purchases. More recent work has developed scales to measure individual differences in attitudes toward charitable organizations and attitudes toward helping others (Webb et al., 2000). Several studies have also addressed the impact CRM has on consumer perceptions of the sponsoring brand or firm.
* Corresponding author. Tel.: +1-780-492-2023; fax: +1-780-4923325. E-mail address: [email protected] (J.W. Pracejus). 0148-2963/$ – see front matter D 2002 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00306-5
Ross et al. (1992), for example, found that CRM had a positive impact on perceptions of the sponsoring firm and that these effects were stronger for women than for men. This positive impact of CRM was also found to be greater when the association was presented as a local, as opposed to a national, ad. Brown and Dacin (1997) also demonstrate that corporate social responsibility associations can influence product evaluations. This impact, however, was primarily through overall corporate evaluations, as opposed to attribute perceptions. Finally, three articles have directly addressed trade-offs consumers may make when weighing CRM contributions against other considerations. Strahilevitz and Meyers (1998) directly compare price reductions against ‘‘charity incentives’’. Their paradigm involves comparing the effectiveness of a unit contribution to a charity (per item purchased) with a unit reduction in the price of the item. They find that charity incentives work better for frivolous products (e.g., chocolate truffles, theme park tickets) then for practical products (pocket dictionaries, correction fluid). They replicate these experimental findings with a field test using actual coupons. They propose that this effect is due to the fact that ‘‘the altruistic utility offered by charity incentives may be more complimentary with the feelings generated from frivolous products than with the more functional motivations associated with practical products’’ (p. 444). They also consider the possibility that some of their findings may have been driven by ‘‘product –charity complementarity’’ (e.g., if the bookstore in Study 3 was giving money to a literacy fund instead of the March of Dimes, the utilitarian
636
J.W. Pracejus, G.D. Olsen / Journal of Business Research 57 (2004) 635–640
product might have benefited more from CRM; p. 444). However, they leave the exploration of whether brand/cause fit can impact choice to future research. Continuing in this stream, Strahilevitz (1999) explores the role of donation magnitude in CRM success. Specifically, she finds that when the amount of the discount/ donation becomes large (i.e., 25 – 50% of the purchase price), a majority of choices are for the discount, especially when the product is practical, as opposed to frivolous. When the amount of the discount/donation remains small, however, (i.e., 1– 5% of the purchase price), more choices are for the donation, and there is no difference between the frivolous and practical products in this regard. Finally, Barone et al. (2000) explore the trade-offs consumers make among CRM, price and quality. They do not make consumers pick choices between a brand engaging in CRM and one that is not, under varying levels of price and quality. Rather, they have subjects trade-off price or quality against varying underlying motivations (selfish/altruistic) of two brands that are both engaged in CRM. They conclude that CRM can impact choice, but only if it does not result in higher prices or lower product performance. We applaud these initial studies in that they (1) demonstrate that CRM can impact consumer choice, and (2) demonstrate that price and quality trade-offs can diminish or eliminate the positive impacts of CRM. We intend to elevate current knowledge about CRM by (1) replicating the finding that CRM can impact choice, and (2) extending these findings by demonstrating that perceived fit between sponsoring brand and cause can also impact consumer choice. Additionally, we use choice-based conjoint to calculate the actual trade-offs consumers are willing to make in order to support a brand that supports a cause. As mentioned above, Strahilevitz and Meyers (1998) have proposed that fit between brand and cause may impact the success of CRM. With the exception of Bendapudi et al. (1996), who address the issue of donor/recipient similarity in the context of charitable giving, no other conceptual or empirical work has been done in the area of brand/cause fit in CRM. Other domains, however, have found robust effects of fit on other types of brand associations. Fit between brand extensions and core brands, for example, has been found to enhance the evaluation of the extension when the core brand is well liked (Aaker and Keller, 1990, 1993). Likewise, in the context of ‘‘composite branding alliances,’’ Park et al. (1996) find that when the two brands in the composite have greater fit in terms of attribute complementary, success is facilitated. Also for celebrity endorsers, increased spokesperson/product fit results in a more favorable product attitude (Kamins and Gupta, 1994) and allows information to be processed in a shorter time (Speck et al., 1988). CRM is similar to these other domains in that brands are intentionally associating themselves with some other object in order to improve brand performance along some dimension. Across all the domains mentioned above, fit has
generally been found to facilitate transfer of positives from an object (celebrity, core brand, etc.) to the object-associated brand. That is, fit is necessary, but not sufficient for success. In keeping with this theoretical tradition, we explore the role of fit between brand and cause on choice when the cause is well liked. We are not only interested in determining if fit affects choice, but also how much it affects choice. We do this by employing discrete choice analysis (choice-based conjoint), a technique which is widely used in industry but has received surprisingly little application to the exploration of theoretical constructs in consumer research. Because many consumer researchers are not fully aware of the assumptions and capabilities of choice-based conjoint, a brief overview is provided below.
2. Choice-based conjoint Conjoint analysis has now come to refer to a broad class of decompositional models that determine utility (part – worth) estimates of individual attribute levels from global responses to a series of alternatives (e.g., a rating of each alternative) or set of alternatives (e.g., a ranking of profiles, or a choice of one from among a set). These techniques allow researchers to (1) present people with novel attribute combinations; (2) obtain demand estimates and price elasticities; and (3) calculate main effects for, and interactions among, product attributes. Discrete choice tasks are analysed using Multinomial Logistic (MNL) regression. While this method may be more difficult to use than regression programs designed for interval data (e.g., Ordinary Least Squares regression), a number of benefits exist, including: (1) the ability to estimate market share predictions directly, rather than passing them through a choice simulator (thereby eliminating one stage of the process and the introduction of more error); and (2) market share estimates are based on choice data, which is what people do in the real world (for a thorough review of the history of discrete choice, see Batsell and Louviere, 1991).
3. Experiment 1 3.1. Purpose The purpose of this study was to determine whether fit between the charity and the brand can impact the success of a CRM campaign in terms of choice behaviour, as well as to measure the magnitude of this effect, if observed. 3.2. Pretesting Since evidence from extant research in other domains suggests that fit generally facilitates or enhances the impact
J.W. Pracejus, G.D. Olsen / Journal of Business Research 57 (2004) 635–640
of associating a brand with a well-liked object, our strategy was to find two causes that were equally well liked, but differed in terms of fit with some sponsor. Given that previous research had shown that CRM is more likely to impact frivolous, as opposed to utilitarian, brands (Strahilevitz, 1999; Strahilevitz and Meyers, 1998), we decided to focus on frivolous brands. For Study 1, we chose ‘‘Theme Parks,’’ which was one of the frivolous categories shown by Strahilevitz and Meyers (1998) to be significantly impacted by CRM. Theme parks were also a highly salient category for our subjects as it was near the end of the term and many local parks were advertising summer discounts for Florida residents. We first attempted to find a set of causes that were well liked by the target population. Twenty subjects rated their liking of 124 sponsored events, organisations and causes (the pretest was also used for an unrelated study of event sponsorship). Fifteen additional subjects then rated perceived fit between theme parks and the 30 best-liked events, organisations and causes from the first part of the pretest. Fit was assessed using a seven-point scale anchored by very low fit and very high fit. The two causes chosen to be paired with theme parks were the ‘‘Kennedy Center for the Performing Arts’’ (Fit 3.2/7; Liking 6.3/7) and the ‘‘Children’s Miracle Network’’ (Fit 6.1/7; Liking 6.1/7). There is a significant difference in perceived fit of theme parks with the two causes ( F = 45.81, P < .001), but no difference in overall liking of the two causes ( F < 1, P > .38). 3.3. Respondents One hundred twenty-eight undergraduate subjects from the University of Florida took part in Experiment 1. Subjects received course extra credit in exchange for their participation. None of the 35 pretest subjects participated in the main study. 3.4. Procedure and experimental design Prior to beginning the discrete choice task, respondents were presented with sample advertising copy for the two theme parks. Embedded within the copy for one of the parks (either ‘‘Hudson’’ or ‘‘Davis’’) was the sentence ‘‘As part of our continuing support of the Children’s Miracle Network (Kennedy Center for the Performing Arts), we’ll donate five dollars to this worthy cause for every ticket sold until the end of the month.’’ Therefore, each subject was randomly assigned to one of four cells in the 2(Fit: high/low) 2(Name: Hudson/Davis) between-subjects design. Pretesting for an unrelated study found ‘‘Hudson’’ and ‘‘Davis’’ to be neutral and equivalently liked ‘‘brand names’’ in general. However, due to the possibility of category-specific associations, ‘‘name’’ was included in the design so as to provide a check that, within the context of the choice task, there was no confound between CRM and the specific name used (i.e., Hudson vs. Davis).
637
Following the description of the theme park and the charity to which the donation would be made, respondents completed a series of 16 discrete choice tasks. All choice tasks required the respondent to choose between the ‘‘Hudson’’ and ‘‘Davis’’ theme park alternatives. These alternatives were described along five different characteristics: admission price ($24.95 and $34.95); driving distance (45 vs. 90 min); number of rides available (32 or 46); food quality (fair vs. good); and hours of operation (8 am – 8 pm vs. 7 am – 11 pm). Subjects were told to consider what they had read about the two hotels in the advertisements, as well as the attribute information, which changed from task to task, in arriving at each of their 16 choices. A full factorial design was used to create all possible profiles (32) for the first option in each choice task. The choice sets were created by matching profiles from this set with a second set of all possible product profiles. This was done in a manner that permitted all attribute levels for the first alternative to remain orthogonal to all attribute levels for the second alternative. Specifically, the matching was conducted in such a way that all attribute levels for each of the attributes in the first set of profiles were paired an equal number of times with all attribute levels for each of the attributes in the second set of profiles. In an effort to reduce respondent fatigue, a blocking technique was used, randomly assigning each of the choice tasks into one of two blocks. Hence, each respondent was only required to complete 16 choice tasks (as opposed to all 32). The order of the choice tasks was randomised within each block. 3.5. Analysis and results MNL regression was employed to arrive at parameter estimates (see Table 1). The r2 value for the model as a whole was 0.70, indicating that it was very effective at explaining the choices made by respondents. The Donation Intercept parameter is based on using a dummy code to represent the amusement park that offered the charitable donation, and reflects the additional amount that an individual would be willing to pay for the low-fit donation. By dividing the coefficient for the Donation Intercept by the Admission Price coefficient, it is possible to directly determine the point at which a consumer would be ambivalent between an amusement park that offered the promotion, but charged more, and an amusement park that offered no promotion. In this particular case, results suggest that the dollar value of offering the low-fit donation is equivalent to $0.62. In order to examine the value of offering a charity that is of high-fit, a similar calculation may be performed by first summing the parameter estimates for Donation Intercept and Fit Donation Intercept, then dividing this value by the parameter estimate for the Amusement Park Admission Price. This calculation suggests that if the high-fit donation alternative was present, an amusement
638
J.W. Pracejus, G.D. Olsen / Journal of Business Research 57 (2004) 635–640
Table 1 Choice task coefficients for amusement park study Variable
Coefficient
Donation Intercept Admission Price (per $) Distance (per minute) Number of Rides Food Hours of Operation Fit Donation Intercept Fit Admission Price ($) Fit Distance Fit Number of Rides Fit Food Fit Hours of Operation Brand Name Intercept Brand Name Admission Price Brand Name Distance Brand Name Number of Rides Brand Name Food Brand Name Hours of Operation
0.333* 0.183* 0.039* 0.105* 0.479* 0.229* 0.219* 0.017 0.007** 0.021** 0.191*** 0.102* 0.136*** 0.004 0.002 0.003 0.148 0.016
* P < .001. ** P < .01. *** P < .05.
park could charge an additional $3.01. Hence, the dollar value of the high-fit alternative, relative to the low-fit alternative, is $2.39. The interaction terms between fit and park features were estimated to examine whether donating to a high-fit charity would have a differential impact on the parameter estimates for the product characteristics. It is interesting to note that fit interacted at conventionally significant levels with a number of these attributes. Specifically, relative to low-fit conditions, when the charity is high-fit, consumers are significantly less sensitive to the driving distance to the park, the number of rides available, the quality of the food and the hours of operation. Examining the parameters reported further, it is of note that an amusement park offering either donation type would be able to compensate for fewer rides or shorter operating hours relative to a competitor. Further, if the charity was high-fit, such a donation would tend to encourage the consumer to select an amusement park with fair food quality over a competitor with good food quality, all other features being equal. In addition to allowing for the calculation of dollar metric trade-offs, the choice conjoint technique also allows for the direct estimation of market shares. Predicted market shares for the high-fit and low-fit CRM campaigns can be calculated using the following equation: eUia Pia ¼ P Uja e j2Ci
where Pia = is the probability that an individual will choose the ath alternative in choice set i consisting of j alternatives; Uia = is the utility associated with the ath alternative in
choice set i; = BXia + eia, where B is a row vector of parameters and Xia is the design matrix for the ath alternative in choice set i; Ci = is the set of all alternatives. Under conditions where there are two alternatives that are equal on all features, if one alternative offered a low-fit charitable donation, market share for this alternative would go from 50% to 58.3%. However, if the charitable donation is high-fit, the market share would be elevated to 67.3%. In the event that both parks offered donations, relative to a low-fit park, a high-fit park would obtain a 59.6% share.
4. Experiment 2 4.1. Purpose Experiment 2 was designed to accomplish two goals. First, we wanted to replicate the findings of Study 1 with a different product class to increase generalizability. Second, we sought to find a product category for which the Kennedy Center for the Performing Arts would be highfit and the Children’s Miracle Network would be lower in fit. In this way, we unconfound fit with cause across the two experiments. 4.2. Pretesting In order to determine a product class meeting the above criteria, 18 respondents rated the fit between the Kennedy Center and 38 product categories. Twenty additional respondents rated the fit between the same 38 categories and the Children’s Miracle Network. ‘‘Luxury Hotels in Washington, DC’’ were found to be a significantly better fit with the Kennedy Center (4.9/7) than with the Children’s Miracle Network (2.8/7; t = 3.74; P < .001). 4.3. Respondents One hundred twenty-eight additional subjects took part in Experiment 2. They received course extra credit for participation. None of them had participated in Experiment 1 or any of the pretests. 4.4. Procedure and experimental design Just as in Experiment 1, respondents were informed that one of the two hotels would offer a donation for each night stayed ($10 in this experiment). A 2(Fit: high/low) 2(Name: Hudson/Davis) between-subjects design was used. Respondents then completed a discrete choice task. In all cases, the respondent selected from two hotels described along five objective characteristics: price per night ($115 and $95); exercise facilities (fair vs. good); number of free movie channels (one vs. three); complimentary afternoon wine and cheese (present vs. absent); and room service
J.W. Pracejus, G.D. Olsen / Journal of Business Research 57 (2004) 635–640 Table 2 Choice task coefficients for hotel study
639
5. Discussion
Variable
Coefficient
Donation Intercept Hotel Price Per Night (per $) Exercise Facilities Number of Movies Wine and Cheese Room Service Hours Fit Donation Intercept Fit Hotel Price Per Night ($) Fit Exercise Facilities Fit Number of Movies Fit Wine and Cheese Fit Room Service Hours Brand Name Intercept Brand Name Hotel Price Per Night Brand Name Exercise Facilities Brand Name Number of Movies Brand Name Wine and Cheese Brand Name Room Service Hours
0.255* 0.128* 0.531* 0.675* 1.741* 0.113* 0.241* 0.006 0.041 0.064 0.005 0.004 0.103 0.004 0.125 0.099 0.191 0.006
* P < .001.
hours (6 am to midnight vs. 24 h/day). The same design used in Study 1 was employed to create 32 choice sets. As before, each subject made 16 choices. 4.5. Analysis and results As with the first study, MNL regression was used to analyse the choice frequencies. The r2 statistic for the model was 0.73, suggesting that the parameters generated were extremely effective at capturing an understanding of the choice behaviour observed. Results of this analysis are presented in Table 2. Once again, both donation and brand/cause fit were significant. The impact of offering a charitable donation promotion, in general, and under high fit was calculated as in Study 1. In this case, the value associated with donation to the low-fit charity was $0.38. The value associated with donation to a high-fit charity was $3.88, which equates to $3.50 more than they could charge if the donation was of low-fit. Of further note is that the parameter estimates suggest that either donation type would compensate for offering reduced room service hours relative to a competitor offering room service 24 h/day. It would also be possible to compare these parameter estimates to other hotel features such as exercise facilities. Examining market share issues, if two alternatives were equal on all other parameters, offering a low-fit charitable donation would increase market share to 56.3% and assuming that the competitor offered no donation, a hotel offering a high-fit donation would result in an increase of market share to 67.6%. In the event that both hotels were offering a donation, the one with the high fit would garner 61.8%. Neither the main effect of Brand Name, nor interactions involving Brand Name were significant at conventional levels.
We have demonstrated that CRM may have a significant impact on choice behaviour (as indicated by the associated dollar equivalencies, as well as market share estimates), and that brand/cause fit substantially amplifies this effect. These effects were obtained across two product categories (Theme Parks and Luxury Hotels). While the overall positive effect of CRM had been demonstrated previously (Barone et al., 2000; Strahilevitz, 1999; Strahilevitz and Meyers, 1998), it has not been previously demonstrated in a way that allows the precise calculations of either dollar metric trade-offs or market share. More importantly, the positive impact of brand/cause fit, which we demonstrate in both studies, has not been previously found. While similar positive effects of fit have been found for brand extensions, composite branding alliances and celebrity endorsers, this effect has not been shown in the domain of CRM. Using choice conjoint techniques, we were also able to calculate the magnitude of CRM impact, both in terms of market share and tradeoffs with price. We believe that this is a major improvement over previously employed techniques. The ability of the method employed to estimate dollar trade-offs is especially salient given our results. In both studies, even the high-fit CRM campaign cost more in donations than could be justified in terms of short-term sales (see Table 3). One might interpret these findings as suggesting that the value of CRM is limited. We argue that, without considering the possible long-term impact of CRM and the limited range of donation values explored in the two studies, such a negative interpretation of the results would be premature. While the results of these studies show that even the donations to the high-fit charity have a smaller impact on choice than would an equivalent reduction in price, one must also consider the long-term benefits to brand equity, which may accrue from the CRM option. Individuals who use the brand for the first time during a CRM campaign may become loyal customers, thus creating a long-term income stream. While a price reduction may get a one-time customer, the positive associations with the charity may allow the CRM option to facilitate more long-term customer relationships. Future research addressing whether CRM is especially effective at initiating or enhancing long-term relationships, therefore, seems particularly worthwhile. There is, however, a much more important reason not to assume that these results argue against the viability of
Table 3 Amount donated and value of CRM
Study 1 Study 2
Amount donated ($)
Value of low-fit CRM ($)
Value of high-fit CRM ($)
Value of fit ($)
5.00 10.00
0.62 0.38
3.01 3.88
2.39 3.50
640
J.W. Pracejus, G.D. Olsen / Journal of Business Research 57 (2004) 635–640
CRM. This reason is that the amounts donated per transaction in the two studies (10 – 20% of the purchase price) are quite large, and possibly much larger than is typical for actual CRM campaigns. We set the level of donation at this relatively high level so as to maximise the potential for statistically observable effects of both CRM and fit. In this respect, our manipulations were successful. In doing so, however, we may have utilised donation levels far in excess of optimal ‘‘real world’’ levels. In fact, Strahilevitz (1999) found that as the donation/discount amount becomes a high percentage of product price, consumers tend to prefer the discount. The price trade-offs in Table 3 hint at the possibility that the level of donation used in the two studies is well beyond a theoretical ‘‘optimal point.’’ Note that between Studies 1 and 2, donation level goes from $5 to $10 (100% increase), while the dollar value of the high-fit CRM only goes from $3.01 to $3.88 (a 29% increase). While it would be highly speculative to draw any conclusions about the functional relationship between donation magnitude and impact on choice from these two data points, it certainly seems possible that some sort of asymptote may exist. If it does, the 10 – 20% donation levels used here may well be approaching that asymptote. Further research is clearly warranted to determine (1) if optimal donation levels exist, and (2) how to calculate these optimal values. In fact, it is possible that the ‘‘optimal value’’ for CRM donation may not be a number at all. A recent study of actual wording of CRM promotions finds that a minuscule fraction of CRM campaigns reveals an actual dollar amount given for each purchase (Olsen and Pracejus, 2002). The vast majority of campaigns utilise a vague quantifier (e.g., ‘‘A portion of the proceeds’’) to describe how much will be given. Many others state the amount in terms of a percent of profits, which consumers may confuse with percent of price. Finally, many CRM campaigns have a ‘‘cap’’ such that a percent of each sale is donated ‘‘up to’’ some maximum amount (e.g., ‘‘50 cents from each box sold will be donated, up to $10,000’’). Future research should explore the degree to which consumers are capable of understanding and processing CRM donation amounts stated in these ways, as well as the impact of such promotional verbiage on choice. Clearly, such studies would have important managerial and public policy implications. None of this, however, should take away from our primary finding that the fit between the brand and the charity can have a tremendous impact on the success of CRM campaigns. In terms of dollar value trade-offs, the high-fit CRM campaign had roughly 5 times the impact of the low-fit campaign in Study 1 ($3.01 vs. $0.62), and 10 times the impact in Study 2 ($3.88 vs. $0.38). These are not small effects. Clearly, perceived fit between the company
and the charity is an important measure that should always be taken prior to any CRM campaign.
Acknowledgements The authors acknowledge the J.D. Muir fund for financial support of this research and thank Joffre Swait for assistance with the experimental design.
References Aaker DA, Keller KL. Consumer evaluations of brand extensions. J Mark 1990;54(1):27 – 41. Aaker DA, Keller KL. Interpreting cross-cultural replications of brand extension research. Int J Res Mark 1993;10:55 – 9. Barone MJ, Miyazaki AD, Taylor KA. The influence of cause related marketing on consumer choice: does one good turn deserve another. J Acad Mark Sci 2000;28(2):248 – 62. Batsell RR, Louviere JJ. Experimental analysis of choice. Mark Lett 1991; 2:199 – 214. Bendapudi N, Sing SN, Bendapudi V. Enhancing helping behaviour: an integrative framework for promotional planning. J Mark 1996;60: 33 – 49 (July). Brown TJ, Dacin P. The company and the product: corporate associations and consumer product responses. J Mark 1997;61:68 – 84 (January). Drumwright ME. Company advertising with a social dimension: the role of noneconomic criteria. J Mark 1996;60:71 – 87 (October). IEG Sponsorship Reports 2001;2(24):1, 4 – 5. Kamins MA, Gupta K. Congruence between spokesperson and product type: a matchup hypothesis perspective. Psychol Mark 1994;11(6):569 – 86. Olsen GD, Pracejus JW. Consumer estimation of actual donation by companies to charities during cause related marketing campaigns. Working paper. Department of Marketing Business Economics and Law, University of Alberta, 2002. Park CW, Jun SY, Shocker AD. Composite branding alliances: an investigation of extension and feedback effects. J Mark Res 1996;33(4):453 – 66. Ross JK, Patterson LT, Stutts MA. Consumer perceptions of organizations that use cause-related marketing. J Acad Mark Sci 1992;20(1):93 – 8. Speck PS, Schumann DW, Thompson C. Celebrity endorsements—scripts, schema, and roles: theoretical framework and preliminary tests. In: Houston MJ, editor. Advances in consumer research, vol. 15. Provo (UT): Association for Consumer, 1988. pp. 69 – 76. Strahilevitz M. The effect of product type and donation magnitude on willingness to pay more for a charity-linked brand. J Consum Psychol 1999;8(3):215 – 41. Strahilevitz M, Meyers JG. Donations to charity as purchase incentives: how well they work may depend on what you are trying to sell. J Consum Res 1998;24:434 – 46 (March). Varandarajan PR, Menon A. Cause related marketing: a coalignment of marketing strategy and corporate philanthropy. J Mark 1998;52: 58 – 74 (July). Webb DJ, Mohr LA. A typology of consumer responses to cause related marketing: from skeptics to socially concerned. J Public Policy Mark 1998;17(2):226 – 38. Webb DJ, Green CL, Brashear TG. Development and validation of scales to measure attitudes influencing monetary donations to charitable organizations. J Acad Mark Sci 2000;20(2):299 – 309.
Journal of Business Research 57 (2004) 641 – 646
Recognition confidence, recognition accuracy and choice Elizabeth Cowley School of Marketing, University of New South Wales, John Goodsell Building, Sydney, NSW 2052, Australia
Abstract Individuals only use a subset of the information available in memory at choice. It is proposed here that in a memory-based decision setting, information that is most confidently retrieved from memory, regardless of its accuracy, will be most influential in decision making. The results reveal that recognition confidence predicts choice, although there is a relationship between recognition accuracy and choice, it is completed mediated by recognition confidence. Recognition confidence is a function of domain specific knowledge and encoding instruction. D 2002 Elsevier Inc. All rights reserved. Keywords: Memory; Product knowledge and expertise; Confidence; Recognition; Metacognition
1. Introduction While on a protracted shopping trip, Susan was standing in the fifth camera shop of the day, trying to remember if the Pentax camera she was looking at was the same model she had seen in another shop in the morning. She was also trying to remember whether the price was the same. It was late, the other shop was in another shopping mall, she felt pretty sure that the camera model was the same, and that the price was higher in the other store. She bought the camera, which was actually a different model with fewer features. Consumers generally only use a subset of the information they have accumulated and stored in memory when making decision. Identifying the most likely information to be used in a decision is of interest to both psychologists and marketers (e.g., Armacost and Hosseini, 1994; Bastardi and Shafir, 1998; Fischhoff and Downs, 1997; Kardes, 1994; Lee et al., 1999). In the accessibility– diagnosticity framework (Feldman and Lynch, 1988), more accessible information is more likely to be used in decision-making (also see Cohen, 1966). Why? Because more accessible information comes to mind easily, and is, therefore, presumed to be correct (Kelley and Lindsay, 1993; Koriat, 1993, 1994, 1995; Metcalfe, 1996). This feeling of accuracy, or recognition confidence, is an efficient way to determine what information to use in choice because checking the accuracy of recognised information often requires a fair amount of effort, if it is possible at all.
E-mail address: [email protected] (E. Cowley). 0148-2963/$ – see front matter D 2002 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00307-7
Like all metacognitive assessments, recognition confidence is prone to error (Metcalfe, 1996). In fact, in the eyewitness testimony literature, the correlation between recognition accuracy and recognition confidence has been found to be low (Bothwell et al., 1987; Cutler and Penrod, 1988). The conclusion drawn in an extensive review by Cutler and Penrod (1995) is that the confidence in one’s ability to retrieve information accurately from memory is a poor predictor of recognition accuracy. Busey et al. (2000) have recently provided an explanation of the lack of correlation between recognition confidence and accuracy with their extension of cue utilization theory. Their results support a multidimensional model of the confidence – accuracy relationship (Fig. 1). In their model, the memory representation includes two dimensions, memory strength and memory certainty. Recognition accuracy varies with memory strength, while confidence varies with both strength and certainty. Hence, any change in certainty can cause a dissociation between accuracy and confidence. If consumers use the confidence – accuracy heuristic, they may be naively using inaccurate information in consumption decisions. Not all representations of accuracy – confidence relationship predict a low correlation; for instance, in the trace access theory (Burke et al., 1991; Hart, 1967; King et al., 1980), retrieval is a process of direct access to memory during confidence and accuracy judgements (see the single dimension model in Fig. 1). Since both confidence ratings and recognition judgements are made using the same information, they can be used to predict each other. In this
642
E. Cowley / Journal of Business Research 57 (2004) 641–646
Fig. 1. Two models of the confidence and accuracy relation. Adapted from Busey et al. (2000).
case, consumers should be very confident when accurate, and not at all confident when inaccurately recognising information. The role of accessibility in choice has been investigated by manipulating vividness (Herr et al., 1991), retrieval cues (Keller, 1987; Tybout et al., 1981), interference (Keller, 1987) and frequency (Menon et al., 1995). It is important to note, however, that in these studies confidence is not measured, and incorrectly retrieved information is not included in the analyses. The relationship between recognition accuracy and recognition confidence is unclear in a consumer context. It is proposed here that encoding instructions affects memory strength which, in turn, affects both recognition accuracy and recognition confidence, while domain-specific knowledge affects memory certainty which, in turn, affects recognition confidence and the use of the information in choice.
2. The determinants of recognition accuracy and confidence 2.1. Task at study 2.1.1. Recognition accuracy One of the cornerstones of cognitive psychology is that processes that occur during encoding influence the ability to retrieve information later (Craik and Lockhart, 1972; Hunt and Einstein, 1981; Tulving and Thomson, 1973). Encoding affects retrieval because elaboration occurs during encoding, and cues in the encoding environment affect accessibility. In the material appropriate processing (MAP) framework, people learn information by considering each fact independently (individual-item processing), or by comparing new information with other information either in the learning environment or in memory (relational processing). In a marketing context, individual-item processing occurs when consumers are accumulating knowledge about brands, while relational processing occurs when consumers are evaluating or comparing brands. Although each type of processing facilitates learning and memory for information, memory performance on a recognition test, which is used in
the study presented here, improves with individual-item processing (Einstein and Hunt, 1980; Hunt and Einstein, 1981; McDaniel et al., 1988; Meyers-Levy, 1991). Individual item processing focuses on how each item is different or distinct from other items. Distinctiveness has been demonstrated to improve recognition in other paradigms as well (Guynn and Roediger, 1995; Tulving, 1969; Wallace, 1965). The first hypothesis is, therefore, a replication hypothesis. Hypothesis 1: Recognition will be more accurate when individual-item processing instructions, versus relational instructions, are provided at encoding. 2.1.2. Recognition confidence Individual-item processing encourages the encoding of information that differentiates an item from others in memory (Hunt and Einstein, 1981; Roediger and Guynn, 1996). When consumers are asked to try and recognise information learned with individual-item instructions, they should be more confident because the task at study has facilitated the type of memory necessary for the task at test. The extended cue utilisation theory would predict that the consistency between the task and the test should result in a greater level of confidence in recognition responses. Hypothesis 2: Individual-item processing instructions versus relational processing instructions will result in higher recognition confidence scores. 2.2. Knowledge level 2.2.1. Recognition accuracy Individuals with more domain knowledge can recall more newly learned information than lower knowledge individuals (Alba, 1963; Alba and Hasher, 1983; Ausubel and Fitzgerald, 1963; Chiesi et al., 1979; Voss et al., 1980), particularly when they can use their prior knowledge to organise new information (Srull, 1983). The results for recognition performance are not as clear. The dramatic differences in recall between knowledge levels are attenu-
E. Cowley / Journal of Business Research 57 (2004) 641–646
ated in recognition and, in some cases, disappear (Alba et al., 1981; Chiesi et al., 1979; Zeitz, 1994). Consumer researchers have demonstrated that low-knowledge consumers (LKCs) use explicitly stated goals or usage situations (Huffman and Houston, 1993) to identify differences between items, and are more likely to consider each fact independently (Alba and Hutchinson, 1987; Srull, 1983). Since LKCs tend to process each new statement independently, an encoding style that facilitates recognition, and high-knowledge consumers (HKCs) have a propensity to reorganise information (Srull, 1983), focusing on the gist of the information (Zeitz, 1994), any advantage resulting from more knowledge in the domain will be attenuated. Although HKCs may have a small advantage in recognition performance, knowledge level is not expected to influence recognition performance. 2.2.2. Recognition confidence HKCs are more likely to elaborate on information during encoding (Robertson, 1996; Zeitz, 1994). Elaboration can result in overconfidence because individuals tend to become more confident with the addition of their own inferential reasoning (Wagenaar, 1988). In addition, HKCs are aware that they know a lot about a product category, as is evidenced by the reported significant correlation between subjective and objective measures of expertise (Mitchell and Dacin, 1996). Hence, HKCs are more likely to be confident in their ability to retrieve information (Cowley, 1997). In which case, they may be aware that they are able to more effectively learn information in that same product category. Busey et al. (2000) suggest that a person’s belief that they are knowledgeable ‘‘do[es] not or cannot influence the recognition judgment but give[s] the illusion of accuracy and thus affect[s] confidence’’ (p. 30). Hypothesis 3: Individuals with more domain-specific knowledge will be more confident in their ability to recognise brand information than individuals low in domainspecific knowledge.
643
Fig. 2. The hypothesised relationships between retrieval confidence, retrieval accuracy and choice.
the information they most confidently retrieve from memory in their decisions. Hypothesis 4: The probability of choice will increase with recognition confidence. Although there will be a relationship between recognition accuracy and choice, the relationship will be completely mediated by recognition confidence. The hypothesised relationships are shown in Fig. 2.
4. Method Eighty-four undergraduate students participated in the study in exchange for course credit in an introductory marketing course. The sample was 56% male and 44% female. There were two encoding instruction conditions (individual-item, relational). The encoding conditions were manipulated within subject and domain-specific knowledge level was measured. Subjects were exposed to eight brand – attribute statements in each encoding condition. Each of the 16 statements included a different brand and different attribute to reduce inference generation. After a 10-min filler task, subjects participated in a recognition test, a forced choice decision task and a knowledge questionnaire. Subjects were debriefed and thanked for their participation. 4.1. Materials
3. Choice 3.1. Recognition confidence In the extended cue utilisation model, accuracy and confidence are both affected by memory strength, therefore accuracy should have some ability to predict confidence. The degree to which accuracy predicts confidence is expected to be positive and significant. Attitude confidence serves as a ‘‘psychological gatekeeper of sorts, systematically determining whether people translate their beliefs into action (e.g. Berger and Mitchell, 1989; Fazio and Zanna, 1978; Pieters and Verplanken, 1995)’’ (Swann and Gill, 1997, p. 747). If recognition confidence has the same property then consumers will use
4.1.1. Brand names Hypothetical brand names were used to avoid familiarity effects on recognition that could occur with existing brand names. The brand names are five-letter nonsense words. The names were selected from a larger pool because they were found in pretests not to have associations with any existing brand name or to differ significantly in accessibility. 4.1.2. Attribute statements Eight to 10 word statements were selected from promotional materials such that LKCs would understand how the attribute would affect camera performance. For example, ‘‘Brand X has an automatic shutterlock to avoid accidental exposures.’’
644
E. Cowley / Journal of Business Research 57 (2004) 641–646
4.2. Encoding instruction condition The MAP processing instruction provided during study varies dramatically from study to study. Individual-item tasks have ranged from pleasantness ratings (Hunt and Einstein, 1981; Hunt and Seta, 1984; Hunt et al., 1986; McDaniel et al., 1988), to imagery (Marschak and Hunt, 1989; Marschak and Surian, 1992; Meyers-Levy, 1991). Relational tasks have ranged from using advance organisers (McDaniel, 1984) to categorising the items, either mentally (Meyers-Levy, 1991), or with sorting and comparison rating tasks (McDaniel et al., 1986, 1988). The encoding instructions used here are relevant to consumption settings. The individual-item encoding instructions was ‘‘Think for a moment about the brand and rate how much you like the brand,’’ on a 110-mm continuous scale anchored with ‘I do not like the brand’ and ‘I do like the brand’. The relational encoding instruction was ‘‘Compare the brand to other brands that you have seen on the screen and rate whether it is better or worse than the other brands,’’ on a 110-mm continuous scale anchored with ‘One of the best brands’ and ‘One of the worst brands’. [To check that the manipulations induced the appropriate type of processing, 21 undergraduate students read either relational or individual-item instructions before they read four of the brand – attribute statements used in the study. They indicated agreement on a sevenpoint scale to the following three statements: (1) I found myself comparing the brands during learning, (2) I thought about each brand independently during learning, and (3) I was thinking about how difficult the statements were to understand. The results indicate that there is a significant difference between the responses to the scales in both Section 1 [ F(1,20) = 17.13, P < .0001, individual-item = 1.95, relational = 5.02], and Section 2 [ F(1,20) = 32.71, P < .0001, individual-item = 5.58, relational = 1.87].] The order of the encoding condition was counterbalanced such that each brand – attribute statement was seen with both encoding conditions. There were no order effects. 4.3. Recognition test Subjects saw 16 statements: eight from the study lists, and eight new statements. Subjects indicated whether they had seen the statement before with a yes/no response, and recognition confidence on a 50-mm continuous scale anchored with ‘‘very sure’’ and ‘‘not at all sure.’’ 4.4. Choice Subjects then made eight choices. Each choice required the subject to choose between a pair of brands. The scale was developed to force a comparison between brands, and to allow subjects a choice between: (1) indicating the magnitude of their preference for one of the brands, and (2) indicating that on the basis of the information accessible for the choice, the brands were essentially indistinguishable.
In each case, subjects indicated their brand preference on a continuous scale anchored with a brand on either end. One brand was processed with relational instructions and the other with individual-item instructions. The pairs of brands used in the choices were counterbalanced on three sets of choices. There was no effect for choice set. 4.5. Knowledge level assessment Subjects responded to both objective and subjective measures. The objective measure included multiple choice questions, terminology questions and brand name listing. The subjective measure included self-reported ratings of familiarity, product usage and product knowledge on 11point scales anchored with ‘expert’ and ‘novice’ for the product knowledge and familiarity measures, and ‘very frequently’ and ‘not at all frequently’ for the product usage measure. The coefficient alpha was .82 indicating an acceptable degree of internal consistency. All of the correlations between the individual objective and subjective measures were significant at P < .0001. The distribution of objective and subjective knowledge scores were approximately normal.
5. Results Hypothesis 1 states that participants will be more accurate in the individual-item processing condition. As expected, subjects were better able to recognise brand-attribute statements when they were encouraged to process each item individually. A simple regression run on recognition accuracy reveals a significant effect for processing condition (t = 2.90, p < .01). The hit rate in the individual-item condition (67%) was significantly higher than in the relational condition (54%). Also expected, the dramatic advantage in recall performance by HKCs reported elsewhere, is attenuated when the task is recognition. When knowledge is added to the regression processing condition remains significant (t = 2.79, p < .01), and as predicted knowledge level was not significant (t = 1.71, p=.09). Hypothesis 2 states that recognition confidence will be higher in the individual-item condition. Hypothesis 3 states that knowledge level will also influence recognition confidence with HKCs indicating more confidence than LKCs. A regression run on recognition confidence including processing condition, knowledge level and accuracy reveals a significant effect for processing condition (t = 5.02, p < .0001). Participants were significantly more confident in the individual-item condition (35.0 mm on a 50 mm scale) compared to the relational condition (30.1 mm). The relationship between knowledge and recognition confidence was positive and significant (t = 2.31, p < .05). Also interesting was that subjects were more confident when they believed that they had seen the statement before (yes = 33.9,
E. Cowley / Journal of Business Research 57 (2004) 641–646
no = 28.44, p < .0001). This finding is consistent with theories of recognition that include the use of a feeling of familiarity in the recognition judgement (Kinstch, 1967; Mandler, 1980; Tulving, 1976). When the feeling surpasses a threshold, the individual believes that they must have seen the item before. Finally, replicating previous research, accuracy was not significant (t = 0.14, p=.87). Hypothesis 4 asserts that there will be a relationship between recognition accuracy and choice, but that the relationship will be mediated by recognition confidence. Following the three steps outlined in Baron and Kenny (1986) to demonstrate mediation, a simple regression between recognition accuracy and choice reveals a significant effect for accuracy (t = 1.96, p.05), a simple regression between recognition confidence and accuracy shows a significant effect for accuracy (t = 5.43, p < .0001), and finally, a multiple regression on choice including both recognition confidence and accuracy reveals a significant effect for recognition confidence (t = 3.29, p < .001) and a non-significant effect for accuracy (t = 1.53, p=.13). Recognition confidence completely mediates the relationship between accuracy and choice.
645
cipants in the study indicate their intended choice. Second, to control for previous attitudes toward a brand, the brand names used in the study were fictitious. It could be that HKCs would be more accurate in retrieving information from memory when real brand names are involved because it would allow for full use of their more elaborate knowledge structures. Third, learning did not occur over time and contexts; there might be many other influences on recognition confidence not explicitly tested here. Limitations aside, the study contributes by considering the role of accessible information in decision-making regardless of its accuracy. Consumers can be very confident that they have seen information about a brand earlier, when the information was actually associated with a different brand. This is important because recognition confidence plays an important role in choice. The implications of the results reported here are twofold. First, recognition confidence is an important measure when testing consumer memory. Second, consumers appear to use confidence as a proxy for accuracy, therefore encouraging consumers to focus on how the brand is different or distinct from other brands should improve recognition confidence and consequently increase the probability of the use of the information in choice.
6. Discussion of the results The results of this study demonstrate that recognition confidence is an important factor in choice. The results also indicate that recognition confidence is a function of both knowledge and the task at study, while recognition accuracy is a function of the task at study only. These results do not support the trace access theory, which suggests that confidence and accuracy be assessed by directly accessing a memory trace. Instead, the results are consistent with the accessibility hypothesis and the multidimensional model. There are many occasions when consumers may be using incorrect information in the subset of information used in choice. It appears that LKCs are better able to judge their accuracy in the individual-item condition, although they are less accurate in this condition. LKCs are actually more accurate in recognition, but not as confident when encouraged to compare across brands. For these consumers, it may be advised to encourage the consideration of individual facts. Of course, consumers need to compare the brands to make a decision, but the comparisons may be more effective after the information has been learned. HKCs are more confident than LKCs, without being more accurate. Although not tested here, this confidence may lead HKCs to confidently make decisions based on inaccurately recognised information.
7. Limitations and contributions There are some obvious limitations to the study. First, the choice task in this study was not an actual choice; parti-
Acknowledgements This research was funded by an ARC Grant 02440-7220, and the School of Marketing at University of New South Wales, Sydney, Australia.
References Alba JW. The effects of product knowledge on the comprehension, retention, and evaluation of product information. In: Bagozzi RP, Tybout AM, editors. Advances in consumer research, vol. 10. Ann Arbour (MI): Association for Consumer Research, 1963. p. 577 – 80. Alba JW, Hasher L. Is memory schematic? Psychol Bull 1983;93:203 – 31. Alba JW, Hutchinson JW. Dimensions of consumer expertise. J Consum Res 1987;13:411 – 54. Alba JW, Alexander SG, Hasher L, Caniglia K. The role of context in the encoding of information. J Exp Psychol: Hum Learn Mem 1981;7: 283 – 92. Armacost RL, Hosseini JC. Identification of determinant attributes using the analytic hierarchy process. J Acad Mark Sci 1994;22(4):383 – 92. Ausubel DP, Fitzgerald D. Organizer, general background, and antecedent learning variables in sequential verbal learning. Mem Cogn 1963;12: 243 – 9. Baron RM, Kenny PA. The moderator mediator variable distinction in social psychological research: conceptual, strategic and statistical considerations. J Pers Soc Psychol 1986;51:1173 – 82. Bastardi A, Shafir E. On the pursuit and misuse of useless information. J Pers Soc Psychol 1998;75(1):19 – 32. Berger IA, Mitchell AA. The effect of advertising on attitude accessibility, attitude confidence, and the attitude – behavior relationship. J Consum Res 1989;16:269 – 79. Bothwell RK, Deffenbacher KA, Brigham JC. Predicting eyewitness accu-
646
E. Cowley / Journal of Business Research 57 (2004) 641–646
racy from confidence: the optimality perspective. J Appl Psychol 1987; 72:691 – 5. Burke DM, MacKay DG, Worthley JS, Wade E. On the tip of the tongue: what causes word finding failures in young and old adults? J Verbal Learn Verbal Behav 1991;6:325 – 37. Busey TA, Tunnicliff J, Loftus GR, Loftus EF. Accounts of the confidence – accuracy relation in recognition memory. Psychon Bull Rev 2000;7(1):26 – 48. Chiesi HL, Spilich GJ, Voss JF. Acquisition of domain-related information in relation to high and low domain knowledge. J Verbal Learn Verbal Behav 1979;18:257 – 73. Cohen L. The level of consciousness: a dynamic approach to the recall technique. J Mark Res 1966;3(2):142 – 8. Cowley E. The effect of message format and content on consumers’ confidence in their memory: Another take on comparative advertising. In: Englis BG, Olofsson A, editors. European Advances in Consumer Research, vol. 3. Ann Arbor (MI): Association for Consumer Research, 1997. p. 108 – 13. Craik FIM, Lockhart RS. Levels of processing: a framework for memory research. J Verbal Learn Verbal Behav 1972;11:671 – 84. Cutler BL, Penrod SD. Improving the reliability of eyewitness identification: lineup construction and presentation. J Appl Psychol 1988;2:281 – 90. Cutler BL, Penrod SD. Mistaken identification: the eyewitness, psychology and the law. New York: Cambridge University Press, 1995. Einstein GO, Hunt RR. Levels of Processing and Organization: Additive Effects of Individual-Item and Relational Processing. J Exp Psychol Hum Learn Mem 1980;5(5):588 – 98. Fazio RH, Zanna MP. Attitudinal qualities related to the strength of the attitude – behavior relationship. J Exp Soc Psychol 1978;14:398 – 408. Feldman JM, Lynch JG. Self-generated validity and other effects of measurement on belief, attitude, intention and behavior. J Appl Psychol 1988;73(3):421 – 35. Fischhoff B, Downs J. Accentuate the relevant. Psychol Sci 1997;8(3): 154 – 8. Guynn MJ, Roediger HL. The impact of distinctive events on implicit and explicit tests of memory. Psychol Res 1995;57:192 – 202. Hart JT. Memory and the memory-monitoring process. J Verbal Learn Verbal Behav 1967;6:685 – 91. Herr PM, Kardes FR, Kim J. Effects of word-of-mouth and product – attribute information on persuasion: an accessibility – diagnosticity perspective. J Consum Res 1991;17:454 – 62. Huffman C, Houston MJ. Goal-oriented experiences and the development of knowledge. J Consum Res 1993;20:190 – 207. Hunt RR, Einstein GO. Relational and individual item processing in memory. J Verbal Learn Verbal Behav 1981;19:497 – 514. Hunt RR, Seta CE. Category size effects in recall: the roles of relational and item specific information. J Exp Psychol: Learn Mem Cogn 1984;10(3): 454 – 64. Hunt RR, Ausley JA, Schultz EE. Shared and item-specific information in memory for event descriptions. Mem Cogn 1986;14(1):49 – 54. Kardes FR. Consumer judgment and decision processes. In: Wyer RS, Srull TK, editors. Handbook of social cognition, vol. 1: basic processes; vol. 2: applications (2nd ed). Hillsdale (NJ): Lawrence Erlbaum Associates, 1994. p. 399 – 466. Keller KL. Memory factors in advertising: the effect of advertising retrieval cues on brand evaluations. J Consum Res 1987;14(3):316 – 33. Kelley CM, Lindsay D. Stephen: remembering mistaken for knowing: ease of retrieval as a basis for confidence in answers to general knowledge questions. J Mem Lang 1993;32(1):1 – 24. King J, Zechmeister EB, Shaunessy JJ. Judgments of knowing: the influence of retrieval practice. Am J Psychol 1980;95:329 – 43. Kinstch W. Memory and decision aspects of recognition learning. Psychol Rev 1967;74:496 – 504.
Koriat A. How do we know what we know? The accessibility model of the feeling of knowing. Psychol Rev 1993;100:609 – 39. Koriat A. Memory’s knowledge of its own knowledge: the accessibility account of the feeling of knowing. In: Metcalfe J, Shimura AP, editors. Metacognition: knowing about knowing. Cambridge: MIT Press, 1994. p. 115 – 35. Koriat A. Dissociating knowing and the feeling of knowing: further evidence for the accessibility model. J Exp Psychol: Gen 1995;124:311 – 33. Lee H, Herr PM, Kardes FR, Kim C. Motivated search: effects of choice accountability, issue involvement, and prior knowledge on information acquisition and use. J Bus Res 1999;45(1):75 – 88. Mandler G. Recognising: the judgement of previous occurrence. Psychol Rev 1980;87:252 – 71. Marschak M, Hunt RR. A reexamination of the role of imagery in learning and memory. J Exp Psychol: Learn Mem Cogn 1989;15(4):710 – 20. Marschak M, Surian L. Concreteness effects in free recall: the roles of imaginal and relational processing. Mem Cogn 1992;20(6):612 – 20. McDaniel MA. The role of elaborative and schema processes in story memory. Mem Cogn 1984;12(1):46 – 51. McDaniel MA, Einstein GO, Dunay PK, Cobb RE. Encoding difficulty and memory: toward a unifying theory. J Mem Lang 1986;25:645 – 56. McDaniel MA, Einstein GO, Lollis T. Qualitative and quantitative considerations in encoding difficulty effects. Mem Cogn 1988;16(1):8 – 14. Menon G, Raghubir P, Schwarz N. Behavioral frequency judgments: an accessibility – diagnosticity framework. J Consum Res 1995;22(2): 212 – 28. Metcalfe J. Metacognitive processes. In: Bjork EL, Bjork RA, editors. Memory. London: Academic Press, 1996. p. 382 – 411. Meyers-Levy J. Elaborating on elaboration: the distinction between itemspecific and relational elaboration. J Consum Res 1991;18:358 – 67. Mitchell AA, Dacin PF. The assessment of alternative measures of consumer expertise. J Consum Res 1996;23:219 – 39. Pieters RGM, Verplanken B. Intention – behaviour consistency: effects of consideration set size, involvement and need for cognition. Eur J Soc Psychol 1995;25(5):531 – 44. Robertson LJ. Clinical reasoning: Part 2. Novice/expert differences. Br J Occup Ther 1996;59(5):212 – 6. Roediger HL, Guynn MJ. Retrieval processes. In: Bjork EL, Bjork RA, editors. Memory. San Diego (CA): Academic Press, 1996. p. 197 – 236. Srull TK. The role of prior knowledge in the acquisition, retention, and use of new information. In: Bagozzi R, Tybout AM, editors. Advances in consumer research, vol. 10. Ann Arbor (MI): Association for Consumer Research, 1983. p. 572 – 6. Swann WB, Gill MJ. Confidence and accuracy in person perception: do we know what we think we know about our relationship partners? J Pers Soc Psychol 1997;73(4):747 – 57. Tulving E. Retrograde amnesia in free recall. Science 1969;164:88 – 90. Tulving E. Ecphoric processes in recall and recognition. In: Brown J, editor. Recall and recognition. London: Wiley, 1976. p. 19 – 25. Tulving E, Thomson DM. Encoding specificity and retrieval processes in episodic memory. Psychol Rev 1973;80(5):352 – 73. Tybout AM, Calder BJ, Sternthal B. Using information processing theory to design market strategies. J Mark Res 1981;18:73 – 9. Voss JF, Vesonder GT, Spilich GJ. Text generation and recall by highknowledge and low-knowledge individuals. J Verbal Learn Verbal Behav 1980;19:651 – 67. Wagenaar WA. Calibration and the effects of knowledge and reconstruction on retrieval from memory. Cognition 1988;28:277 – 96. Wallace WP. Review of the historical, empirical and theoretical status of the von Restorff phenomenon. Psychol Bull 1965;63:410 – 24. Zeitz CM. Expert – novice differences in memory, abstraction, and reasoning in the domain of literature. Cogn Instr 1994;12(4):277 – 312.
Journal of Business Research 57 (2004) 647 – 656
Emotion and reason in persuasion $ Applying the ARI model and the CASC Scale Ross Bucka,*, Erika Andersonb, Arjun Chaudhuric, Ipshita Rayc a
Communication Sciences U-1085, University of Connecticut, Storrs, CT 06269-1085, USA b University of Southern Maine, Portland, ME, USA c Fairfield University, Fairfield, CT, USA
Abstract Whereas practitioners in advertising and marketing clearly appreciate the importance of affect and emotion, traditional academic approaches to the analysis of persuasion tend to stress rational ‘‘central route’’ or ‘‘systematic’’ processing. However, the notion of two sorts of cognitive process—one rational, the other affective—has gained increasing support. This paper presents a view of the conceptualization and operationalization of the interaction of affect and reason based upon MacLeans’s triune theory of the brain, distinguishing reptilian, individualist and prosocial biological emotions, as well as ‘‘higher level’’ social, cognitive and moral emotions. The interactive role of affect and reason in involvement is described by the affect – reason – involvement (ARI) model, and emotions are operationalized by versions of the Communication via Analytic and Syncretic Cognition Scale (CASC Scale) tuned to the requirements of a given area of investigation. Examples of studies analyzing emotional factors in response to common consumer products and condom use/nonuse are presented. D 2002 Elsevier Inc. All rights reserved. Keywords: Emotion; Reason; Advertising; Persuasion
1. Introduction Traditional academic models of attitude change and persuasion typically approach their subject matter from ‘‘cold,’’ rational, analytic – cognitive points of view. Even those specifically attempting to deal with the role of emotion in persuasion tend to view the ‘‘central route’’ involving analytic cognition as the subject of ultimate interest and emotion as a ‘‘peripheral’’ factor at best. However, inspection of a few television commercials is enough to convince an objective observer that advertising and marketing professionals are well aware of the importance of emotion in the practical realm of persuasion. Arguably, the global village associated with new electronic media has greatly increased the importance of emotion in persuasion. At the same time, in the academic realm, advances in the neurosciences and in the study of nonverbal – emotional commun-
$ Presentation at the Fourth International Research Seminar on Marketing Communications and Consumer Behavior La Londe les Maures (French Riviera), June 5 – 8, 2001. * Corresponding author. Tel.: +1-860-486-4494; fax: +1-860-486-5422. E-mail address: [email protected] (R. Buck).
0148-2963/$ – see front matter D 2002 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00308-9
ication have made possible a new level of specificity in the definition of emotion. This paper seeks to apply the newfound academic understanding of emotion and nonverbal communication to the practical realm of marketing communication and persuasion. It addresses the persuasion process in general, and marketing communication in particular, from the viewpoint of the affect – reason – involvement model (ARI model), which defines and describes the relationships between affect, reason and involvement, arguing that both emotional and rational involvement are important in attitude change and persuasion. ‘‘Affect’’ (also termed Emotion III) is defined as the subjective aspect of ‘‘emotion’’ in Developmental Interactionist Theory (Buck, 1985, 1999). That is, emotion is seen to have three aspects, arousal (Emotion I), expression (Emotion II) and subjective experience (Emotion III or affect). Emotional involvement is based upon holistic syncretic cognition, while rational involvement is based upon linear and sequential analytic cognition. Affect is operationalized using the Communication via Analytic and Syncretic Cognition Scale (CASC), developed to capture both analytic rational knowledge and a large number of varieties of syncretic – affective knowledge. That is, the CASC Scale
648
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
is designed with the potential to assess a wide variety (potentially hundreds) of emotions, including biological emotions and many ‘‘higher-level’’ social, cognitive and moral emotions distinguished in recent theoretical analyses (see Buck, 1999). Biological emotions include reptilian emotions (eroticism and power), fearful individualistic emotions (fear, nervous, uncomfortable and unsure), angry Individualistic emotions (angry, insulted and selfish), positive individualistic emotions (confident, secure and satisfied), negative prosocial emotions (guilty, ashamed and embarrassed) and positive prosocial emotions (intimate, loving/loved and caring). Social emotions (e.g., pride, guilt, shame, pity, jealousy) are conceptualized as being based biologically upon prosocial attachment needs and cognitive emotions as being based biologically upon curiosity and interest. Moral emotions are seen as being based upon both cognitive and social emotions, involving the understanding of rules and the judgment that they should be followed. In the research described in this paper, the ARI conceptualization and CASC Scale were applied to studying the beliefs and emotions associated with common consumer products, and with safe sex and condom use. In this research, the CASC Scale was developed into the Brief-CASC, an efficient form applicable to consumer research, and the Safe Sex Communication Scale (SAFECOM) centered on emotions involved in sexual behavior. More recently, a version of CASC termed the Competence – Credibility –Compassion –Charisma Scale (C4 Scale) has been developed to focus upon emotional involvement in the perception of leadership, in a study of the 2000 Presidential election campaign in the US (Buck and Vieira, 2001). This introductory section considers (a) the treatment of emotion and reason in persuasion theory and research, (b) the conceptualization and operationalization of involvement and (c) specific definitions of affect, reason and involvement. It concludes with a description of an application of the ARI model using the Brief-CASC Scale to analyze emotion and reason associated with familiar consumer products.
2. Emotional and rational cognition 2.1. Hot, cold and indifferent: emotion and persuasion research Emotion is a ‘‘hot topic’’ in the social and behavioral sciences, due in part to new capabilities to observe and measure emotional phenomena. These developments have greatly increased the ability to measure emotion in its various manifestations. In particular, the study of ‘‘subjective’’ affective phenomena has long been eschewed by many behavioral scientists because of the apparent impossibility of objective measurement. The new developments in measuring activities of central neurochemical systems put the study of such events on a newly objective basis (Buck, 1993, 2000). Indeed, affect can be defined explicitly in
terms of potentially observable neurochemical systems, leading to the possibility of an affective neuroscience (Buck, 1985, 1999; Panksepp, 1993, 1998). 2.1.1. Two-factor theories and persuasion ‘‘Two factor’’ models of persuasion have recognized that emotion plays a role of in attitude change. The present model suggests instead that there are in effect two persuasion processes that take place simultaneously and interactively: a rational influence process involving analytic cognition that works much as conventional theories in attitude change envision, and an emotional influence process involving a qualitatively different sort of cognition. Emotional cognition involves memory and processing systems that are separate from those of analytic cognitive processing, are organized differently and obey different rules. Moreover, emotional and rational persuasion are associated with two qualitatively different but simultaneous and interacting ‘‘streams’’ of communication: spontaneous and symbolic communication, respectively (Buck, 1984). 2.1.2. Affect as syncretic cognition Based upon neuropsychological theory and research, Tucker (1981) distinguished two sorts of cognition. Syncretic cognition is ‘‘hot,’’ direct and immediate, while analytic cognition involves ‘‘cold,’’ sequential and linear information processing. He related the distinction between analytic and syncretic cognition to processing modes characteristic of the left and right cerebral hemispheres, respectively. Tucker’s distinction is similar in many respects to that made by Le Doux (1994) between cortico-cognitive processes based on the hippocampus and neocortex vs. emotional processing involving the amygdala. Le Doux (1996) distinguished a ‘‘high road’’ and ‘‘low road’’ to cognition, showing that emotion-related structures associated with the amygdala receive input about events that is earlier than and potentially independent of input to relevant neocortical sensory systems. Furthermore, Le Doux outlined two central memory networks that operate simultaneously and in parallel: explicit or declarative memory, which involves the hippocampus, and implicit or emotional memory which involves the amygdala (1994, p. 312). The differentiation of analytic and syncretic cognition blurs the usual distinction between emotion and cognition: the subjective experience of emotion, or ‘‘affect,’’ becomes a type of cognition: a type of knowledge. In the present view, affect is defined as the direct knowledge of feelings and desires, based upon readouts of specifiable neurochemical systems evolved by natural selection as phylogenetic adaptations functioning to inform the organism of bodily events important in self-regulation (Buck, 1985, 1994, 1999). Human beings experience affects immediately and directly, the phenomenological subjective reality of affect is self-evident. Among the strongest and most fundamental of affects is, of course, sex.
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
2.1.3. Prosocial affects Theories of emotion emphasize ‘‘individualistic’’ emotions associated with individual survival, and often ignore sex and other ‘‘prosocial’’ emotions associated with species survival posited in MacLean’s (1993) Triune Theory of the brain. Recent studies have supported MacLean’s approach, in that they have focused upon neurobiological systems that regulate social organization involving attachment and bonding (see Buck, 1999; Carter et al., 1997; Panksepp, 1993). The subjectively experienced affects associated with these systems appear to involve specific neurochemicals including serotonin, oxytocin, gonadotropin releasing hormone and the endorphins. These participate in the regulation of a vast array of positive social behaviors involving feelings of erotic arousal, intimacy, caring, playfulness and love. 2.2. Conceptualizing affect, reason and involvement: the ARI model The notion that persuasion involves parallel and interactive processes of affective and rational influence is at the heart of the ARI model of persuasion. The ARI model describes the relationship of affect and reason with one another and with involvement (Buck and Chaudhuri, 1994; Chaudhuri and Buck 1995a,b, 1998). 2.2.1. The A/R continuum The relationship between affect and reason is expressed by Fig. 1. Affective persuasion is associated with spontaneous communication and rational persuasion with symbolic communication (Buck, 1984). The continuum at the base of Fig. 1 is the affect/reason continuum (A/R continuum). It describes the mix of affective and rational persuasion, and therefore also the mix of spontaneous and symbolic communication. At the left of the continuum, the influence of affect is total: reason has no influence. This is the case of pure spontaneous communication. As one goes to the right, reason exerts an increasing influence relative to affect, but affective influence never falls to zero. Attitude objects and messages can be placed on the A/R continuum, reflecting the ratio of affect to reason in the possession and use of that object, or in the message advocating such possession and use (Chaudhuri and Buck, 1993).
Fig. 1. The A/R continuum.
649
Fig. 2. The ARI (affect – reason – involvement) solid.
2.2.2. Level of involvement Chaudhuri and Buck (1993) define involvement conceptually following Batra and Ray (1983) as the ‘‘depth and quality of cognitive response’’ (p. 309), but suggest that both affective and rational involvement are possible, constituting the depth and quality of syncretic and analytic cognitive responding, respectively. This can refer to a temporary state, to a general disposition to respond, and/ or to the tendency of an attitude object or message to ‘‘pull out’’ affective or rational responses in people. Given this definition, we suggest that the level of involvement (LI) can be defined operationally as the average of affective and rational involvement: that is, LI=(A + R)/2. In this way, involvement is defined both conceptually and operationally as a combination of affective and rational cognitive processing: if cognitive processing is measured, involvement is measured by definition. This suggests that one can be ‘‘hot, cold or indifferent’’ in response to attitude objects and messages. ‘‘Hot processing’’ is high in affect (high A/R ratio) and high in involvement, ‘‘cold processing’’ is high in reason (low A/R ratio) and high in involvement, ‘‘indifference’’ is low in affect and reason, and low in involvement. 2.2.3. The ARI solid The ARI solid models the relationships between affect, reason and involvement. The ARI solid is a three-dimensional figure bounded on one side by a low-high LI dimension and on the other by the A/R continuum (see Fig. 2). An ARI slice, in which the relative influence of affect and reason remains constant as involvement varies, represents the relative influence of affect and reason at any point on the A/R continuum. The ‘‘floor’’ of the ARI solid is a two-dimensional space with an involvement dimension on the y-axis and the A/R continuum on the x-axis (Fig. 3A). LI and the A/R ratio represent the position of an object on this floor. The floor of the ARI solid is similar in form to the Foot, Cone and Belding grid (FCB grid; Fig. 3B), which contrasts high and low involvement with ‘‘feel’’ and ‘‘think’’ categories (Vaughn, 1980, 1986).
650
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
2.2.5. Evaluation and the ARI model The ARI model aims to conceptualize and measure how attitude objects and messages are cognitively processed, and the level of involvement as defined by the depth and quality of this processing. It provides a representation of the depth of thought and feeling about an object or message. Again, this can be conceptualized in several ways: as a temporary state affected by the situation, an enduring tendency of a person to respond carried across situations and/or the tendency of an object or message to ‘‘pull out’’ a mix of rational and emotional responding in people. The Brief-CASC Scale was designed with the latter in mind: assessing attitude objects. However, the ARI model does not in itself represent evaluation: whether the object or message is liked or disliked, approached or avoided, loved or loathed. The position of an object or message at any point in time may thus be described in three dimensions: LI, A/R ratio and evaluation. This paper concentrates on the former two dimensions. 2.3. Measuring affect and reason: the CASC Scale
Fig. 3. The floor of the ARI solid (A) and the FCB grid (B).
2.2.4. Affective and rational responses to consumer objects: the Brief-CASC Scale Chaudhuri (1993) tested the usefulness of this conceptualization to the analysis of the processing of consumer products. He developed three-item scales from the McQuarrie and Munson (1987) modification of the Zaichkowsky Personal Involvement Inventory: three ‘‘risk’’ items comprised the measure of analytic cognition and three ‘‘hedonic’’ items comprised the measure of syncretic cognition. The analytic items were: ‘‘Not risky vs. risky,’’ ‘‘Easy to go wrong vs. hard to go wrong’’ and ‘‘Hard to pick vs. easy to choose.’’ The syncretic items were: ‘‘Appealing vs. unappealing,’’ ‘‘Unexciting vs. exciting’’ and ‘‘Fun vs. not fun.’’ This ‘‘Brief-CASC Scale’’ was used to assess affective and rational responses to 30 consumer objects ranging from cars and computers, to candy and paper products, to insurance policies (Buck et al., 1995). As expected, affectively loaded objects had high A/R ratios, while objects to be dealt with ‘‘mindfully’’ had low A/R ratios. For example, candy and snack foods had high A/ R ratios (1.36 and 1.43, respectively), automobiles, airline services and paper products had and relatively balanced in A/R scores (1.18, 1.00 and 0.90), and appliances, insurance policies and laundry products had low A/R ratios (0.59, 0.44 and 0.77). At the same time, candy and insurance had relatively high LI scores (3.99 and 4.00), suggesting hot and cold processing, respectively, while paper products had a low LI score (2.60), suggesting indifference.
Affect is usually thought of as undifferentiated relative to reason, but the measurement of affect can be highly specific, involving for example, happiness, anger, eroticism and shame (Buck, 1999). MacLean’s (1993) Triune Theory of the brain suggests that there are reptilian emotions involving ‘‘raw’’ sex and aggression (eroticism, power) based upon subcortical parts of the brain. Paleocortical (limbic system) areas are associated with more complex motivational – emotional systems: individualistic emotions involving selfpreservation (anger, fear) and prosocial emotions involving species preservation (love, caring, intimacy). The prosocial emotions serve as the biological basis of a range of higherorder social emotions, including embarrassment, guilt and shame. The individualistic emotions serve as the biological basis for higher-order cognitive emotions involved in the structuring of the cognitive system (curiosity, surprise, interest, boredom), including emotions such as confidence, security and satisfaction (see Buck, 1988, 1999). These specific emotions can be assessed by the CASC Scale, which can be modified to fit those emotions relevant to a specific area of study.
3. The SAFECOM Scale: application to the study of condom use/nonuse 3.1. Persuasion and HIV/AIDS prevention Conventional persuasion campaigns have succeeded in convincing many if not most in the target audience about the importance of condom use in preventing AIDS and other STDs. However, these messages have often been unsuccessful in altering behavior (Baldwin et al., 1990; DeBro et al., 1994), and there is little evidence that knowledge about HIV/AIDS and its prevention increases condom use
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
(DiClemente, 1991; Valdiserri et al., 1989; Rikert et al., 1989). In a meta-analysis of 26 cross sectional studies, Gerrard et al. (1996) concluded that there is ‘‘virtually no support for the hypothesis that perceptions of vulnerability to HIV/AIDS motivate subsequent precautionary sexual behavior’’ (p. 398). We suggest that the consideration of emotional factors may be useful in the promotion of safe sex behavior. Conventional methods of public service communicationbased as they are upon assumptions of rational judgment— may not be adequate in meeting the hazard of the pandemic of HIV/AIDS and other STDs. Sex is hot: it is rarely cold and never indifferent. To explore this, the CASC Scales were applied to the analysis of the emotions and perceptions involved in the purchase and use or nonuse of condoms in specific personal relationships varying in exclusivity. 3.2. Emotions, condom use/nonuse and relationship exclusivity As noted, the CASC Scale can be modified to fit a specific area of study, and the SAFECOM Scale was developed from the CASC Scale to assess emotions particularly relevant to safe sex. Based upon MacLean’s (1993) theory, we used 18 emotion items expecting six emotion clusters to emerge (Buck, 1999). Reptilian emotions included erotic and power; fearful individualistic emotions included fear, nervous, uncomfortable and unsure; angry individualistic emotions included angry, insulted and selfish; positive individualistic emotions included confident, secure and satisfied; negative prosocial emotions included guilty, ashamed and embarrassed, and positive prosocial emotions included intimate, loving/loved and caring. Each of the 18 emotions was measured by a seven-point scale anchored on the left by Not at all and on the right by Very. Also, the questionnaire explicitly included relational exclusivity in questioning Ss about condom use or nonuse. 3.3. Hypotheses 3.3.1. Relational exclusivity We expected that the exclusivity of the relationship would have a major impact upon feelings about condom use, with fears of HIV/AIDS and other STDs being mitigated to the extent that the relationship is perceived as being exclusive. We operationally defined levels of increasing exclusivity by the three sorts of relational context: from a ‘‘one-night-stand’’ to a ‘‘person you know’’ and are considering dating exclusively, to a partner in a long-term ‘‘monogamous’’ relationship. 3.3.2. Reptilian rewards The SAFECOM Scale explicitly asked participants about the reptilian emotions: feelings of eroticism and power. We expected that these feelings play an important role in the motivation for condom use/nonuse, and specifically that
651
condom nonuse would be associated with higher feelings of both eroticism and power. 3.3.3. The ASP hypothesis Based upon research and theory regarding gender differences in evolutionary sexual strategies as well as cultural gender roles, we expected reported feelings of anger, selfishness and power to differ for women and men. Specifically, we expected that condom use would elicit feelings of higher anger and lower selfishness and power among men than women, and that nonuse of condoms would elicit the opposite patterns. This was the anger – selfishness –power (ASP) hypothesis. 3.4. Participants, materials and design Participants were 188 undergraduates from three universities in the northeastern United States, including 95 males, 86 females and 7 unknown/missing data. The anonymous questionnaire was administered in classroom settings. The questionnaire initially asked about knowledge about HIV/ AIDS and condoms, and related beliefs, attitudes, perceived risks, behavior intentions and behaviors. The SAFECOM Scale then asked how ‘‘a person would feel’’ (1) discussing condoms with a new sexual partner, (2) purchasing a condom and (3) using or not using a condom within sexual relationships of three levels of exclusivity: a ‘‘one-nightstand,’’ an ‘‘acquaintance’’ and a ‘‘long-term partner.’’ A total of 18 emotions were assessed for each of these situations, with one situation per page for a total of eight pages. These latter, SAFECOM data are the subject of the present analysis. 3.5. Results: feelings about condom use or nonuse across relationships 3.5.1. Structure of emotions Exploratory factor analysis was used in a confirmatory manner to analyze emotion structure. Results indicated that ratings of how ‘‘a person would feel’’ discussing condom use with a new sexual partner produced a factor structure consistent with MacLean’s (1993) theory. Results from a replication study conducted at the University of Hyderabad to the same question produced a factor structure closely comparable to that produced with American participants (Kowta, 1996). The combined factor structure is presented in Table 1. For each of the emotions, data on feelings about condom use/nonuse were examined by 2(2 3) mixed model analyses of variance, with gender of participant crossed with repeated measures on condom use/nonuse and relationship. Results indicated many significant effects with effect sizes ranging from small to quite large. The meanings of significant main effects were often tempered by strong interactions. In the following, main effects are discussed first, followed by the interaction effects, organized by emotion type. Effect sizes
652
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
Table 1 Affect structure associated with how a person would feel discussing condom use with a new sexual partner (SAFECOM version of CASC) Affect
Positive Negative Positive Negative prosocial prosocial individualist individualist Reptilian
Loving/loved .89 Caring .74 Intimate .71 Afraid Nervous Ashamed Embarrassed Guilty Uncomfortable Satisfied Confident Secure Selfish Insulted Angry Unsure Powerful Erotic
.86 .83 .72 .67 .65 .57
.51 .75 .74 .65 .78 .68 .66 .64 .78 .71
Only factor loadings above .50 are illustrated.
(h’s) are given to illustrate the strength of the different effects and to convey something of their relative importance. 3.5.2. Relationships and emotions Main effects indicated that, overall, the increasing exclusivity of relationships was strongly associated with increased positive and decreased negative emotion ratings. Effect sizes were generally high (median h=.68). 3.5.3. Condom use and emotions Main effects indicated that, overall, condom nonuse compared to use was associated with higher ratings of all of the angry and fearful individualistic emotions and the negative prosocial emotions, with high effect sizes (median h=.72). Regarding positive individualistic emotions, condom nonuse was associated with lower ratings of secure and confident (h’s=.73 and .56, respectively). For positive prosocial emotions, condom nonuse was associated with lower ratings of caring (.26), but higher intimacy (.39). As expected from the ‘‘reptilian rewards’’ hypothesis, condom nonuse was associated with higher ratings on reptilian emotions, weakly with power (h=.15, P=.042) and strongly with erotic (h=.54, P < .001). 3.5.4. Negative prosocial emotions Negative prosocial emotions (feeling ashamed, embarrassed and guilty) showed the following pattern, illustrated for ‘‘ashamed’’ in Fig. 4A. When condoms were used, these emotions were low levels across relationships, but when condoms were not used these emotions were high levels in ‘‘one night stands’’ decreasing, as the relationship became more exclusive. The linear components of interactions between condom use and relationship were highly significant in all cases (h=.69 for shame, h=.52 for embarrassment,
h=.67 for guilt; all P’s < .001). Women reported generally higher levels of the negative prosocial emotions. These emotions also showed significant interactions between gender and condom use. Women reported higher levels of shame, embarrassment and guilt relative to men when condoms were not used (h=.23 for shame, h=.17 for embarrassment, h=.17 for guilt; all P’s < .03). Moreover, for less exclusive relationships, women reported relatively higher levels of shame and embarrassment than did men (h=.21 for shame, h=.20 for embarrassment; P’s < .02). 3.5.5. Fearful individualistic emotions Fearful individualistic emotions (feeling afraid, nervous and uncomfortable) showed a pattern similar to those of the negative prosocial emotions. Significant main effects indicated that, overall, these emotions were stronger when condoms were not used. When condoms were used, these emotions were at relatively low levels across relationships. However, when condoms were not used, these emotions showed high levels in ‘‘one night stands,’’ which decreased dramatically as the relationship became more exclusive. The linear components of interactions between condom use and relationship were highly significant in all cases (h’s for fear=.61, for nervous=.58, for uncomfortable=.46; all P’s < .001). Women reported higher overall feelings of nervousness and discomfort, but not of fear. Unlike nervousness and discomfort, fear showed a significant interaction between gender and condom use. Women reported higher levels of fear than men did when condoms were not used, but less when they were used (h=.21, P=.007). 3.5.6. Angry individualistic emotions The angry individualistic emotions (feeling angry, unsure, insulted and selfish) showed a pattern generally similar to that of the negative prosocial emotions, illustrated for anger in Fig. 4B. Again, as noted significant main effects indicated that all of these emotions were stronger when condoms were not used, but there were highly significant interactions with relationship exclusivity. When condoms were used, these emotions were at relatively low levels, but when condoms were not used, they showed high levels in ‘‘one night stands’’ that decreased, as the relationship became more exclusive. Again, the linear components of interactions between condom use and relationship were highly significant (h=.61 for anger, h=.50 for unsure, h=.48 for insulted, h=.35 for selfish; all P’s < .001). Again as noted, main effects indicated women to report feeling more insulted and unsure than men. These emotions and anger also showed significant interactions between gender and condom use. When condoms were not used, women relative to men reported higher levels of anger, being unsure and insulted (h=.26 for anger, h=.16 for unsure, h=.18 for insult; all P’s < .04). Also, for the less exclusive relationships, women reported relatively higher levels of anger and being unsure (h=.16 for anger, h=.16 for unsure; P’s < .04).
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
653
Fig. 4. Emotional responses to using or not using condoms in one-night-stand, acquaintance and long-term relationships.
3.5.7. Positive individualistic emotions Reports of feeling secure, confident and satisfied generally showed a ‘‘mirror image’’ of the pattern shown with the fearful, angry and negative prosocial emotions, illustrated for secure in Fig. 4C. As noted, significant main effects indicated that security and confidence were stronger when condoms were used, but the main effect for satisfaction was not significant, and indeed the tendency was in the opposite direction. Inspection of the data revealed that this was primarily due to males’ greater reported satisfaction associated with condom nonuse. Highly significant interactions indicated that, when condoms were used, these emotions were at relatively high levels, but when condoms
were not used, they showed low levels in ‘‘one night stands’’ that increased, as the relationship became more exclusive. The linear components of interactions between condom use and relationship were again highly significant in all cases (h=.54 for secure, h=.49 for confident, h=.44 for satisfied; all P’s < .001). Similarly, the positive individualistic emotions showed significant interactions with gender. When condoms were not used, women reported relatively lower levels of confidence and satisfaction relative to men (h=.22 for confidence, h=.19 for satisfaction; P’s < .02). Also, for the less exclusive relationships, women reported relatively lower levels of security and satisfaction (h=.18 for security, h=.24 for satisfaction; P’s < .02).
654
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
3.5.8. Positive prosocial emotions The positive prosocial emotions (feeling loving/loved, caring and intimate) increased as relationships became more exclusive, illustrated for intimacy in Fig. 4D. As noted, main effects indicated that rated caring was stronger when condoms were used, but rated intimacy was stronger when condoms were not used, and the main effect for feeling loving/loved was not significant. In the less exclusive relationships, these emotions were at roughly equivalent levels whether condoms were used or not used, but in the long-term relationship these emotions were consistently higher, for both females and males, when condoms were not used. Again, the linear components of interactions between condom use and relationship were highly significant in all cases (h=.33 for loving/loved, h=.40 for caring, h=.44 for intimate; all P’s < .001). Positive prosocial emotions showed no significant interactions between gender and condom use: both for females and males, loving, caring and intimacy tended to be higher in the long-term relationship when condoms were not used. For the less exclusive relationships, women reported relatively lower levels of loving/loved and caring relative to men (h=.18 for loving/ loved, h=.24 for caring; P’s < .02). 3.5.9. Reptilian emotions: eroticism Results for erotic emotions are illustrated in Fig. 4E. As noted, these increased as relationships became more exclusive, erotic feelings were stronger when condoms were not used, and males reported generally higher erotic feelings. Both the linear and quadratic components of the interactions between condom use and relationship were significant in eroticism (h=.26, P=.001 for linear; h=.42, P < .001 for quadratic), indicating that erotic feelings were relatively higher with condom nonuse in the one-nightstand and long-term relationships, particularly the latter. The ratings of erotic feelings showed no significant interactions between gender and condom use or gender and relationship exclusivity. 3.5.10. Reptilian emotions: power Results for power are illustrated in Fig. 4F. As noted, ratings of power increased slightly as relationships became more exclusive, and feelings of power were marginally stronger when condoms were not used. As with eroticism, both the linear and quadratic components of the interactions between condom use and relationship were significant in power (h=.17, P=.031 for linear; h=.29, P < .001 for quadratic), indicating that feelings of power were relatively higher with condom nonuse in the one-night-stand and long-term relationships, particularly the latter. As expected, power ratings showed a significant interaction between gender and condom use/nonuse: females indicated relatively greater power when condoms were used and males indicated relatively greater power when condoms were not used (h=.26, P=.001). This finding is relevant to the ASP hypothesis.
Fig. 5. Gender differences in anger ratings related to condom use/nonuse.
3.5.11. The ASP hypothesis The finding that power was higher for men when condoms were not used and higher for women when condoms were used was consistent with the ASP hypothesis. Examination of the data revealed that ratings of anger were also consistent with expectations: men reported more anger than women when a condom was used and women reported being more angry when it was not used (see Fig. 5). Ratings of selfishness did not however show the expected pattern.
4. Discussion 4.1. Summary To summarize the major findings, results using the SAFECOM scale indicate that positive emotions are high and negative emotions low when condoms are used; when condoms are not used these emotions vary widely with the exclusivity of the relationship. Also, feeling erotic, powerful and intimate are higher when condoms are not used. Finally, when condoms are not used feelings of power are higher for men than women and feelings of anger are higher for women than men. The pattern of results was noteworthy in several respects. First, as a general point, the very strong relationships between reported emotions, relationship exclusivity and condom use/ nonuse reflected in the high effect sizes are consistent with the notion that emotional variables exert important influences on the decision to use or not use condoms. A second noteworthy result concerns the intricacy and subtlety of the influence of specific emotions, including reptilian and prosocial emotions not often recognized in many contemporary emotion theories. As expected, both men and women perceived condom nonuse to be associated with higher reptilian emotions of eroticism and power, but the complex results concerning positive prosocial emotions was not expected. That is, condom use as opposed to nonuse was seen to be associated positively with caring feelings but negatively with intimacy, and loving feelings were not significantly associated with condom use/nonuse. Clearly,
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
although condom use is associated with feelings of confidence and security, and lower negative emotions, condom use is also associated with significant emotional penalties as well. It is east to see how condom nonuse could occur when negative emotions are reduced by, for example, the effects of alcohol or other drugs. A related point concerns how the negative emotions associated with condom nonuse are almost entirely mitigated by perceptions that a relationship is exclusive. In this regard the finding of significant quadratic components in the Condom Use Relationship interactions—indicating that erotic and powerful feelings were relatively higher with condom nonuse in the one-night-stand and long-term relationships—is particularly noteworthy. The ‘‘reptilian rewards’’ appear to be higher in precisely those relationships where the emotional costs of condom nonuse are likely to be mitigated: by the use of drugs/alcohol in the case of the one-night-stand and by the perception of exclusivity in the long-term relationship. This is a case in which specific messages might be targeted to warn of these situations of particular peril. A fourth point concerns gender differences in reported emotion. Overall, women reported feeling more insulted, unsure, nervous, uncomfortable, ashamed, guilty and embarrassed than men, and men reported feeling more satisfied, confident, loving, caring and erotic than women in responding to these questions. These results are generally consistent with expectations from both evolutionary and cultural approaches to gender differences, although as noted the effect sizes in these cases were low to moderate at best. More revealing were the interactions between gender and condom use, with women relative to men reporting greater feelings of shame, embarrassment, guilt, fear, anger, unsureness and insult, and lower feelings of confidence, satisfaction and power when condoms were not used. The interactions with anger and power were consistent with the ASP hypothesis, but it is clear that many other emotions evoked by condom use –nonuse are different in women and men. For example, it is noteworthy that men’s perceptions of high erotic feelings when condoms were not used were particularly strong in the less exclusive relationships. Again, this suggests a need for special messages directed to women and men to help facilitate accurate understanding and communication between women and men. A fifth point is that the positive prosocial emotions and erotic feelings did not show interactions with gender and condom use. Women and men agreed that, not only the ‘‘reptilian reward’’ of erotic feelings, but also the prosocial emotions of intimacy, loving and caring, are higher when condoms are not used in long-term relationships. These perceived emotional rewards associated with condom nonuse, and corresponding perceived emotional penalties associated with condom use, must be recognized and addressed. A specific implication is that there is a need for messages associating condom use with emotional rewards, including eroticism and intimacy as well as loving and caring.
655
4.2. Implications One of the major conclusions of the present studies is that an emotion typology based upon brain theory and research worked successfully in an application to human persuasion (Chaudhuri and Buck, 1994). These basic emotion control systems are organized to respond rapidly and coherently, reflecting primal survival needs: thus, as noted, Le Doux (1994) found evidence that the amygdala is ‘‘involved in processing the emotional significance of simple sensory events’’ (p. 220), even in the absence of corticocognitive processing. The present studies demonstrate that these highly conserved neurochemical systems are intimately involved in the persuasion process in a way and to an extent that has not heretofore been appreciated. The neurological approach used here led us to emphasize the importance of emotions that are not widely recognized in contemporary emotion theory, but are arguably crucial in understanding human behavior, particularly behavior relevant to STDs and to political persuasion. Foremost among these are the reptilian emotions: ‘‘raw’’ sex and aggression. The importance of sex is assumed in many psychological theories and is palpable in much commercial advertising: its relative absence in contemporary emotion or persuasion theories is difficult to account for. It is useful to compare the ARI/CASC approach with other attempts to conceptualize and measure emotional and rational involvement, such as the FCB grid (Ratchford, 1997; Vaughn, 1980, 1986) and the Laurent and Kapferer (1985) consumer involvement profile (CIP). The FCB grid classifies purchase decisions into four quadrants: high involvement/thinking, low involvement/thinking, high involvement/feeling and low involvement/feeling. The CIP distinguishes five antecedent conditions producing involvement: personal interest, sign value, pleasure value, risk importance and risk probability (Kapferer and Laurent, 1993). These approaches tend to contrast ‘‘emotional’’ vs. ‘‘cognitive’’ factors, as with the FCB grid contrast of ‘‘feeling’’ vs. ‘‘thinking’’ and the CIP’s distinction between ‘‘pleasure’’ and other variables. The implication is that emotion and cognition are distinct, an assumption with deep roots in Western thought (Buck, 1988, 2000). In contrast, the ARI approach explicitly considers emotion/affect to be a type of cognition (syncretic cognition), and does not consider syncretic cognition to be in any way incompatible with analytic cognition. Indeed, empirically the two are highly correlated: Chaudhuri (1993) found that if participants report making rational distinctions about a product they also report gaining pleasure from it (r=.72). Moreover, the ARI approach does not conceptualize ‘‘involvement’’ as constituting a variable distinct from analytic and syncretic cognition: instead, involvement is defined as the combination of analytic and syncretic cognition: ‘‘involvement is composed of feeling and thinking’’ (Chaudhuri, 1993, p. 154; italics in the original). As noted, this definition is compatible with Batra and Ray’s (1983) definition of in-
656
R. Buck et al. / Journal of Business Research 57 (2004) 647–656
volvement as the quality and depth of cognitive processing: in the view of the ARI model such cognitive processing can be both emotional/affective and rational. In conclusion, the ARI/CASC conception is quite unlike traditional cognitive models of persuasion. However, it is compatible with the new evidence of the importance of emotion in persuasion, and may serve to aid in its conceptualization and measurement in research on health information campaigns that have had uneven success in the past. We submit that the ARI model has the potential to serve as an integrative viewpoint, which will clarify the role of emotion and reason in persuasion, that the CASC Scale represents a new way to conceptualize and operationalize emotion in the context of persuasion research, and that the ARI model and CASC Scale can have a significant theoretical and empirical impact upon the field of persuasion in general. The studies presented here are primarily exploratory. We know too little—and perhaps have not thought sufficiently—about emotional dynamics to make specific predictions or even to explain the difference in patterns obtained in the present studies. The beginning of wisdom is to accept the reality that highly differentiated systems of emotion have profound consequences for attitude formation, maintenance and change.
References Baldwin JD, Whitely S, Baldwin JI. Changing AIDS and fertility related behaviors: the effectiveness of sexual education. Journal of Sex Research 1990;27:245 – 62. Batra R, Ray ML. Conceptualizing involvement as depth and quality of cognitive response. In: Bagozzi RP, Tybout AM, editors. Advances in consumer research, vol. 10. Ann Arbor, MI: Association for Consumer Research, 1983. p. 309 – 13. Buck R. The communication of emotion. New York: Guilford Press, 1984. Buck R. Prime theory: an integrated view of motivation and emotion. Psychol Rev 1985;92:389 – 413. Buck R. Human motivation and emotion. New York: Wiley, 1988. Buck R. What is this thing called subjective experience? Reflections on the neuropsychology of qualia. Neuropsychology 1993;7(4):490 – 9. Buck R. Social and emotional functions in facial expression and communication: the readout hypothesis. Biol Psychol 1994;38:95 – 115. Buck R. The biological affects: a typology. Psychol Rev 1999;106:301 – 36. Buck R. The epistemology of reason and affect. In: Borod JC, editor. The neuropsychology of emotion. New York: Oxford Univ. Press, 2000. p. 31 – 55. Buck R, Chaudhuri A. Affect, reason, and involvement in persuasion: the ARI model. Konsumentenforschung. Munchen: Franz Vahlen, 1994. p. 107 – 17 (Forschungsgruppe Konsum und Verhalten (Hrsg.)). Buck R, Vieira E. Competence, credibility, compassion, and charisma: a study of the emotional bases of leadership. Paper submitted for publication, 2001. Buck R, Chaudhuri A, Georgson M, Kowta S. Conceptualizing and operationalizing affect, reason, and involvement in persuasion: the ARI model and the CASC Scale. Adv Consum Res 1995;22:440 – 7. Carter CS, Lederhendler II, Kirkpatrick B, editors. The integrative neurobiology of affiliation. Annals of the New York Academy of Sciences, vol. 807. New York: New York Academy of Sciences, 1997. p. 260 – 72. Chaudhuri A. Advertising implications of the pleasure principle in the classification of products. In: van Raay WF, Bamossy GJ, editors. Euro-
pean advances in consumer research, vol. 1. Provo, UT: Association for Consumer Research, 1993. p. 154 – 9. Chaudhuri A, Buck R. The relationship of advertising variables to analytic and syncretic cognitions. In: Varadarajan R, Jaworski B, editors. Marketing theory and applications, vol. 4. Chicago, IL: American Marketing Association, 1993. p. 193 – 8. Chaudhuri A, Buck R. Are advertisers using brain theory? Introducing the CASC Scale. In: Park CW, Smith D, editors. Marketing theory and applications, vol. 5. Chicago, IL: American Marketing Association, 1994. p. 161 – 2. Chaudhuri A, Buck R. Affect, reason, and persuasion: advertising variables that predict affective and analytic – cognitive responses. Hum Commun Res 1995a;21(3):422 – 41. Chaudhuri A, Buck R. Media differences in rational and emotional responses to advertising. J Broadcast Electron Media 1995b;39(1):109 – 25. Chaudhuri A, Buck R. CASC—Eine Skala zur Messung emotionaler und rationaler Reaktionen auf Werbebotschaften (CASC—a scale for measuring emotional and rational responses to advertising). Sozialpsychologie 1998;29(2):194 – 206. DeBro SC, Cambell SM, Peplau LA. Influencing a partner to use a condom. Psychol Women Q 1994;18:165 – 82. DiClemente RJ. Predictors of AIDS—preventive sexual behavior in a highrisk adolescent population: the influence of perceived peer norms and sexual communication on incarcerated adolescents’ consistent use of condoms. J Adolesc Health 1991;12:385 – 90. Gerrard M, Gibbons FX, Bushman BJ. Relation between perceived vulnerability to HIV and precautionary sexual behavior. Psychol Bull 1996;119(3):390 – 409. Kapferer J-N, Laurent G. Further evidence on the consumer involvement profile: five antecedents of involvement. Psychol Mark 1993;10(4): 347 – 55. Kowta S. Structure of emotions: evidence for biologically-based and categorical affects and implications for construction of safe sex messages. Presented at the symposium ‘‘Condom use, emotion, and health communication.’’ Meeting of the Eastern Communication Association, New York, NY, April, 1996. Laurent G, Kapferer J-N. Measuring consumer involvement profiles. J Mark Res 1985;22(1):41 – 53. Le Doux J. Memory versus emotional memory in the brain. In: Ekman P, Davidson R, editors. The nature of emotion. New York: Oxford, 1994. p. 311 – 2. Le Doux JE. The emotional brain: the mysterious underpinnings of emotional life. New York: Simon and Schuster, 1996. MacLean PD. The cerebral evolution of emotion. In: Lewis M, Haviland J, editors. Handbook of emotions. New York: Guilford, 1993. p. 67 – 83. McQuarrie EF, Munson JM. The Zaichkowsky personal involvement inventory: modification and extension. In: Advances in consumer research, vol. 14. Ann Arbor: Association for Consumer, 1987. p. 36 – 40. Panksepp J. Neurochemical control of moods and emotions: amino acids to neuropeptides. In: Lewis M, Haviland J, editors. Handbook of emotions. New York: Guilford, 1993. p. 87 – 107. Panksepp J. Affective neuroscience: the foundations of human and animal emotions. New York: Oxford Univ. Press, 1998. Ratchford B. New insights about the FCB grid. J Advertising Res 1997; 27(4):24 – 38. Rikert VI, Jay MS, Gottleib A, Bridges C. Adolescents and AIDS: females’ attitudes and behaviors toward condom purchase and use. J Adolesc Health Care 1989;10:313 – 6. Tucker D. Lateral brain function, emotion, and conceptualization. Psychol Bull 1981;89:19 – 46. Valdiserri RO, Arena VC, Proctor D, Bonati FA. The relationship between womens’ attitudes about condoms and their use: implications for condom promotion programs. Am J Public Health 1989;79:499 – 501. Vaughn R. How advertising works: a planning model. J Advertising Res 1980;20(5):27 – 33. Vaughn R. How advertising works: a planning model revised. J Advertising Res 1986;26(1):57 – 66.
Journal of Business Research 57 (2004) 657 – 664
The impact of cognitive and/or affective processing styles on consumer response to advertising appeals Salvador Ruiz*, Marı´a Sicilia Department of Marketing, School of Economics and Business, University of Murcia, Campus Universitario de Espinardo, 30100 Murcia, Spain
Abstract As advertisers increasingly seek greater communication effectiveness and new forms of media emerge, psychological differences amongst individuals are becoming essential criteria in the design of advertising appeals. The present study considers whether individuals differ in their propensity to rely on affective, cognitive or both systems to process information. This research suggests that persuasive appeals tend to be more effective when the nature of the appeal matches, rather than mismatches, the individual personality-type preferences for processing information. Results show that informational and informational-emotional advertising appeals, which match consumer’s processing style (thinking and thinking-feeling processors, respectively), can generate more positive attitudes toward the brand, purchase intention (PI) and brand choice. D 2004 Elsevier Inc. All rights reserved. Keywords: Advertising; Processing style; Cognitive/affective; Attitudes; Purchase intention; Brand choice
1. Introduction As advertisers increasingly seek greater communication effectiveness, more careful consideration needs to be given to the selection of the type of advertising appeal used for each target group. Qualitative factors associated with the content and/or execution of an ad have an impact on its eventual effectiveness (MacKenzie et al., 1986). There exists a long tradition in advertising research of exploring the relationship between personality and response to advertising messages (LaBarbera et al., 1998). In particular, marketing researchers have long been intrigued by two individual characteristics, affect and cognition, which may have an influence on advertising effectiveness (Fabrigar and Petty, 1999; Harris and Moore, 1990; Zhang and Buda, 1999). In the last 20 years, many similar measures have been developed to account for the individual tendencies to engage in affective or cognitive activities (Hadddock and Zanna, 1993; Sojka and Giese, 1997). This distinction has also been a popular means of classifying types of persuasive com-
* Corresponding author. Tel.: +34-968-363802; fax: +34-968-367986. E-mail address: [email protected] (S. Ruiz). 0148-2963/$ – see front matter D 2004 Elsevier Inc. All rights reserved. PII: S 0 1 4 8 - 2 9 6 3 ( 0 2 ) 0 0 3 0 9 - 0
munication (Fabrigar and Petty, 1999), and some work has investigated the extent to which the two types of persuasive communication change attitudes by different processes (Edell and Burke, 1987; Roselli et al., 1995). Previous research suggests that persuasive appeals tend to be more effective when the nature of the appeal matches, rather than mismatches, these two processing styles (La Barbera et al., 1998). Such an idea can provide a rationale for understanding the reasons why some individuals differ in their responses to ad stimuli. This paper presents a framework based on an independent but interactive relationship between affect and cognition as suggested by Sojka and Giese (1997). Our objective is to illustrate how advertising effectiveness, in terms of attitudes and intentions, is affected by the consumer’s processing style. The main contribution of this paper lies in the analysis of the interactive affect-cognition relationship (i.e., thinking-feeling processing style) effects on advertising effectiveness across different executions of advertising stimuli. In the next sections we review the literature on affective and cognitive processing styles, and present our research hypothesis followed by a detailed description of the methodology. Analysis and discussion are presented at the end of the paper.
658
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
2. Affective and cognitive processing styles Consumer behavior researchers have pointed out that individual differences among message recipients may lead to wide variations in the manner in which people respond to advertising appeals (Moore et al., 1995). Some individuals, when exposed to an emotionally charged advertising appeal, may exhibit a characteristic tendency to experience their emotions with greater magnitude of intensity (Aaker et al., 1986). Similarly, some individuals may exhibit a tendency to engage in and enjoy thinking when exposed to an ad (Cacciopo and Petty, 1982). Personality classifications that focus on the cognitive style (tendency to engage in and enjoy thinking) have usually been referred to as need for cognition (NFC) (Cacciopo and Petty, 1982). The NFC scale allows researchers to distinguish between two groups of individuals, high and low NFC individuals. Haugtvedt et al. (1992) demonstrated that individual differences in the NFC significantly influenced the formation of attitudes toward consumer products. High NFC individuals (i.e., those who are more likely to enjoy thinking) have been shown to process and evaluate advertising information more thoroughly than low NFC individuals (Mantel and Kardes, 1999; Peltier and Schibrowsky, 1994). Previously, Petty et al. (1983) indicated that under a low involvement condition, people process information through the peripheral route rather than the central route. To encourage consumers’ acceptance of claim assertions, advertisers often provide arguments as rationale for the ad (Munch et al., 1993). But also, as the NFC increases, diagnostic information increasingly becomes the focus of consumer attention (Maclnnis and Jaworski, 1989; Petty et al., 1983; Suri and Monroe, 2001). More positive responses are found to be obtained in high NFC individuals with factual appeals or more informationally dense ads and those with high-quality arguments (Geuens and De Pelsmacker, 1998). On the contrary, when there is very little information in the ad and it is mainly emotional, both low and high NFC individuals will present similar advertising effectiveness, given that the latter do not have any information to be processed. Therefore, we propose: Hypothesis 1: Higher levels of advertising effectiveness are associated to high NFC individuals, compared with low NFC individuals, when the ad is mainly informational, and no differences appear when the ad is mainly emotional. However, cognition represents only one mode of information processing. Individuals may also differ in their tendencies to process affective or emotional stimuli (Raman et al., 1995). There are several concepts and measurements developed for the affective processing style such as the affect intensity (Geuens and De Pelsmacker, 1999; Larsen, 1984; Moore et al., 1995), the Feeling-Belief measure (Hadddock and
Zanna, 1993) and the Need For Emotion (Raman et al., 1995). Recently, the Preference for Affect (PFA) scale (Sojka and Giese, 1997) was developed to measure affective processing across all situations, given that the affect intensity scale only measures the intensity with which an individual responds to an affective stimulus (Raman et al., 1995) and the other two, the Feeling-Belief measure and the Need for Emotion scale, are situationally bounded. That is, the latter two consider that the affect processing is a function of the situation as opposed to an individual processing trait, which is relatively stable with respect to situations (Sojka and Giese, 1997). Despite this controversy with the measurement of affective processing in the literature, their results are consistent and give us an idea about the way in which different people respond to emotional appeals. In this sense, Larsen and Diener (1987) confirmed that when people are exposed to equal levels of affect producing stimuli, some individuals consistently respond with high levels of emotional intensity while others respond with only moderate levels. Furthermore Moore et al. (1995) obtained that high affect intensity individuals may be more likely to be persuaded by emotionally charged advertising appeals. Harris and Moore (1990) suggested to investigate the influence of AI on emotional versus informational ads. Moore et al. (1995) found that high AI individuals, compared with low AI, manifested significantly stronger emotional responses to the emotional ad appeal and showed no differences in emotional response intensity when exposed to a nonemotional appeal. However, they did not obtain concluding results related to attitudes. We believe it may be due to the fact that they used a scale that may not be as valid for all situations as the PFA scale. Therefore, based on this and using PFA, we propose: Hypothesis 2: Higher levels of advertising effectiveness are associated to high PFA individuals, compared with low PFA individuals, when the ad is mainly emotional, and no differences appear when the ad is mainly informational.
3. The interaction between affective and cognitive processing styles The link between cognition and affect in an advertising setting has been demonstrated in several studies (Burke and Edell, 1989; Zajonc and Markus, 1982), which suggested that the cognitive and affective processes may proceed independently as well as work together. For that reason, more recent models integrate affective responses (Vakratsas and Ambler, 1999). There are individual differences in the tendency to use affective and cognitive information in guiding attitude (Hadddock and Zanna, 1993) and the scales of NFC and its
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
counterpart, PFA, can be very good instruments in measuring those individual differences as they measure the intensity of cognitive and affective processing styles (Sojka and Giese, 1997). Furthermore, following Sojka and Geise (1997) classification, it is possible to identify four types of individuals according to these two processing styles (affective and cognitive). Their results suggested that affective and cognitive processing systems are independent yet can operate interactively. So, individuals unbounded by a specific situation may demonstrate a propensity to process information using affect, cognition, both or neither (see Fig. 1). Well-known advertising-planning models recommend matching the advertising appeal to the attitude basis, i.e., use rational and informational advertising for ‘‘thinking’’ or ‘‘functional’’ products and emotional appeals for ‘‘feeling’’ or ‘‘transformational’’ products (Dube et al., 1996). In this sense, past research has uniformly suggested that persuasive appeals tend to be more effective when the nature of the appeal matches, rather than mismatches, the basis of the attitude (Fabrigar and Petty, 1999). Extending these findings to the processing styles and using Sojka and Giese’s (1997) classification, we propose that the matching, rather than the mismatching, of the advertising appeal to the groups of processors described in Fig. 1 will produce higher advertising effectiveness. Apart from the two groups strictly based on affect (feeling processors) and cognition (thinking processors) for which the matches represent an advance over the two previous hypotheses, the most interesting issue comes from the combination processors, who tend to use both processing styles, for which ads containing both emotional and informational elements at a significant level should produce higher advertising effectiveness than ads containing only one of these elements. Lastly, for those individuals that present low levels in both variables (i.e. have low interest in processing affective or cognitive information) we cannot predict their behavior. They are passive processors. The foregoing discussion leads to the formulation of the following hypothesis: Hypothesis 3: Higher levels of advertising effectiveness are associated with ads whose composition in terms of emo-
659
tional and/or informational elements matches the individual processing style. Hypothesis 3a: For thinking processors, higher levels of advertising effectiveness are associated with ads that are mainly informational. Hypothesis 3b: For feeling processors, higher levels of advertising effectiveness are associated with ads that are mainly emotional. Hypothesis 3c: For combination processors, higher levels of advertising effectiveness are associated with ads that are both emotional and informational.
4. Method 4.1. Design and subjects Two hundred and sixty undergraduate students were recruited to participate in the study. Most of the studies about advertising have been tested with students, as they are very homogeneous and convenient samples for data collection (Brown and Stayman, 1992). The participation in the experiment was voluntary, and students were told that they were going to participate in a testing project about three camera advertisements. A 2 (between subjects) 2 (between subjects) 3 (within subjects) mixed-factorial experiment design comprising two levels of cognitive processing (high and low), two levels of affective processing (high and low) and three advertisements for one product category (camera) was used to test the study hypotheses. 4.2. The product used A compact camera was used in the experiment. In the three advertisements presented to the sample, there was a different compact camera, though all of them share most of the technical characteristics. We used three existing brands for the advertisements (Bronica, Centon and Soligor). However, none of them is a powerful brand in the Spanish market in order to avoid any contamination due to differences in prior advertising exposure or brand usage (Derbaix, 1995; Singh, Lessig and Kim, 2000). We got the names from several photography magazines and tested, after collecting data, that no one student have heard about them. 4.3. The advertisements
Fig. 1. Classification of the individuals according to their processing style. Source: Sojka and Giese (1997).
Each individual was exposed to three camera advertisements. None of them were real advertisements, but most of the characteristics and types were similar from the existing
660
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
ads that appeared in the magazines (most of them advertised well-known brands). Three advertisements were created that differed in their levels of information and emotions elicited. The first advertisement (Centon) was very informational (strong and objective arguments). The second (Bronica) was very emotional (we used a picture with a bear family in a very pleasant scene) without any technical or objective information about the camera. The third ad (Soligor) was both very informational and emotional at the same time. This third ad included the strongest and most objective arguments of the first ad plus a picture depicting some people watching whales jumping in the sea. 4.4. Measures Our dependent variables are related to the concept of advertising effectiveness, being attitude toward the ad (Aa), attitude toward the brand (Ab), purchase intention (PI) and brand choice (BC). The survey used previously developed scales to measure the three first dependent variables. Each of them was assessed using seven-point semantic differential scales. Positive – negative, good –bad, favourable – unfavourable, nice – not nice and I like it –I do not like it were used to measure Aa with a selection of items extracted from the relevant literature (Coulter and Punj, 1999; Lafferty and Goldsmith, 1999; MacKenzie and Luzt, 1989). Attractive – not attractive, bad – good, nice – not nice, it is worse than competing brands – it is better than competing cameras, it is worthy – it is not worthy and I like it –I do not like it were used to measure Ab using also items that appeared in the literature (Batra and Stayman, 1990; Krishnamurthy and Sujan, 1999; Pham, 1996). Finally, Unlikely – likely, improbable – probable and impossible –possible were used for PI (Zhang and Buda, 1999). For the last dependent measure, brand choice, they had to select among the three cameras, Bronica, Soligor or Centon, supposing they had to buy one at that moment. According to our study design, we have three independent variables, NFC, PFA and type of advertisement. NFC was measured via the 18-item scale developed by Cacciopo et al. (1984) to identify a person’s inherent desire to engage in elaborate processing. PFA was measured with the 13 items developed by Sojka and Giese (1997).
At the end of the questionnaire, students had to choose one of the three cameras they had seen.
5. Results Multiple item scales were averaged and reliability was high for both individual characteristics scales, NFC (a=.80) and PFA (a=.86). Subjects then were rank ordered and divided into thirds. For the NFC scale, one major concern in the decision to use a tripartite partitioning was that university students may be more homogeneous than the general population in this characteristic (Zhang and Buda, 1999). For PFA, we opted by the tripartite partitioning in order to be consistent with the procedure applied for NFC, although median splits (Mantel and Kardes, 1999, Zhang, 1996), tripartite partitioning (Geuens and De Pelsmacker, 1999) and quartile splits (Moore et al., 1995) have been used for this variable. The quartiles split was rejected because of cell size reasons. For both variables, the middle one-third of students were teased out from the experimental sample for each independent variable. After dividing the sample using NFC scores, low NFC individuals rated an average of 0.34 and high NFC individuals rated 2.60 in the nine-point scale (ranging from 4 to 4). Similarly, these rates were 0.09 and 2.61 for PFA. These groups (Fig. 2) were similar to those obtained by Sojka and Giese (1997) indicating four types of processing style. Mean differences for each variable (NFC and PFA) are statistically significant ( F = 137.06, P < .01 and F = 148.6, P < .01, respectively) for the four different groups, indicating that respondents can be categorized as high in cognition but low in affect, low in cognition but high in affect and high and low on both affect and cognition scales. For the dependent measures, all multiple-item measures (Aa, Ab and PI) were also averaged, and the resulting composites were used in all subsequent analyses. Reliability analyses showed that these scales achieved high levels of internal consistency, with coefficient alpha’s ranging from .80 to .90. A manipulation check was performed by measuring the information and emotions that were transmitted by the ads.
4.5. Procedure In the first stage of the experiment, subjects were administered the NFC and PFA scales. After that, subjects viewed each ad for 1 minute, and before viewing the next one, they rated the ad according to its informational and emotional contents and then responded to the questions concerning their Aa, Ab and PI. This process was done for the three ads. In order to prevent order effects, the position of the advertisements was varied through the experiment.
Fig. 2. NFC, PFA and cell sizes for the four groups of subjects. Note: The four categories are based on a tripartite partitioning using the 33 and 66 percentiles of the actual state data.
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664 Table 1 Information and emotions transmitted by the ads Ad type
Information transmission
Emotions transmission
Emotional ad (Bronica) Informational ad (Centon) Emotional and informational ad (Soligor)
2.04 * 6.50 * 5.50 *
5.44 * 1.02 * 5.11 *
661
showed significantly higher advertising effectiveness than their low counterparts through Aa, Ab and PI when exposed to the emotional ad. Again, for the second part of Hypothesis 2, no difference was found with the informational ad between high and low PFA individuals. Results supporting this hypothesis are presented in Table 3. Finally, Hypothesis 3 was tested comparing the effectiveness of each advertisement within each group of individuals described in Fig. 2. For this hypothesis, we used the three advertisements included in the experiment: Centon (mainly informational), Soligor (informational-emotional) and Bronica (mainly emotional). Results are shown in Fig. 3. Fig. 3a shows that thinking processors have a higher attitude toward Centon (C) at the 5% level, while there is no significant difference at the 10% level between Bronica (B) and Soligor (S) (t = 3.27, 2.28 and 1.18 for C vs. B, C vs. S and B vs. S, respectively). The same happens with PI (t = 3.28, 2.43 and 1.17 for C vs. B, C vs. S and B vs. S, respectively) and brand choice (t = 4, 3.12 and 0.63 for C vs. B, C vs. S and B vs. S, respectively). These results partially confirm Hypothesis 3a, given the exception of Aa that shows opposite results to those previously hypothesized. On the opposite corner of Fig. 2, feeling processors (Fig. 3b) also present a consistent response pattern. But for these individuals, all the three pairwise comparisons are not statistically significant at the 10% level neither in terms of Ab (t = 0.0, 0.31 and 0.25 for C vs. B, C vs. S and B vs. S, respectively), PI (t = 1.15, 1.51 and 0.31 for C vs. B, C vs. S and B vs. S, respectively) nor brand choice (t = 1.65, 0.40 and 1.22 for C vs. B, C vs. S and B vs. S, respectively). Aa shows similar results to those obtained for thinking processors. Therefore, Hypothesis 3b is not supported. Thirdly, Fig. 3c shows that combination processors have lower attitude toward Bronica at the 5% level, while there was no significant difference between Soligor and Centon
* Represents significance of difference with respect to the medium point of the scale (3.5): P < .01.
Using two seven-point scales (none –a lot), we determined the extent to which the target ads were informational, emotional or both. t test analysis revealed that the three ads were correctly transformed, that is, our manipulations in emotions and information transmitted were effective. As shown in Table 1, the emotional ad (Bronica) had a mean score of 5.44 in the emotions and 2.04 in the information transmitted. All mean scores differed significantly (in the expected direction) with respect to the medium point of the scale (3.5). In order to test Hypothesis 1, we used Centon (informational) and Bronica (emotional) ads. As shown in Table 2, high NFC subjects showed significantly higher advertising effectiveness when that effectiveness is measured using Ab and PI but not with Aa. The lower rates found for the informational ad compared with the emotional ad in Aa may be explained by the fact that the ad only contained text without any color or design. For the second part of Hypothesis 1, no difference was expected for the emotional ad among high and low NFC individuals, which is supported with the data included in Table 2. The same approach was used in the second hypothesis, but this time emotional ad was considered first. As could be expected on the basis of previous results, high PFA subjects
Table 2 Advertising effectiveness measures (means) for the two NFC groups Ad type
Informational ad (Centon)
NFC variable
Aa
High NFC individuals (n = 89) Low NFC individuals (n = 86) t test
4.40 4.57 0.86
Emotional ad (Bronica)
Ab
PI 5.21 4.83 2.51* *
4.64 4.29 1.79 *
Aa
Ab
PI
5.28 5.21 0.17
4.63 4.65 0.13
3.82 3.93 0.17
* P < .10. * * P < .05.
Table 3 Advertising effectiveness measures (means) for the two PFA groups Ad type
Informational ad (Centon)
PFA variable
Aa
High PFA individuals (n = 97) Low PFA individuals (n = 86) t test * * * P < .01.
Ab 4.46 4.40 0.11
Emotional ad (Bronica) PI
5.06 4.94 1.15
Aa 4.47 4.38 0.61
5.56 4.92 4.28* * *
Ab
PI 4.88 4.37 3.74* * *
4.13 3.66 2.77* * *
662
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
Fig. 3. Advertising effectiveness and processing style.
(t = 2.52, 2.17 and 0.02 for B vs. S, B vs. C and S vs. C, respectively). The same happens with PI (t = 3.77 1.691 and 1.48 for B vs. S, B vs. C and S vs. C, respectively) and BC (t = 3.85, 2.78 and 0.78 for B vs. S, B vs. C and S vs. C, respectively). Higher scores are therefore obtained by the two ads that provide information (strong and objective arguments, although in different amounts) about the two cameras. Based on these results and the behavior of the variable Aa, Hypothesis 3c is not supported.
6. Discussion After years of researching affect and cognition separately, recent studies on preference formation and attitude change have focused on how cognition and affect work together. Considering affect and cognition as generators either of personality types or of processing styles, these two classifications become a powerful tool for segmenting consumers, which can give marketing managers an idea about the ad type best suited to each group of consumers. In this study, results show that when individuals were exposed to ads congruent with their processing styles, in terms of affect and cognition (Hypotheses 1 and 2), higher advertising effectiveness is obtained. Significant effects 1
P < .10.
were found for Ab and PI. Concerning Aa, there are no concluding results in these two first hypotheses because no differences in this variable were found between high and low NFC subjects. The lack of visual attractiveness of the highly informational advertisement (just a list of written strong and objective claims) provoked low scores in attitude toward it, especially because two-thirds of the subjects had previously seen (during the experiment) another camera advertisement with attractive images, and this comparison effect may be the reason for the absence of significant differences between high and low NFC subjects. Furthermore, the findings of this study offer additional support that affective and cognitive processing systems can operate independent but also interactively (combination processors). While different ads do not seem to generate different advertising effectiveness for feeling processors, informational advertisements are associated with higher effectiveness for thinking processors. For combination processors, advertising appeals based on information or information plus emotions obtained the better results. In summary, informational and informational-emotional advertising appeals, which match consumer’s processing style (thinking and thinking-feeling processors, respectively), can generate more positive Ab, PI and BC. The latter result may be explained through some studies that affirm that the relative importance of cognition and affect depends on the context (Vakratsas and Ambler, 1999).
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
Furthermore, according to Sojka and Giese (1997), it is also conceivable that other variables, such as product involvement and risk level, may impact on individual’s actual processing. This may be the reason why the ads containing information have shown greater acceptance. The product used, camera, is a product for which decision making may be more cognitive than emotional, and for that reason, the emotional ad has not demonstrated as much effectiveness as the informational one. Additionally, it is interesting to point out the consistent pattern of results obtained with the variable Aa for the three groups analyzed. Advertisements including pictures (either alone or with information) are always associated with higher levels of Aa than those including only information (text). Even for thinking processors, the informational ad showed the lowest Aa, although brand attitude, PI and probability of choice were the highest for the advertised brand. This result prevent us to think that the simultaneous questioning about attitudes, PI and brand choice may have induced cognitive consistency in the answers. The study findings have several interesting implications for creative advertising decisions assuming knowledge of target personality-type preferences. Although for some subjects the combination of information and emotions can have the same value than only information in the advertisements, especially for consumers who tend to use cognitive as well as emotional information processing systems interactively, others (thinking processors) can consider those emotions as capable of reducing the highly positive effect of strong and objective arguments. Even more, as long as arguments are strong and images are highly emotive, both elements, independently and together, produce the same effect for another group of subjects (feeling processors). Finally, future research should further investigate these four groups of information processors. There is no research concerning passive processors, what types of ads or what combination of emotions and arguments will be more effective for them. Furthermore, considering the four types of processing styles, hypotheses concerning emotional products more than cognitive ones and referring to low versus high involvement products need to be tested.
References Aaker DA, Stayman DM, Hagerty MR. Warmth in advertising: measurement, impact and sequence effects. J Consum Res 1986;12:365 – 81 (March). Batra R, Stayman DM. The role of mood in advertising effectiveness. J Consum Res 1990;17:203 – 14 (October). Brown SP, Stayman DM. Antecedents and consequences of attitude toward the ad: a meta-analysis. J Consum Res 1992;19:24 – 51 (June). Burke MC, Edell JA. The impact of feelings on ad-based affect and cognition. J Mark Res 1989;26:69 – 83 (February). Cacciopo JT, Petty RE. The need for cognition. J Pers Soc Psychol 1982; 42:116 – 31 (January). Cacciopo JT, Petty RE, Kao CF. The efficient assessment of need for cognition. J Pers Assess 1984;48:306 – 7 (June).
663
Coulter KS, Punj G. Influence of viewing contents on the determinants of attitude toward the ad and the brand. J Bus Res 1999;45:47 – 58. Derbaix C. The impact of affective reactions on attitudes toward the advertisement and the brand: a step toward ecological validity. J Mark Res 1995;32:470 – 9 (November). Dube L, Chattopadhyay A, Letarte A. Should advertising appeals match the basis of consumer attitudes? J Advert Res 1996;36:82 – 9 (November/ December). Edell JA, Burke MC. The power of feelings in understanding advertising effects. J Consum Res 1987;14:142 – 3 (December). Fabrigar LR, Petty RE. The role of the affective and cognitive bases of attitudes in susceptibility to affectively. Pers Soc Psychol Bull 1999;25: 363 – 81. Geuens M, De Pelsmacker P. Need for cognition and the moderating role of the intensity of warm and humorous advertising appeals. Asia Pac Adv Consum Res 1998;23:74 – 80. Geuens M, De Pelsmacker P. Affect intensity revisited: individual differences and the communication effects of emotional stimuli. Psychol Mark 1999;16:195 – 209 (May). Haddock G, Zanna MP. Predicting prejudicial attitudes: the importance of affect, cognition, and the feeling-belief dimension. Adv Consum Res 1993;20:315 – 8. Harris WD, Moore DJ. Affect intensity as an individual difference variable in consumer response to advertising appeals. Adv Consum Res 1990; 17:792 – 7. Haugtvedt CP, Petty RE, Cacioppo JT. Need for cognition and advertising: understanding the role of personality variables in consumer behavior. J Consum Psychol 1992;1:239 – 60. Krishnamurthy P, Sujan M. Retrospection versus anticipation: the role of the ad under retrospective and anticipatory self-referencing. J Consum Res 1999;26:55 – 69 (June). LaBarbera PA, Weingard P, Yorkston EA. Matching the message to the mind: advertising imagery and consumer processing styles. J Advert Res 1998;38:29 – 43 (September/October). Lafferty B, Goldsmith R. Corporate credibility’s role in consumers’ attitudes and purchase intentions. When a high versus a low credibility endorser is used in the ad. J Bus Res 1999;44:109 – 16. Larsen R. Theory and measurement of affect intensity as an individual difference characteristic. J Res Pers 1984;21:1 – 39. Larsen R, Diener J. Affect intensity as an individual difference characteristic: a review. J Res Pers 1987;21:1 – 39. MacKenzie SB, Lutz RJ. An empirical examination of the structural antecedents of attitude toward the ad in an advertising pre-testing context. J Mark 1989;53:48 – 65 (April). MacKenzie SB, Lutz RJ, Belch GE. The role of attitude toward the ad as a mediator of advertising effectiveness: a test of competing explanations. J Mark Res 1986;23:130 – 43 (May). Maclnnis DJ, Jaworski BJ. Information processing from advertisements: toward an integrative framework. J Mark 1989;53:1 – 23 (October). Mantel SP, Kardes FR. The role of direction of comparison, attribute-based processing, and attitude-based processing in consumer preference. J Consum Res 1999;25:335 – 52 (March). Moore DJ, Harris WD, Chen HC. Affect intensity: an individual difference response to advertising appeals. J Consum Res 1995;22:154 – 64 (September). Munch JM, Boller GW, Swasy JL. The effects of argument structure and affective tagging on product attitude formation. J Consum Res 1993; 20:294 – 302 (September). Peltier JW, Schibrowsky JA. Need for cognition, advertisement viewing time and memory for advertising stimuli. Adv Consum Res 1994;21: 244 – 50. Petty RE, Cacioppo JT, Schumann DW. Central and peripheral routes to advertising effectiveness: the moderating role of involvement. J Consum Res 1983;10:135 – 6 (September). Pham M. Cue representation and selection effects of arousal on persuasion. J Consum Res 1996;22:373 – 87 (March). Raman NV, Chattopadhyay P, Hoyer WD. Do consumers seek emotional
664
S. Ruiz, M. Sicilia / Journal of Business Research 57 (2004) 657–664
situations: the need for emotion scale. Adv Consum Res 1995;22: 537 – 42. Roselli F, Skelly J, Mackie DM. Processing rational and emotional messages: the cognitive and affective mediation of persuasion. J Exp Soc Psychol 1995;31:168 – 90. Singh SN, Lessig PV, Kim D. Does your ad have too many pictures? J Advert Res 2000;40:11 – 27 (January/April). Sojka JZ, Giese JL. Thinking and/or feeling: an examination of interaction between processing styles. Adv Consum Res 1997;24:438 – 42. Suri R, Monroe KB. The effects of need for cognition and trait anxiety on price acceptability. Psychol Mark 2001;18:21 – 42 (January).
Vakratsas D, Ambler T. How advertising works: what do we really know. J Mark 1999;63:26 – 43 (January). Zajonc RB, Markus H. Affective and cognitive factors in preferences. J Consum Res 1982;9:123 – 31 (September). Zhang Y. Responses to humorous advertising: the moderating effect of need for cognition. J Advert 1996;25:15 – 32 (Spring). Zhang Y, Buda R. Moderating effects of need for cognition on responses to positively versus negatively framed advertising messages. J Advert 1999;28:1 – 15 (Summer).
Journal of Business Research 57 (2004) 665 – 670
Closing the gap between values and behavior— a means–end theory of lifestyle Karen Brunsø*, Joachim Scholderer, Klaus G. Grunert Centre for Research on Customer Relations in the Food Sector (MAPP), The Aarhus School of Business, Haslegaardsvej 10, Aarhus DK-8210, Denmark
Abstract Means – end chain theory and lifestyle are reconstructed within a dual-process framework, incorporating bottom – up and top – down information processing routes. The bottom – up route is defined as a hierarchical categorization process, and the top – down route as goaldirected action. Lifestyle, then, is a system of individual differences in the habitual use of declarative and procedural knowledge structures that intervene between abstract goal states (personal values) and situation-specific product perceptions and behaviors. Access to the intervening knowledge structures is considered a necessary condition for both information processing routes to reach their ends, predicting a strict mediation model. The model is tested on survey data gathered in France in 1998, using the list of values as a measure of abstract goal states, the food-related lifestyle instrument as a measure of intervening knowledge structures, and a newly constructed behavior list as a measure of behavior. Data were analyzed by means of structural equation modeling. Compared against five alternative model structures, the strict mediation model fitted the data best, thus confirming the predictions derived from the reconstructed theory. D 2003 Elsevier Inc. All rights reserved. Keywords: Values; Lifestyle; Behavior; Structural equation modeling; Mediation
1. Introduction The lifestyle construct has a longstanding history in marketing research. First introduced by Lazer (1964), it was mainly used as an umbrella term for arbitrary assortments of ‘‘activities, interests and opinions’’ items (AIO; Wells and Tigert, 1971) by which marketing researchers sought to describe how consumer segments differed from another. The vagueness of such a definition has annoyed many marketing scholars over the years (e.g., Anderson and Golden, 1984; Lastovicka, 1982). Whilst the theoretical status of the lifestyle construct has recently been elaborated within a means – end framework (Brunsø and Grunert, 1998; Grunert, 1993), its construct validity is still highly uncertain. The aim of the present paper is to fill this gap. 1.1. A means –end theory of lifestyle Brunsø and Grunert (1995, 1998) have proposed a lifestyle definition that clearly breaks with the AIO tradition. Their framework is consistent with the means –end approach * Corresponding author. Tel.: +45-89-486-487; fax: +45-86-150-177. E-mail address: [email protected] (K. Brunsø). 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00310-7
to consumer behavior (Olson and Reynolds, 1983), especially in its hierarchical cognitive structure formulation (Grunert and Grunert, 1995). On the top level of their hierarchy, personal values are defined as abstract, transsituationally aggregated cognitive categories. On the bottom level, product perceptions are defined as situation-specific input to a categorization process. Lifestyle is then defined as an intervening system of cognitive structures that link situation-specific product perceptions to increasingly abstract cognitive categories and finally to personal values. The system comprises declarative knowledge (categories, concepts, associative networks) and procedural knowledge (scripts and skills), shaping information processing on bottom – up as well as top –down routes. From a bottom – up perspective, the knowledge structures guide a comprehension process. Declarative knowledge attaches meaning to an incoming product sensation, and procedural knowledge provides behavioral routines to act upon the mental representation of a product. From a top –down perspective, the knowledge structures guide goal-directed action. Personal values are top-level goals that are situationally unspecific (Rokeach, 1968) and not instantiated with respect to a certain object. Declarative knowledge, when activated, yields specific goal instantia-
666
K. Brunsø et al. / Journal of Business Research 57 (2004) 665–670
tions (Gollwitzer, 1999; Pieters et al., 1995) on the level of a certain product category. Procedural knowledge, when activated, provides behavioral routines for acting upon these instantiations so that a behavioral intention can be formed, adapted to situational constraints, and finally be acted upon (Kuhl, 1985). 1.2. Testable predictions Both information processing routes are formulated on the individual level, and both imply a particular sequence of activation. The bottom – up route is driven by external input (product perception), which is thought to trigger a hierarchical categorization process (activating declarative and procedural knowledge structures) that finally results in the activation of the most abstract conceptual level (personal values). Hence, a valid test of the assumed hierarchy would have to manipulate appropriate stimulus characteristics in an experimental design, and assess whether the observed sequence of activation is consistent with the predicted sequence of activation. The top – down route, on the other hand, is driven by stable individual differences in personal values. Stable individual differences in such superordinate goals imply stable individual differences in the activation of subordinate goals and behavior routines that are instrumental in achieving superordinate goals. Finally, frequent activation of particular subordinate goals and behavior routines implies a higher frequency of observable behaviors that are instrumental in achieving these goals as compared to behaviors that are not instrumental in achieving them. Unlike the bottom – up route (which is driven by external input), the top – down route implies stable individual differences, and can thus be tested by means of survey methods. A valid test would have to establish that lifestyle, as defined above, in a strict sense mediates the relation between personal values and the frequency of instrumental behaviors. In a path model, this would imply that values predict lifestyle, and lifestyle predicts behavior, but that there is no direct effect of values on behavior when lifestyle is included in the model. 1.3. Extensions of the general means –end approach At a first glance, our lifestyle model does not seem to add much to the general means –end approach. On closer inspection, however, three important differences become apparent. The first one is conceptual: The general means – end approach presupposes a certain structure in the subjective ‘‘meaning’’ of product. Due to a lack of theoretical rigor, however, it is not quite clear whether ‘‘meaning’’ is to be understood in a semantic or in a motivational sense (see Grunert and Grunert, 1995 for a detailed discussion). The present approach assumes that semantic aspects are predominant when a means – end chain is primed from the bottom end (initiating a categorization process), but that motiva-
tional aspects are predominant when a means – end chain is primed from the top end (initiating goal-directed action). Secondly, our lifestyle model extends the general means –end approach by including procedural knowledge. Scripts and skills add a pragmatic dimension, linking the mere mental representation of a product or a consumption goal to certain behavioral routines so that goal-directed behavior is also predicted by the model. Thirdly and most importantly, our lifestyle model assumes that lifestyle is a strict mediator of the relationship between values and behavior. As Grunert and Grunert (1995) show, the general means –end approach takes quite a liberal stance on the particular mode of mediation—it is just not clear whether direct links from product perceptions to values are consistent with the theory or not. In contrast to this, our lifestyle model makes explicit predictions that are empirically testable. 1.4. Measurement Operationalized in terms of the top – down perspective outlined above, lifestyle becomes a construct that transcends individual products or brands, but may still be specific to a particular class of consumer goods. Brunsø and Grunert (1995, 1998) have developed a survey instrument to measure food-related lifestyle (FRL). The questionnaire measures 23 lifestyle dimensions in five domains: . Ways of shopping. How do people shop for food products? Is their decision making characterized by impulse buying, or by extensive deliberation? Do they read labels and other product information, or do they rely on the advice of experts, like friends or sales personnel? How do they do their shopping—one-stop shopping versus specialty food shops? . Cooking methods. How are the products purchased transformed into meals? How much time is used for preparation? Is preparation characterized by efficiency, or by indulgence? Is it a social activity, or one characterized by family division of labor? To which extent is it planned or spontaneous? . Importance of quality aspects. This refers not to concrete attributes of individual products, but to attributes that may apply to food products in general. Examples may be healthy, natural, fresh, and tasty. . Consumption situations. How are meals spread over the day? How important is eating out? . Purchasing motives. What is expected from a meal, and what is the relative importance of these various consequences? How important are social aspects, hedonism, tradition, and security? The purchasing motives domain measures individual differences in the importance attached to food-specific instantiations of personal values. The importance of quality aspects domain measures a generalized schema for the evaluation of
K. Brunsø et al. / Journal of Business Research 57 (2004) 665–670
667
product attributes. The ways of shopping, cooking methods, and consumption situations domains measure individual differences in the habitual use of scripts and skills. Extensive cross-national analyses of the psychometric properties of the FRL (Grunert et al., 1997; Scholderer et al., in press) indicate that the factorial structure is practically universal to European food cultures and remarkably stable over time.
imposed on region. Face-to-face interviews were then conducted with the person mainly responsible for food shopping and cooking in the household, with an additional quota on age. The mean age of the respondents was 48.17 years (S.D.=15.45), 87% were female.
1.5. Aims of the study
Three instruments were used in this study. The list of values (LOV; Kahle, 1983) consists of nine items measuring sense of belonging, fun and enjoyment, warm relationships with others, self-fulfillment, excitement, being well respected, security, sense of accomplishment, and selfrespect. All items were answered on nine-point scales ranging from not at all important (1) to very important (9). Previous analyses suggest a three-factorial structure for the LOV (Homer and Kahle, 1988; Grunert and Scherhorn, 1990; Grunert et al., 1994). The FRL instrument (Brunsø and Grunert, 1995) is a 69item questionnaire measuring 23 lifestyle dimensions in five major life domains, including ways of shopping (WS; six subscales), cooking methods (CM; six subscales), importance of quality aspects (QA; six subscales), consumption situations (CS; two subscales), and purchasing motives (PM; three subscales). Each subscale consists of three items, to be answered on seven-point scales ranging from completely disagree (1) to completely agree (7). The construct validity of the FRL dimensions has been tested extensively (Grunert et a., 1997; Scholderer et al., in press), indicating that the factor structures of the subscales in each FRL domain conform to simple structure models that are stable across cultures and over time. The food-related behavior (FR-BEH) list was newly constructed to measure a broad range of shopping, cooking, and eating behaviors. It consists of altogether 37 behavioral frequency items, to be answered on seven-point scales with scale points never (1), less frequent (2), one to five times every 6 months (3), one to three times a month (4), one to two times a week (5), three to four times a week (6), and every day or almost every day (7). Preliminary analyses suggest that 30 of the FR-BEH items can be reliably represented by a six-factor simple structure model (Scholderer, Brunsø and Grunert, in press). The rest of the items will be excluded from the present analysis.
The empirical relation between personal values and behavior is generally low (Munson, 1984). A number of studies have tried to bridge the gap with different mediating constructs (e.g., Goldsmith et al., 1997; Homer and Kahle, 1988; van Raaij and Verhallen, 1994; Valette-Florence and Jolibert, 1990), intending to show that there is in fact a link from values to behavior, even if it may not be a direct one. In contrast to this, the theory we outlined above predicts the absence of a direct value-to-behavior link. From our point of view, abstract personal values have to be transformed into specific goals and linked to behavioral routines before they can initiate goal-directed action. Thus, the aim of the present study is to test whether lifestyle, as defined above, is a strict mediator of the value-to-behavior relation, or whether other model structures fit the data better. Altogether, six models will compete against each other:
a ‘‘no-effects’’ model, assuming that values and lifestyle are independent, and that neither affects behavior at all; a ‘‘value-effects’’ model, again assuming that values and lifestyle are independent, that values influence behavior directly, but lifestyle does not influence behavior at all; a ‘‘lifestyle-effects’’ model, still assuming that values and lifestyle are independent, and that lifestyle influences behavior directly, but values do not influence behavior at all; an ‘‘additive-effects’’ model, assuming that values and lifestyle are independent, and that both influence behavior directly; an ‘‘indirect-effects’’ model, assuming that values influence lifestyle, and lifestyle influences behavior, but that there is no direct effect of values on behavior (strict mediation); and a ‘‘total-effects’’ model, assuming that values influence lifestyle, and lifestyle influences behavior, but that there are also direct effects of values on behavior (partial mediation).
2. Method 2.1. Data collection A sample of N=1000 consumers was drawn in France in 1998. Households were selected at random with a quota
2.2. Measures
3. Results To check whether the distributional assumptions of maximum likelihood estimation were met, multivariate skewness and kurtosis statistics were computed for the joint distribution of the 108 observed variables. The distribution deviated significantly from normality. To account for the violation of assumptions, all observed variables were normalized using Tukey’s proportion estimation formula.
668
K. Brunsø et al. / Journal of Business Research 57 (2004) 665–670
Table 1 Goodness-of-fit statistics for value, lifestyle, and behavior measurement models Model MM.1 MM.2 MM.3 MM.4 MM.5 MM.6 MM.7
List of values (9 items, 3 factors) FRL—ways of shopping (18 items, 6 factors) FRL—cooking methods (18 items, 6 factors) FRL—quality aspects (18 items, 6 factors) FRL—consumption situations (6 items, 2 factors) FRL—purchasing motives (9 items, 3 factors) FR-BEH list (30 items, 6 factors)
c2
df
AGFI
RMSEA
133.080 537.648 504.039 902.362 47.568 180.566 2001.491
22 120 120 120 8 24 391
0.941 0.917 0.921 0.869 0.960 0.929 0.842
0.071 0.060 0.058 0.081 0.069 0.080 0.070
All models were estimated by maximum likelihood. Total N=1000; missing data deleted pairwise.
3.1. Measurement models Measurement models were specified separately for the LOV (MM.1), the FRL domains ways of shopping (MM.2), cooking methods (MM.3), importance of quality aspects (MM.4), consumption situations (MM.5), and purchasing motives (MM.6), and the FR-BEH list (MM.7). All models were estimated by means of maximum likelihood using LISREL 8.30 (Jo¨reskog and So¨rbom, 1996). Goodness-offit statistics are presented in Table 1. The RMSEA values remained within conventional acceptance limits (RMSEA 0.900). Taken together, all measurement models showed acceptable fit. 3.2. Structural models Six alternative structures were specified for the relationship among values, lifestyle, and behavior (Fig. 1). In Models 1 – 4, values and lifestyle are exogenous constructs, influencing behavior independently. Value factors, lifestyle factors, and equation errors in the behavior factors were allowed to correlate within their domain boundaries, but not across. In Models 5 and 6, lifestyle is endogenous, assumed to moderate the value – behavior relation. To maintain a
Fig. 1. Alternative structural models of the values – lifestyle – behavior relation.
K. Brunsø et al. / Journal of Business Research 57 (2004) 665–670
comparable error structure, correlations were allowed among value factors, among equation errors in lifestyle factors, and among equation errors in behavior factors, but not across their domain boundaries. All models were specified separately for each of the five FRL domains, providing a test of the hypothesis that each lifestyle domain is a necessary mediator of the value-to-behavior relation. Again, all models were estimated by means of maximum likelihood using LISREL 8.30. Results are presented in Table 2. The RMSEA indicates acceptable fit for all models and is therefore of limited use here. Since the models are not completely nested either, the usual c2 difference tests cannot be applied so that information-theoretic measures have to be used instead. The consistent Akaike information criterion (CAIC) is a maximum entropy measure (Bozdogan, 1987). In situations where several competing models are specified a priori, the one yielding the lowest CAIC is to be selected. The CAIC puts a moderate penalty on model complexity. Consistent Table 2 Goodness-of-fit statistics for alternative value – lifestyle – behavior structures Model
c2
df
RMSEA
CAIC
WS.1 WS.2 WS.3 WS.4 WS.5 WS.6
No-effects model Value-effects model Lifestyle-effects model Additive-effects model Indirect-effects model Total-effects model
5871.715 5726.589 4915.975 4793.297 4798.102 4703.375
1505 1487 1469 1451 1451 1433
0.059 0.058 0.053 0.052 0.052 0.051
7865.488 7767.630 7000.221 6933.971 6913.220 6941.131
CM.1 CM.2 CM.3 CM.4 CM.5 CM.6
No-effects model Value-effects model Lifestyle-effects model Additive-effects model Indirect-effects model Total-effects model
6338.551 6193.426 4923.215 4875.429 4777.425 4738.529
1505 1487 1469 1451 1451 1433
0.061 0.060 0.054 0.054 0.053 0.053
8264.902 8122.289 7158.790 7221.417 7056.619 7131.512
QA.1 QA.2 QA.3 QA.4 QA.5 QA.6
No-effects model Value-effects model Lifestyle-effects model Additive-effects model Indirect-effects model Total-effects model
6526.321 6381.195 5677.567 5602.455 5428.950 5366.468
1505 1487 1469 1451 1451 1433
0.063 0.062 0.058 0.057 0.056 0.056
8643.352 8473.298 7782.685 7832.330 7546.337 7614.946
CS.1 CS.2 CS.3 CS.4 CS.5 CS.6
No-effects model Value-effects model Lifestyle-effects model Additive-effects model Indirect-effects model Total-effects model
4436.375 4191.250 3784.683 3722.979 3602.332 3569.530
925 907 913 895 907 889
0.067 0.065 0.061 0.060 0.058 0.059
5941.752 5774.985 5214.186 5291.890 5007.315 5114.258
PM.1 PM.2 PM.3 PM.4 PM.5 PM.6
No-effects model Value-effects model Lifestyle-effects model Additive-effects model Indirect-effects model Total-effects model
4416.729 4271.604 3959.157 3844.512 3798.784 3697.290
1058 1040 1040 1022 1031 1013
0.064 0.063 0.059 0.059 0.057 0.057
6314.332 6193.995 5745.796 5736.853 5567.578 5583.033
Models were fitted separately for the FRL domains ways of shopping (WS), cooking methods (CM), importance of quality aspects (QA), consumption situations (CS), and purchasing motives (PM). All models were estimated by maximum likelihood. Total N=1000; missing data deleted pairwise.
669
across all five FRL domains, the indirect-effects model yielded the lowest CAIC values (Models WS.5, CM.5, QA.5, CS.5, PM.5). Notably, the total-effects model—also assuming direct effects of values on behavior—did not further improve the fit to the data, so that the strict mediation role of lifestyle assumed by the indirect-effects models can be maintained.
4. Discussion This paper has made an ambitious attempt to explain the overall organization of consumer behavior in terms of superordinate goals, mediating cognitive structures, and observable behavior. The risk of mere theorizing in such a way is all too often that it can explain everything, but predict nothing. Only when a theory introduces particular restrictions does it become amenable to empirical falsification and gains the status of a model. We believe that the present study provides an example on how to overcome such limitations. 4.1. Construct validity problems in means – end chain theory and lifestyle We have related two concepts that have proven of great heuristic value to marketing researchers: means – end chain theory and consumers’ lifestyle. In the past, both constructs have suffered from problems of unconfirmed validity. In the case of lifestyle, the absence of construct validation studies is easy to explain: AIO and related approaches have never attempted anything close to a theory, let alone a nomological network (Anderson and Golden, 1984; Lastovicka, 1982; Roos, 1986). And construct validation without a properly defined construct would have been a pointless exercise anyway. In the case of means – end chain theory, the absence of construct validation studies is more difficult to explain. We believe that there are two main reasons (part of them already discussed by Grunert and Grunert, 1995). Firstly, means – end chain theory is formulated in vague terms. For example, it is not clear whether a hierarchical value map is to be read from the top (as a motivational process), or from the bottom (as a categorization process). In such a situation, it is indeed difficult to derive hypotheses that could test the fundamental implications of the theory. Secondly, the methods employed by means – end researchers impose a presupposed structure on the data. In hard laddering as well as in soft laddering, the interviewing technique forces respondents to elaborate their answers in an increasingly abstract way. A particular sequence is assumed (attributes – consequences – values), but earlier levels of abstraction are apparently not considered necessary for later levels of abstraction to be reached. Moreover, reversed paths are not considered to reflect a different cognitive process. Hence, there is no way data gathered by means of a
670
K. Brunsø et al. / Journal of Business Research 57 (2004) 665–670
laddering interview could disconfirm means – end chain theory. 4.2. A reconstructed model In the present paper, we have reconstructed means –end chain theory and lifestyle within a dual-process model, incorporating a bottom – up and a top – down information processing route. The bottom – up route is defined as a hierarchical categorization process, and the top –down route as goal-directed action. Lifestyle, in our reconstruction, is then a system of individual differences in the habitual use of certain declarative and procedural knowledge structures that on both routes intervene between abstract goal states (personal values) and situation-specific product perceptions and behaviors. Access of the intervening knowledge structures is considered a necessary condition for both information processing routes to reach their ends. This restriction is crucial since it transforms means –end chain theory from a heuristic concept into a falsifiable model. We have tested the model using representative survey data gathered in France in 1998. Predictions were confirmed: Using the LOV (Kahle, 1983) as a measure of consumers’ superordinate goals, the FRL instrument (Brunsø and Grunert, 1995) as a measure of intervening knowledge structures, and a newly constructed instrument (Scholderer, Brunsø and Grunert, in preparation) as a measure of a broad range of food-related consumption behaviors, we could establish for five different subsets of intervening knowledge structures that they were strict mediators of the relationship between goals and behaviors.
References Anderson WT, Golden LL. Life style and psychographics: a critical review and recommendation. In: Kinnear TC, editor. Advances in consumer research, vol. 11. Provo (UT): Association of Consumer Research, 1984. pp. 405 – 11. Bozdogan H. Model selection and Akaike’s information criterion (AIC): the general theory and its analytical extensions. Psychometrika 1987;52: 345 – 70. Brunsø K, Grunert KG. Development and testing of a cross-culturally valid instrument: food-related lifestyle. In: Kardes F, Sujan M, editors. Advances in consumer research, vol. 22. Provo (UT): Association for Consumer Research, 1995. pp. 475 – 80. Brunsø K, Grunert KG. Cross-cultural similarities and differences in shopping for food. J Bus Res 1998;42:145 – 50. Goldsmith ER, Freiden J, Henderson KV. The impact of social values on food-related attitudes. Br Food J 1997;99(9):352 – 7.
Gollwitzer PM. Implementation intentions and effective goal pursuit: strong effects of simple plans. Am Psychol 1999;54:493 – 503. Grunert KG. Towards a concept of food-related lifestyle. Appetite 1993; 21:51 – 5. Grunert KG, Grunert SC. Measuring subjective meaning structures by the laddering method: theoretical considerations and methodological problems. Int J Res Mark 1995;12:209 – 25. Grunert SC, Scherhorn G. Consumer values in West Germany: underlying dimensions and cross-cultural comparison with North America. J Bus Res 1990;20:97 – 107. Grunert SC, Grunert KG, Kristensen K. Une me´thode d’estimation de la validite´ interculturelle des instruments de mesure: le cas de la mesure des valeurs de consommateurs par la liste des valeurs LOV. Rech Appl Mark 1994;8(4):5 – 28. Grunert KG, Brunsø K, Søren B. Food-related lifestyle: development of a cross-culturally valid instrument for market surveillance. In: Kahle L, Chiagouris C, editors. Values, lifestyles, and psychographics. Mahwah (NJ): Erlbaum, 1997. pp. 337 – 54. Homer PM, Kahle L. A structural equation test of the value – attitude – behavior hierarchy. J Pers Soc Psychol 1988;54(4):638 – 46. Jo¨reskog KG, So¨rbom D. LISREL 8: user’s reference guide. Chicago (IL): Scientific Software International, 1996. Kahle LR. Social values and social change: adaptation to life in America New York: Praeger, 1983. Kuhl J. Volitional mediators of cognitive – behavior consistency: self-regulatory processes and actions versus state orientation. In: Kuhl J, Beckmann J, editors. Action control: from cognition to behavior. Heidelberg: Springer-Verlag, 1985. pp. 101 – 28. Lastovicka JL. On the validation of lifestyle traits. J Mark Res 1982;19: 126 – 38. Lazer W. Life style concepts and marketing. In: Greyser SA, editor. Towards scientific marketing. Chicago (IL): American Marketing Association, 1964. pp. 424 – 38. Munson JM. Personal values: considerations on their measurement and application to five areas of research inquiry. In: Pitts RE, Woodside AG, editors. Personal values and consumer psychology. Lexington Books, 1984. pp. 13 – 33. Olson JC, Reynolds TJ. Understanding consumers’ cognitive structures: implications for advertising strategy. In: Percy L, Woodside AG, editors. Advertising and consumer psychology. Lexington: Lexington Books, 1983. pp. 77 – 90. Pieters R, Baumgartner H, Allen D. A means – end chain approach to consumer goal structures. Int J Res Mark 1995;12:227 – 44. Rokeach M. Beliefs, attitudes, and values: a theory of organization and change New York: Free Press, 1968. Roos JP. On way of life typologies. In: Uusitalo L, editor. Environmental impact of consumption patter. Aldershot: Gower, 1986. pp. 38 – 55. Scholderer J, Brunsø K, Bredahl L, Grunert KG. The cross-cultural validity of food-related lifestyles. in press. Valette-Florence P, Jolibert A. Social values, AIO and consumption patterns. J Bus Res 1990;20:109 – 22. van Raaij WF, Verhallen TMM. Domain-specific market segmentation. Eur J Mark 1994;28:49 – 66. Wells WD, Tigert DJ. Activities, interests, and opinions. J Adver. 1971; 27 – 35.
Journal of Business Research 57 (2004) 671 – 677
Consumer innovativeness Concepts and measurements Gilles Roehrich* Ecole Supe´rieure des Affaires, BP 47X, 38040 cedex 9 Grenoble, France
Abstract Consumer innovativeness, as a force that leads to innovative behavior, has often been cited and studied in research on the diffusion of innovation. Surprisingly, it appears that there is still room for discussion about this concept. This article attempts to take stock of this issue. In the first part, the different theoretical definitions of the notion are introduced critically. The second part is devoted to displaying major measurement scales that have been designed with a view to measuring this construct. This review helps in understanding the diversity of approaches to innovativeness. It raises two main questions: (1) Are the different theoretical conceptualizations of innovativeness equally valid and compatible? (2) Do the scales really express each theoretical standpoint? This suggests that the present scales may be imperfect, and construction of a new one may well be of interest. D 2002 Elsevier Inc. All rights reserved. Keywords: Innovativeness; Measurement scales; Innovative behavior; New product; Innovation
1. Introduction As a marketing concept, innovativeness can at the very least be defined as imprecise. Firm innovativeness, or ‘‘creation of newness,’’ depicts a firm’s ability to develop and launch new products at a fast rate (Hurley and Hult, 1998). Product innovativeness, or ‘‘possession of newness,’’ is the degree of newness of a product (Daneels and Kleinsmith, 2001). Consumer innovativeness, or ‘‘consumption of newness,’’ is the tendency to buy new products more often and more quickly than other people (Midgley and Dowling, 1978). In this article, the word ‘‘innovativeness’’ will be used solely with reference to consumer innovativeness. There is no real consensus on the meaning of innovativeness. It may be described as early purchase of a new product (Cestre, 1996), as well as a tendency to be attracted by new products (Steenkamp et al., 1999). Following the distinction made by Midgley and Dowling (1978) between actualized and innate innovativeness, most authors seem to consider innovativeness a trait, the nature of which is still under question. The first part of this article presents the various
* Tel.: +33-4-76-82-78-66; fax: +33-4-76-82-59-99. E-mail address: [email protected] (G. Roehrich). 0148-2963/$ – see front matter D 2002 Elsevier Inc. All rights reserved. doi:10.1016/S0148-2963(02)00311-9
conceptualizations of the consumer innate innovativeness construct found in the literature. Numerous scales have been created for the purpose of measuring innate innovativeness. A comparative analysis of the main scales found in the European and American literature is presented in the second part of this article.
2. The consumer innate innovativeness concept Innate innovativeness is a ‘‘predisposition to buy new and different products and brands rather than remain with previous choices and consumer patterns’’ (Steenkamp et al., 1999). What forces can explain such a predisposition? Four explanations have been proposed: (1) stimulation need, (2) novelty seeking, (3) independence toward others’ communicated experience and (4) need for uniqueness. 2.1. Innate innovativeness as an expression of the need for stimulation Hebb (1955) and Leuba (1955) seem to be the first to suggest that the individual seeks stimulation, and there is an individual optimal level of stimulation. After a thorough review of the different theories concerning this need, Venkatesan (1973) suggested that a relationship of direct
672
G. Roehrich / Journal of Business Research 57 (2004) 671–677
dependency between the need for stimulation and innovative behavior should be considered. Building on Berlyne’s (1960) approach, he shows how new products can help people maintain their inner stimulation at an optimum level in different situations. Many empirical results (Mittelstaedt et al., 1976; Etzel and Wahlers, 1984; Valette-Florence and Roehrich, 1993, for example) validate this theoretical perspective. Going further, Raju (1980) suggests that innovativeness may intervene between need for stimulation and innovative behavior as a mediator variable. Empirical results showing a positive and significant relationship between need for stimulation and innovativeness support this proposition (Joachimstahler and Lastovicka, 1984; Wahlers et al., 1986; Roehrich, 1993). As a theoretical basis of many human activities, need for stimulation may be perceived as an antecedent of new product adoption, either directly or indirectly, through innovativeness. 2.2. Innate innovativeness as an expression of novelty seeking As proposed by Pearson (1970), inherent novelty seeking is an ‘‘internal drive or a motivating strength,’’ which motivates the individual search for new information. Hirschman (1980) asserts that inherent novelty seeking is ‘‘conceptually indistinguishable from the willingness to adopt new products.’’ She considers it a cardinal trait, linked to different forms of behavioral innovativeness through actualized novelty seeking. Actualized novelty seeking translates into a series of activities aimed at finding new information, which leads to three types of behavioral innovativeness: (1) informative innovativeness is the actual acquisition of new information about a new product, (2) adoptive innovativeness is the adoption of a new product and (3) use innovativeness, which has two expressions: (1) using a product in a different way or (2) knowing all the different uses of a specific product. This proposal broadens the scope of innovativeness from interest in new products to interest in any kind of newness: information, ideas or behavior. Venkatraman and Price (1990) also build on Pearson’s (1970) work to make the distinction between cognitive and sensory innovativeness: cognitive innovativeness is a ‘‘tendency to engage with pleasure in new experiences that stimulate thinking,’’ which may be either internal or external, whereas sensory innovativeness is ‘‘a tendency to engage with pleasure in internal experiences like fantasy, dreaming or stimulating and risky activities like ski jumping.’’ This latter innovativeness may be activated by stimuli, which can be internal (dreaming) as well as external (experiences). By focusing on novelty, Pearson (1970) and Hirschman (1980) push innovativeness beyond the realm of new product consumption. For Mudd (1990), rather than solving
questions about its nature, this adds more ambiguity to the concept. 2.3. Innovativeness as independence toward other’s communicated experience Midgley (1977) makes a clear distinction between innate innovativeness, a trait possessed by every human being, and actualized innovativeness, which is actual innovative behavior. Arguing that an innovator will be the first to use a new product, he defines innate innovativeness as ‘‘the degree to which an individual makes innovation decisions independently from the communicated experience of others.’’ Midgley and Dowling (1978) adopt this approach, but they question whether it might not be better to add ‘‘receptivity to new ideas’’ to Midgley’s definition. They finally choose to consider that ‘‘receptivity to new ideas’’ and ‘‘independence toward others’ communicated experience’’ may probably be equivalent. Certain empirical results tend to invalidate this theoretical position. Hirschman (1980) obtained a negative correlation between ‘‘receptivity to new ideas’’ and ‘‘independence of judgment in innovative decisions.’’ This result leads Hirschman to conclude that ‘‘these two operationalizations of innovativeness address probably two different domains of behavior.’’ Carlson and Grossbart (1984) and Bearden et al. (1986) obtain a positive but weak correlation between independence of judgment and innate tendency toward newness. Finally, the ‘‘independence in innovative decision’’ dimension of Le Louarn’s (1997) innovativeness scale is revealed to be independent of the ‘‘attraction to newness’’ dimension of the scale and of possession of new products. Although attractive, the proposal to consider innovativeness as an expression of independence of judgment lacks empirical support. We conclude that although useful in the innovative decision process, autonomy in decision may probably be neither an antecedent nor a facet of innovativeness. 2.4. Innovativeness as an expression of need for uniqueness Simonson and Nowlis (2000) recall that there is tension between two opposite objectives in decision making: conformity and distinction. According to Fromkin (1968), the need for uniqueness pushes the individual to distinguish himself through the possession of rare items, a socially accepted behavior. Snyder and Fromkin (1980) suggest three consequences of the need for uniqueness: (1) the absence of interest in the reaction of others to one’s own different ideas or acts, (2) the desire not to always follow the rules and (3) the willingness to publicly defend one’s opinions. Fromkin (1971) is the first to suggest a link between innovative behavior and need for uniqueness, whereas Gatignon and Robertson (1985) conclude that ‘‘consumers
G. Roehrich / Journal of Business Research 57 (2004) 671–677
673
who depend highly on normative influences (desire for conformity) adopt more slowly.’’ Burns and Krampf (1991) provide the first empirical validation of this theoretical proposition. They demonstrate positive correlation between need for uniqueness and the number of new products possessed. Moreover, this correlation was higher for new products than for new brands. Although supported by only a few empirical results, Fromkin’s sound theoretical proposal suggests that the need for uniqueness can be considered to be a credible antecedent of innovativeness. Firstly, because innovativeness is an easy way to satisfy the need for uniqueness and, secondly, because need for uniqueness includes independence in judgment, which is necessary for innovative purchasing.
solutions within an organization. Finally, Hurt and Alii define innovativeness as ‘‘change willingness.’’ Moreover, two of these scales have a ‘‘creativity’’ dimension, which indicates that the innovativeness concept they measure is not limited to newness consumption. Little research has been undertaken on the Leavitt and Walton’s (see Bearden et al., 1993 for a presentation) and Hurt– Joseph – Cook’s (see Pallister and Foxall, 1995 for a recent study) scales. Kirton’s innovators –adaptators inventory (KAI) raised a far greater interest in the research community (see Mudd, 1995; Foxall, 1995 or Bagozzi and Foxall, 1996 for an overview). Some general conclusions can be drawn from these studies:
2.5. Discussion
There is no consensus in the definition of innovativeness. From ‘‘inherent novelty seeking,’’ which may have consequences other than new product buying behavior, to ‘‘predisposition to buy new products,’’ which defines the concept by its main consequence, through ‘‘independence in innovative decisions,’’ which could not be empirically validated, various authors have given different views of the concept. There is no consensus either on the roots of innovativeness. Of the need for stimulation, novelty seeking, independence in judgment and the need for uniqueness, which are true antecedents of innovativeness? Analysis of existing innovativeness scales may provide insights into these questions.
3. Operational measurements of innovativeness Since the mid-1970s, a stream of research has led to the design of innovativeness scales through a structured validation process. Most of these scales are different in terms of their theoretical premise and internal structure. The resulting set of scales therefore lacks homogeneity. The most representative scales are presented below in two groups: firstly, ‘‘life innovativeness’’ scales, i.e., the ability to introduce newness in one’s life, will be briefly described. Then ‘‘adoptive innovativeness’’ scales will be more thoroughly presented.
These scales tap innovativeness at a high level: items describe attraction to any kind of newness, not only new product attraction; These scales are multidimensional: seven dimensions for Leavitt and Walton’s 24-item scale, three for Kirton’s 32item inventory, four (or five) for Hurt– Joseph – Cook’s 20-item scale; These scales have good psychometric properties, except for predictive validity: only weak correlations, if any, have been found with new product purchase. These scales are very close to each other: Goldsmith and Nugent (1984), then Goldsmith (1990) obtain very high correlation coefficients between the Hurt – Joseph – Cook’s and Leavitt –Walton’s scales (between 0.64 and 0.82) and between the Hurt– Joseph – Cook’s and Kirton’s scales (0.55). These scales do measure very similar concepts. The way their authors present them, their poor predictive validity with new product purchase and the reading of their items (face validity) suggests that they tap inherent novelty seeking more than specific innovativeness. 3.2. Adoptive innovativeness scales The scales presented under this heading have been specifically designed to measure innovativeness as a tendency to buy new products.
3.1. Life innovativeness scales
Table 1 Item sample of the RAJU’s innovativeness scale
Leavitt and Walton’s (1975), Kirton’s (1976) and Hurt et al.’s (1977) scales are included in this category. They are named ‘‘life innovativeness’’ because their scopes go beyond the sole adoption of new products. For example, Leavitt and Walton view innovativeness as a trait ‘‘that underlies the intelligent, creative, selective use of communication for solving problems.’’ Kirtons defines ‘‘innovators’’ as those who tend to search for new problems and original
When I see a new or different brand on a shelf, I often pick it up just to see what it is like A new store or restaurant is not something I would be eager to find out about I am very cautious in trying new/different products I would rather wait for others to try a new store or restaurant than try it myself Investigating new brands of grocery and other similar products is generally a waste of time
674
G. Roehrich / Journal of Business Research 57 (2004) 671–677
3.2.1. Raju’s (1980) scale Raju’s (1980) innovativeness scale is part of a broader scale designed to measure consumer tendencies toward ‘‘exploratory behavior.’’ It consists of 10 items (see items sample in Table 1), 7 of which are common with other dimensions of his exploratory tendencies scale. Only one of these items has any social content: I would rather wait for others to try a new store or restaurant than try it myself. The author’s results show good internal consistency and high correlation with a sensation seeking scale. These results are partly confirmed by Joachimsthaler and Lastovicka (1984), Wahlers and Dunn (1987) and Wahlers et al. (1986). This scale has been criticized for its structure. After a review of the criticisms (Baumgartner and Steenkamp, 1996), the authors propose a modified scale. 3.2.2. Baumgartner and Steenkamp’s exploratory product acquisition These authors distinguish only two dimensions of exploratory buying behavior: exploratory acquisition of products (EAP) and exploratory information seeking (EIS). For them, ‘‘consumers who are high on EAP enjoy taking chances in buying unfamiliar products, are willing to try out new and innovative products, value variety in making product choices, and change their purchase behavior in an effort to attain stimulating consumption experiences.’’ This EAP 10-item scale (items sample are presented in Table 2) is highly correlated with such constructs as stimulation need (.45) and sensory sensation seeking (.43). Its predictive validity is confirmed by correlations with variety seeking behavior (.25) and innovative behavior (.16). Steenkamp and Van Trijp (1996) confirm a significant correlation between EAP and the possession of 46 new products. 3.2.3. Goldsmith and Hofacker’s scale The originality of the scale designed by these authors is that it measures domain-specific innovativeness, which is a ‘‘tendency to learn about and adopt innovations (new products) within a specific domain of interest.’’ They perceive this construct as intermediary between innate innovativeness and innovative behavior, which is empirically, but moderately, validated by Goldsmith et al. (1995). Four of the six items in this scale (Table 3) describe social innovativeness, as the interviewee is compared with others. This scale proved to be unidimensional and highly reliable. Predictive validity is high, with correlations ranTable 2 Item sample of Baumgartner and Steenkamp’s EAP scale I would rather stick to a brand I usually buy than try something I am not very sure of When I go to restaurant, I feel it is safer to order dishes I am familiar with If I like a brand, I rarely switch from it just to try something different I enjoy taking chances in buying unfamiliar brands just to get some variety in my purchase When I see a new brand on the shelf, I’m not afraid of giving it a try
Table 3 Domain specific innovativeness scale Compared to my friends, I own few rock albums In general, I am the last in my circle of friends to know the titles of the latest rock albums In general, I am among the first in my circle of friends to buy a new rock album when it appears If I heard that a new rock album was available in the store, I would be interested enough to buy it I will buy a new rock album, even if I haven’t heard it yet I know the names of new rock acts before other people do
ging from .38 to .63 with new-product purchase. However, the correlation between that scale and an opinion leadership scale (.78 and .80) questions its discriminant validity. Nyeck et al. (1996) used this scale in an international study (Canada, Israel, France). Their results tend to confirm those of Goldsmith and Hofacker, although predictive validity is lower and factorial structure of lesser quality, as for Goldsmith et al. (1995). 3.2.4. Roehrich’s (1995) scale For Roehrich, innovativeness is an expression of two central needs: need for stimulation (Berlyne, 1960) and need for uniqueness (Snyder and Fromkin, 1980). Consequently, his scale comprises two dimensions: hedonist innovativeness (tied to need for stimulation) and social innovativeness (tied to need for uniqueness). Significant items are displayed in Table 4. Internal consistency, trait validity and nomological validity seem acceptable, except for correlations with mental rigidity (up to .42) and dogmatism (from .19 to .37). Correlation with need for stimulation is as expected (from .16 to .18), but surprisingly no correlation with need for uniqueness is presented. Predictive validity tends to be higher (r=.31) with the number of new products purchased than with innovative purchase intention (between 0 and .30). This result is consistent with Midgley and Dowling’s proposal. Other studies using this scale (Roehrich, 1987; D’Hauteville, 1994) confirm these results. 3.2.5. Le Louarn’s scale Building on the works of Midgley and Dowling on one hand and Hirschman on the other, Le Louarn (1997) defines predisposition to innovate as ‘‘a central predisposition to take innovative decisions, which expresses itself at every Table 4 Item sample of Roehrich’s innovativeness scale (free translation) Hedonist innovativeness
Social innovativeness
I am more interested in buying new than known products I like to buy new and different products New products excite me I am usually among the first to try new products I know more than others on latest new products I try new products before my friends and neighbors
G. Roehrich / Journal of Business Research 57 (2004) 671–677 Table 5 Item sample of Le Louarn’s innovativeness scale (free translation) Attraction to newness
Autonomy in innovative decision
Ability to take risks in trying newness
I am the kind of person who tries every new product at least once When I hear about a new product, I try to know more about at the first occasion Before trying a new product, I try to learn what friends who possess this product think about it I seek out the opinion of those who have tried new products or brands before I try them I’d rather choose a brand that I usually buy rather than try something I am not confident in I never buy something I don’t know anything about with the risk of making a mistake
level of human activity.’’ At product consumption level, this predisposition has three main expressions. Table 5 displays items from the three facets of this scale. This scale has proved to have good psychometric properties (internal consistency and validity). Its predictive validity is good: R2 between the score and early new product purchase intention is up to .23. Of the three dimensions, which surprisingly appear to correlate poorly, only newness attractiveness is correlated with innovative behavior.
4. Discussion Except for a few results (relative to factorial structure or some correlations), the scales reviewed in this second part show good psychometric properties. However, they differ in many dimensions. We will concentrate on four of them: dimensionality, implicit content, level of measurement and predictive validity. As a whole, these scales tap different dimensions, the most specific for innovation diffusion are: newness attraction/repulsion (Leavitt and Walton; Hurt, Joseph and Cook; Raju, Baumgartner and Steenkamp; Goldsmith and Hofacker; Roehrich; Le Louarn scales), creativity/originality (Kirton, Hurt, Joseph and Cook scales), risk attraction/ aversion (Leavitt and Walton; Le Louarn scales), attention to others’ opinion (Leavitt and Walton; Le Louarn scales). The implicit content of the scales refers to the ‘‘individual-social’’ dimension of innovativeness. Some items are
675
centered on individual innovativeness (I like to buy new products), whereas others imply a comparison (I like to buy new products before others). We call this dimension implicit because, except for Roehrich and for Steenkamp and Baumgartner (whose scale does not include any ‘‘social’’ item), it is not explicitly mentioned in the theoretical description of the innovativeness concept. A study of the wording of the items reveals that innovativeness dimensions are measured at different levels: at general level, any kind of newness (products, ideas, behaviors, etc.) is concerned. At product level, the items are about innovations or new products (except for EAP scale, which mainly concentrates on different products, either new or unknown) . At domain-specific level, the items are about new products in a specific product category. It is not obvious whether a ‘‘yes’’ at one level would be equivalent to a ‘‘yes’’ at the other one. Finally, as already mentioned by Foxall (1995), the scales’ predictive validity may be disappointing. Most scales demonstrate very low correlation, if any, with innovative behavior. Goldsmith and Hofacker’s domain-specific scale appears to be an exception with a correlation of up to .64. However, this latter correlation seems to be exceptional, as most of the coefficients obtained with this scale are significantly lower, but still greater than .30. This correlation therefore appears to be a reasonable level for innovativeness, as it is obtained by scales from Le Louarn, Roehrich, Steenkamp and Baumgartner or Raju. Finally, we try to link predictive validity with dimensionality and level of measurement of the scales (Table 6). This table clearly demonstrates that when measured at a general level, innovativeness has no predictive validity. When measured at product consumption level, attraction toward newness and social innovativeness dimensions have average predictive validity. The best predictive validity is reached by domain-specific measurement of social innovativeness, which dominates individual innovativeness in the Goldsmith and Hofacker’s scale. Dimensions such as independence of judgment, attitude toward risk/ change or creativity have no predictive validity. Midgley and Dowling (1978) distinguish three levels of innovative behavior: (1) the purchase of a single new
Table 6 Predictive validity of the innovativeness scales, depending on their subdimensions and level of measurement
Newness attraction (individual) Social context
Independence of judgement Attitude toward risk/change Creativity
General behavior
Product consumption
Domain-specific consumption
No predictive validity (Leavitt and Walton, Hurt, Joseph and Cook) No predictive validity (Leavitt and Walton, Hurt, Joseph and Cook)
Low to average predictive validity (Raju, Steenkamp and Baumgartner, Roehrich, Le Louarn) Low to average predictive validity (Roehrich)
Average to high predictive validity (Goldsmith and Hofacker, 1991)
No predictive validity (Leavitt and Walton) No predictive validity (Kirton, Hurt, Joseph and Cook)
No predictive validity (Le Louarn) No predictive validity (Le Louarn)
676
G. Roehrich / Journal of Business Research 57 (2004) 671–677
Table 7 Different levels of predictive validity
Mono products
Mono category
Roehrich (1995)
Le Louarn (1997)
From 0.09 to 0.30 (about 0.15 on the average) 0.26 (perfumes)
From 0.07 to 0.48 (about 0.15 on the average) 0.32 (electronic appliances) –
Multicategories 0.31
Baumgartner and Steenkamp (1996) 0.16
– –
product, (2) the purchase of new products in a single product category and (3) the purchase of new products in any product category. They hypothesize that the role of innovativeness, at the new product attraction level, becomes easier to isolate as the level of innovative behavior rises. Results displayed in Table 7 provide some support for this hypothesis.
5. Summary and future research Of the dimensions which constitute the internal structure of innovativeness scales, two appear to be of great interest: attraction toward newness (individual innovativeness) and speed of adoption (social innovativeness). The former can be found in all the scales presented here, whereas the latter is explicitly included in only one scale, although implicitly present in most. Although common to every scale, individual innovativeness is theoretically linked to different roots: novelty seeking (Le Louarn) or need for stimulation (Raju, Roehrich, Steenkamp and Baumgartner). However, reading the items shows that they are quite similar despite the theoretical differences. Moreover, each scale contains social items, although only one author explicitly identifies this theoretical dimension of innovativeness. These remarks raise the general question of the link between the theoretical foundations of a scale and the wording of the items. Whether individual or social, innovativeness seems to be able to tap on average only about 10% of innovative behavior. Two possible explanations will be pointed out here. Innovativeness is secondary in explaining innovative behavior: most of the explanatory power may come from the way the new product is perceived (Ostlund, 1974; Roehrich, 2001) or from other intervening variables (Midgley and Dowling, 1978). What does ‘‘new product’’ mean for the interviewee? New products belong to a continuum from ‘‘highly continuous’’ to ‘‘highly discontinuous.’’ Do we really know the level at which respondents give their answer? People who want to change their world may not be interested in buying a new detergent; and people who feel adventurous when buying a new perfume may have a low score on innovativeness.
Consequently, further research into the study of innovativeness and its consequences may be helpful. Firstly, an integrative model of innovativeness is needed. This model should simultaneously offer a structured representation of the different levels at which the innovativeness construct has been conceptualized and the theoretical roots of this construct. It should include the different dimensions of innovativeness. Secondly, this model should provide the theoretical foundation for the construction of innovativeness scales, each tapping the phenomena at a specific level and including items specific to the hypothetical dimensions of innovativeness. References Bagozzi RP, Foxall GR. Construct validation of a measure of adaptive – innovative cognitive styles in consumption. Int J Res Mark 1996;13: 201 – 13. Baumgartner H, Steenkamp J-BEM. Exploratory consumer buying behavior: conceptualization and measurement. Int J Res Mark 1996;13: 121 – 37. Bearden WO, Calcich SE, Netemeyer R, Tell FE. An exploratory investigation of consumer innovativeness and interpersonal influences. In: Richard JL, editor. Advances in consumer research, vol. 13. Provo, UT: Association for Consumer, 1986. p. 77 – 82. Bearden WO, Netemeyer R, Mobley MF. Handbook of marketing scales. Newburg Park, California: Sage Publication, 1993. Berlyne DE. Conflict, arousal and curiosity. New York: McGraw-Hill, 1960. Burns DJ, Krampf RF. A semiotic perspective on innovative behavior. Developments in marketing science. 15th Annual Conference, Academy of Marketing Science, vol. 14. 1991;32 – 5. Carlson L, Grossbart SL. Toward a better understanding of inherent innovativeness. In: Russel WB, Robert AP, editors. Proceeding of the A.M.A. educator’s conference. Chicago: American Marketing Association, 1984. p. 88 – 91. Cestre G. Diffusion et innovativite´: de´finition, mode´lisation et mesure. Rech Appl Mark 1996;11(1):69 – 88. Daneels E, Kleinsmith EJ. Product innovativeness from the firm’s perspective: its dimensions and their relation with project selection and performance. J Prod Innov Manage 2001;18(6):357 – 73 (November). Etzel MJ, Wahlers RG. Optimal stimulation level and consumer travel preferences. In: Russel WB, Robert AP, editors. Proceeding of the A.M.A. educator’s conference. Chicago: American Marketing Association, 1984. p. 92 – 5. Foxall GR. Cognitive styles of consumer initiators. Technovation 1995; 15(5):269 – 89. Fromkin HL. Affective and valuational consequences of self-perceived uniqueness deprivation. Unpublished doctoral dissertation. The Ohio State University, 1968. Fromkin HL. A social psychological analysis of the adoption and diffusion of new products and practices from a uniqueness motivation perspective. In: David MG, editor. 2nd annual conference. Ann Arbor, MI: Association for Consumer Research, 1971. p. 464 – 9. Gatignon H, Robertson TS. A propositional inventory for new diffusion research. J Consum Res 1985;11:849 – 67 (March). Goldsmith RE. The validity of scale to measure global innovativeness. J Appl Bus Res 1990;7(2):89 – 97. Goldsmith RE, Hofacker CF. Measuring consumer innovativeness. J Acad Mark Sci 1991;19(3):209 – 22 (summer). Goldsmith RE, Nugent N. Innovativeness and cognitive complexity: a second look. Psychol Rep 1984;55:431 – 8. Goldsmith RE, Freiden JB, Eastman JK. The generality/specificity issue in consumer innovativeness research. Technovation 1995;15(10):601 – 13.
G. Roehrich / Journal of Business Research 57 (2004) 671–677 D’Hauteville F. Un mode`le d’acceptation du nouveau produit par le consommateur: cas du vin alle´ge´ en alcool. Unpublished doctoral dissertation, Montpellier II, 1994. Hebb DD. Drives and the C.N.S. (central nervous system). Psychol Rev 1995;62:243 – 54. Hirschman EC. Innovativeness, novelty seeking and consumer creativity. J Consum Res 1980;7:283 – 95 (December). Hurley RF, Hult GTM. Innovation, market orientation, and organizational learning: an integration and empirical examination. J Mark 1998;62(3): 42 – 54. Hurt HT, Joseph K, Cook C. Scales for the measurement of innovativeness. Hum Commun Res 1977;4(1):58 – 65. Joachimsthaler EA, Lastovicka JL. Optimal stimulation level: exploratory behavior models. J Consum Res 1984;11:830 – 5 (December). Kirton M. Adaptors and innovators: a description and measure. J Appl Psychol 1976;61(5):622 – 9. Le Louarn P. La tendance a` innover des consommateurs: analyse conceptuelle et proposition d’une e´chelle de mesure. Rech Appl Mark 1997; 12(1):3 – 20. Leavitt C, Walton J. Development of a scale for innovativeness. In: Schlinger MJ, editor. Advances in consumer research, vol. 2. Ann Arbor, MI: Association for Consumer Research, 1975. p. 545 – 54. Leuba C. Toward some integration of learning theories: the concept of optimal stimulation. Psychol Rep 1995;1:27 – 33. Midgley D. Innovation and new product marketing. Londres: Croom Helm, 1977. Midgley D, Dowling GR. Innovativeness: the concept and its measurement. J Consum Res 1978;4:229 – 42. Mittelstaedt RA, Grossbart SL, Curtis WW, Devere SP. Optimal stimulation level and the adoption decision process. J Consum Res 1976;3:84 – 94 (September). Mudd S. The place of innovativeness in models of the adoption process: an integrative review. Technovation 1990;10(2):119 – 36. Mudd S. Kirton adaptation – innovation theory: organizational implications. Technovation 1995;15(3):165 – 76. Nyeck S, Sylvie P, Xuereb J-M, Chebat JC. Standardisation ou adaptation des e´chelles de mesure a` travers diffe´rents contextes nationaux: l’exemple d’une e´chelle de mesure de l’innovativite´. Rech Appl Mark 1996;11(3):57 – 74. Ostlund LE. Perceived innovation attributes as predictors of innovativeness. J Consum Res 1974;1:23 – 9 (September). Pallister JG, Foxall GR. Psychometric properties of the Hurt – Joseph –
677
Cook scales for the measurement of innovativeness. Technovation 1995;18(11):663 – 75. Pearson PH. Relationships between global and specific measures of novelty seeking. J Consult Clin Psychol 1970;34:199 – 204. Raju PS. Optimum stimulation level: its relationship to personality, demographics and exploratory behavior. J Consum Res 1980;7:272 – 82 (December). Roehrich G. Nouveaute´ perc˛ue d’une innovation. Rech Appl Mark 1987; 2(1):1 – 15. Roehrich G. Les consommateurs innovateurs: un essai d’identification. Unpublished doctoral dissertation. Ecole Supe´rieure des Affaires de Grenoble, 1993. Roehrich G. Innovativite´s he´doniste et sociale: proposition d’une e´chelle de mesure. Rech Appl Mark 1995;9(2):19 – 41. Roehrich G. Causes de l’achat d’un nouveau produit: variables individuelles ou caracte´ristiques perc˛ues? Rev Fr Mark 2001;182(2001/2): 83 – 98. Simonson I, Nowlis SM. The role of explanation and need for uniqueness in consumer decision making: unconventional choice based on reason. J Consum Res 2000;27(1):49 – 68 (June). Snyder CR, Fromkin HC. Uniqueness: the human pursuit of difference. New York: Plenum, 1980. Steenkamp JBEM, Hofstede F ter, Wedel M. A cross-national comparison into the national and national cultural antecedents of consumer innovativeness. J Mark 1999;63(2):55 – 69. Steenkamp JBEM, Van Trijp HCM. To buy or not to buy? Modeling purchase of new products using marketing mix variables and consumer characteristics. Gainesville, FL: Marketing Science Conference, 1996. Valette-Florence P, Roehrich G. Une approche causale du comportement innovateur. Econ Soc Se´r Sci Gestion 1993;19:75 – 106 (October, SG). Venkatesan M. Cognitive consistency and novelty seeking. In: Ward S, Robertson TS, editors. Consumer behavior – theoretical sources. Englewood Cliffs: Prentice Hall, 1973. p. 355 – 84. Venkatraman MP, Price LL. Differentiating between cognitive and sensory innovativeness. J Bus Rev 1990;20:293 – 315. Wahlers RG, Dunn MG. Optimal stimulation level measurement and exploratory behavior: review and analysis. AMA Winter Educators’ Conference, San Antonio, TX.1987;249 – 54. Wahlers RG, Dunn MG, Etzel MJ. The convergence of alternative OSL measures with consumer exploratory behavior tendencies. In: Richard JL, editor. Advances in consumer research, vol. 13. Provo, UT: Association for Consumer Research, 1986. p. 398 – 402.
Journal of Business Research 58 (2005) 1407 – 1408
Editorial
Preface to La Londe 2003 special issue: communications and consumer behavior
This special issue represents the best of the 30th International Research Seminar in Marketing organized by the Aix Graduate School of Management (I.A.E. Aix). The seminar, now better known as the bLa Londe SeminarQ, is devoted to Marketing Communication and Consumer Behavior on a biennial basis. The conference includes indepth discussions of current research topics. There are only two tracks, and each paper is allotted forty minutes for presentation. Audience participation is not only encouraged, it is expected, so that authors may get the best feedback on their work and ideas. A total of 71 papers were submitted for double-blind review to the conference, and 36 papers were selected for presentation by a distinguished group of reviewers. The authors and the reviewers covered all five continents. From the 36 papers, we selected a subset for inclusion to the special issue of the Journal of Business Research that represents a diverse sampling of the best papers. Among them is the article awarded best conference paper: bBehavioral Outcomes from On-line Auctions: Price, Reserve Disclosure and Initial Bidding InfluencesQ by Mathew Walley and David Fortin, University of Canterbury, New Zealand. Their paper is outstanding for several reasons. First, they managed to run a clean field experiment with 120 respondents, without knowledge of their participation. Second, their analyses and reporting of the data and results were very well done and extremely clear. And third, their significant findings are counter to other research in the area, and therefore, this work should lead to replication. This carefully designed experiment must be compared against other works for the field to reconcile the differences reported in the literature about reserve prices and disclosure in on-line auctions. Other papers selected reflect vastly different approaches to topics. bThe Role of Affective Expectations in Memory for a Service EncounterQ by Elizabeth Cowley, Colin Farrell and Michael Edwardson of the University of New South Wales involves the application of luck and chance to interpretations of service. Their results showed that indi0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2004.05.005
viduals with different orientations interpreted the same service experience in a different manner. The importance of affective expectations may be just as important as the experience in determining the affective reaction or outcome of the experience for some people. The paper by Flemming Hansen, bDistinguishing between Feelings and Emotions in Understanding Communication EffectsQ, is a synthesis and review of advertising response measures and delves into the distinction between cognitive and emotional information processing. Much of the paper is based on primary studies at the Copenhagen Business School, which makes the leap from pencil and paper reports of feelings to inferences about underlying emotional dispositions. It is suggested that the distinction between central/peripheral, higher/lower involvement or more or less cognitive information processing is useful with respect to measuring advertising effectiveness. It may be that emotional affect makes information processing more efficient. A very interesting partner to the Hansen paper is the Laros and Steenkamp paper, entitled bEmotions in Consumer Research: A Hierarchical ApproachQ. This paper uses the same battery of feelings used by Hansen (Richens CES 1997), but in a different domain, the consumption of food. The results of their analyses expand on the positive and negative emotional response categories to give a richer dimension to each emotion. One of the contributions of the Laros and Steenkamp paper is the exhaustive literature review and classification tables of emotions. Other researchers in this area are sure to find it valuable. The paper by Sintas and Alverez, bFour Characters on the Stage Playing Games: Performing Arts Consumption in SpainQ, looks at the symbolic consumption of art and how it relates to social class. Their paper is based on a fantastic data set of in-home interviews of 12,000 randomly selected households in Spain about their bhabits of cultural consumptionQ. Segments based on consumption habits show how the social status of individuals is linked to the world of arts. Their analyses question some of the classical approaches by Bourdieu.
1408
Editorial
The final contribution is a contrast in research styles to the other papers. bResearching Cultural Metaphors in Action: Metaphors of Computing Technology in Contemporary U.S. LifeQ by Denny and Sunderland shows how the cultural metaphor is useful as an analytic tool. Using ethnographic research, they demonstrate how the computer and the Internet have come to change the meanings within our day-to-day lives. In sum, these six papers represent a diverse array of research topics and styles that add to our understanding of consumers and how they relate to their environment. We have learned from reading and editing these papers and others that were submitted to the conference. We hope you enjoy reading them and encourage you to submit your work
and future the next La Londe conferences. You will find it an enriching academic experience. Gilles Laurent Carrefour Chair, HEC School of Management, Paris Judy Zaichkowsky Simon Fraser University, Canada E-mail address: [email protected]. Corresponding author. Department of Marketing, Groupe H.E.C., 78351 Jouy-en-Josas Cedex, France. Tel.: +33 1 39 67 74 80; fax: +33 1 39 67 70 87. 1 February 2003
Journal of Business Research 58 (2005) 1409 – 1418
Behavioral outcomes from online auctions: reserve price, reserve disclosure, and initial bidding influences in the decision process Matthew J.C. Walley*, David R. Fortin* Department of Management, University of Canterbury, Private Bag 4800, Christchurch 8004, New Zealand Received 1 November 2002; received in revised form 1 August 2003; accepted 1 October 2003
Abstract This study presents a conceptual model of the online auction consumer decision process and empirically tests some of the model relationships by using an experimental design. The factorial design used a combination of reserve price, reserve disclosure, and initial bidding as treatments to impact two behavioral outcome variables: final price sold and auction interest. The data were gathered via a real existing online auction site where products were auctioned, recreating a true market-based environment. The key finding is that having a $1 reserve price that is disclosed is the optimal strategy to adopt when selling goods in an online auction, if the goal of the seller is to maximize sale price and auction interest. The findings of this research provide insight into how and to what degree these factors influence the dependent variables, ultimately providing online auction hosts and sellers with possible methods for maximizing their potential income from online auctions. This research contributes in synthesizing the developed theories underlying traditional bin personQ auctions, with the emerging and rapidly growing medium of online auctions. D 2004 Elsevier Inc. All rights reserved. Keywords: Auctions; Online; Web; Reserve bidding; Hits; Behavior
1. Introduction Online auctions attract thousands, sometimes millions, of bidders for items ranging from commodities to collectibles (Massad and Tucker, 2000). Investors expect a tremendous amount of growth in online auctioning, assigning huge price/earning multiples to the major players in the industry (Massad and Tucker, 2000). Consequently, from the perspective of auction site owners and the sellers of goods online, it is extremely useful to understand what factors of the auction influence the overall interest in, and ultimately the final sale price of auctioned products. This study presents a conceptual model of the online auction consumer decision process and empirically tests * Corresponding authors. Matthew J.C. Walley is to be contacted at Tel.: +64 3 376 4142; fax: +64 3 376 4142. David R. Fortin, Tel.: +64 3 364 2987x7026; fax: +64 3 364 2020. E-mail addresses: [email protected] (M.J.C. Walley)8 [email protected] (D.R. Fortin). 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.10.014
some of the model relationships by using an experimental design tracking observation-based data points. The research focuses on how the factors of auction reserve price, the disclosure or nondisclosure of this reserve price, and the amount of initial bidding history in the auction influence the final sale price, and overall interest in the auction. Practically, the findings of this research provide insight into how and to what degree these factors influence the final sale price, providing online auction hosts and sellers with possible methods for maximizing their potential income from online auctions.
2. Literature review Among the most popular commerce exchange applications of the World Wide Web is the online auction. The availability and convenience of the Internet, in conjunction with the variety of products available has contributed to online auctions sites reporting 600% growth in auctions held
1410
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
(Massad and Tucker, 2000). Forrester Research (2002) predicts that the number of online auction buyers will grow from 3 million in 1998 to at least 14 million by 2003. In addition to the growing number of users, the estimated number of auction sites has also increased dramatically, as there are now more than 1660 auction sites listed on the online auction portal web site Internetauctionlist (2002). Online auctions are a new form of transaction that introduces variability into nearly all aspects of a commercial exchange. The process can be mystifying. Prices are unpredictable, consumers and suppliers are strangers, and comparison is difficult because products come packaged with services, and items on offer may be unique or obsolete (Massad and Tucker, 2000). As a consequence, some researchers believe auction itself is an inadequate term for describing relationships powered by advanced algorithmic models (Herschlag and Zwick, 2000). Online auctions represent a model for the way the Internet is shaping the new economy. In the absence of spatial, temporal, and geographic constraints, these mechanisms provide many benefits to both buyers and sellers. They are now an important component of the portfolio of mercantile processes that are transforming the economy from traditionally hierarchical to free market oriented structures (Gupta and Bapna, 2001). Kauffman and Walden (2001) provide a comprehensive review on this topic. According to Gupta and Bapna (2001), online auctions fall under the category of web-based dynamic pricing mechanisms. Under these mechanisms, consumers become involved in the price-setting process and they can experience the thrill of bwinningQ a product, potentially at a bargain, as opposed to the typically relatively tedious notion of bbuyingQ it. For sellers, these mechanisms are likely to bring access to newer markets, help clear aging or perishable inventory, and provide experiential and at times viral marketing capabilities. Nowhere are these trends as visible as in the hugely popular online auction site, eBay.com. The most popular auction sites experience millions of monthly visitors, which provide the necessary critical mass of buyers and sellers to set market prices for their goods. Massad and Tucker (2000) argue that online auctions are inherently less risky than traditional in-person auctions because the perceived risk in a consumer behavior context is largely moderated by the buyer’s ability to search out information online, thus reducing purchase uncertainty. However, Johns and Zaichkowsky (2003) emphasize some relevance of live auctions in understanding online auction buyer behavior. Commonly, in traditional consumer transactions, the most relevant matter over which buyers are uncertain is the appropriate price. Online auctions offer unprecedented access to information regarding previous sales of products, comparable to those for which they are considering placing a bid. This access allows buyers to analyze the market and develop a bidding price with greater confidence. Due to
the reduced price information risk, bidders with more information are likely to be willing to pay more than bidders with less information (Cox, 1963). The disclosure of a reserve price increases the amount of information available to the buyer; accordingly, there is some justification for believing a disclosed reserve will increase the final sale price of the goods. A key factor that makes electronic markets, such as online auctions, interesting is the potential for achieving higher efficiency. On the surface, e-markets such as eBay, with millions of registered users, would appear to be a close approximation to an economist’s idealization of a frictionless, efficient market (Gupta and Bapna, 2001). Auctioning is among the most efficient and fastest concept available to achieve fair and competitive prices and identify the optimal business partner (Rumpe, 2003). Auctioning survives via its networking capability both from a business-to-business (B2B) and a computer-tocomputer perspective (Varadarjan and Yadav, 2002). One thing for certain is that the use of information technology has brought this sphere of economic activity out of the domain of specialists to that of the technologically layperson (Gupta and Bapna, 2001). 2.1. Market segmentation According to Chen and Wilson (2002), the online auction market can be broken down into three subgroups based upon the roles of buyers and sellers; these are person-toperson, business-to-person (B2P), and B2B. First, host auction sites that serve the person-to-person segment mainly provide sellers and buyers with a public marketplace to trade online and sometimes help both parties to deal with the transactions, including billing, shipping, and handling. Popular sites that service this segment are eBay and Yahoo! The revenues of person-to-person auctions from eBay style auction sites accounted for 70% of the US$1.4 billion in goods sold online in 1997 (Chen and Wilson, 2002). Second, business-to-person host auction sites provide a digital marketplace for business suppliers to sell their excess inventories or refurbished goods to end-users. In addition, suppliers can also serve as retailers that actually get involved in the distribution of auctioned items, such as uBid. Third, the B2B segment transacts with both sellers and buyers and treats them as business units. The purpose of executing an auction can be for resale of those items to individual customers or for their own procurement, online sites that use this model include ComAuction. According to predictions made by several research firms, the B2B electronic commerce market will exceed US$4 trillion by the year 2005 (Johnson and Friedman, 2002). The concept of reverse auctions is also making some inroads, especially with the B2B segment Johnson and Friedman (2002). However, as this research focuses on online auctions in the person-to-person segment, the B2P and the B2B segments will not be elaborated upon.
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
2.2. The traditional English auction model Single-item auctions are asset exchange mechanisms, which have been extensively theoretically studied, beginning with the seminal article by Nobel laureate William Vickrey (1961). More recent coverage can be found by referring to Rothkopf and Harstad (1994). Researchers studying this auction commonly make use of assumptions, such as the independent private values (IPV) assumption to derive its equilibrium characterization. Such an assumption implies that a single indivisible object is to be sold to one of several bidders (Gupta and Bapna, 2001). Each bidder is risk-neutral and knows the value of the object to him or herself, but does not know the value of the object to other bidders. It also implies that there is a finite population of bidders, each of which draws the bidder’s valuation independently from some given continuous distribution (see Milgrom, 1989, for details). For most goods being auctioned, the IPV assumption is robust. However, for collectibles, it is reasonable to assume that an individuals’ valuation will be dependent on the valuations of fellow bidders (Smith, 1989). Presumably, a profiteering collector will have the objective of at least recovering the cost of the item purchased and thus will implicitly carve his or her valuation distribution dependent on that of other bidders (Gupta and Bapna, 2001). In addition to offering single-item products for auction, English auctions enable the bidders to place open bids on the Internet detailing the amount the individual wishes to bid at, as opposed to sealed tender bids. These bids incrementally increase in value with the product being sold to the highest bidder at the end of the predetermined closing date. Auctions can be either short or long term. Long-term auctions may last up to 4 weeks, where the bidders repeatedly review the current situation. Consequently, such kinds of auctions are mainly of interest during the closing phase. On the other hand, short-term auctions simulate these last few hours immediately, creating immediate interest. These auctions may be as short as 30 minutes (Rumpe, 2003). 2.3. Gaps in the literature Interestingly, the advent of auctions over open Internet Protocol-based networks, such as the Internet, has also facilitated the pursuance of a richer set of empirically derived methodologies by today’s researchers. Most preInternet-based auction research was either purely theoretical in nature (McAfee and McMillan, 1987; Milgrom, 1989; Myerson, 1981) or involved laboratory experimentation (Kagel and Roth, 1997). Empirical research was rare, due the lack of meaningful data sets, which in turn could be attributed to the lack of mainstream appeal of online auctions. Lucking-Reiley (1999) acknowledges the difficulty in obtaining field data for testing long-standing hypotheses, such as the supposed revenue equivalence between the basic auction formats. Given the accessibility
1411
of the Internet, it opens the opportunity to contribute empirically to the field of online auctions. The benefits of online auctions to both sellers and buyers are undeniable. However, to date, much of the literature has focused on the benefits of individual consumers and businesses. Researchers are only beginning to question what provides benefits to individual online sellers. Specifically, what combination of factors will enable the seller to gain the maximum price for his or her product? This research proposes to narrow the gap in the literature by empirically measuring the impact of external variables or cues on outcome variables, such as the final sale price and amount of consumer interest in the auction. 2.4. Potential independent variables 2.4.1. Reserve price Waehrer et al. (1988) state that the existence of a reserve price may attract more competing merchandise into auction since reserve prices attract risk-averse sellers. Thus, the effect of reserve pricing may be to lower the average price ultimately paid in seller reserve auctions, while increasing the average size of opening bids (Massad and Tucker, 2000). Regarding auction interest, Vincent (1995) argues that higher reserve prices have the effect of bincreasing bidder participation at higher levels since the reserve would have the effect of eliminating low-valuation bidders from contentionQ (Massad and Tucker, 2000). Horstman and LaCasse (1997) argue that the disclosure of the sellers reserve has the effect of conveying information to the bidder regarding the seller’s estimation of the items value, and should therefore result in a higher selling price. Bajari and Hortacasu (2000) also find that the minimum bid is the most significant determinant of whether a bidder enters an auction. Additionally, recent research shows that a zero reserve price provides higher expected profit than a reserve price greater than or equal to the auctioneer’s salvage value for the good (Gupta and Bapna, 2001). According to Massad and Tucker (2000), when seller reserves are present, initial bids at online auctions will supersede those at in-person auctions, and because the bidders have no information to offset the information advantage of buying online, the final selling price at online auctions will exceed that of in-person auctions. Therefore, given the extensive consideration with respect to the influence of reserve price over the amount of interest and final sale price obtained in the auction, we believe there is a need for further investigation of this variable. 2.4.2. Bidding history Engelbrecht-Wiggans et al.’s (1983) early work in bidding theory proposes that it is badvantageous for a bidder to bid more aggressively when the competition increases, that is, when there are more competitors and/or more aggressive bidding competitors.Q Furthermore, Smith (1989) believes that the social inertia created by increased bidding competition will often increase the final sale price
1412
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
and overall interest in the auction. This is also supported by Johns and Zaichkowsky (2003) in the realm of online auctions as they state that the bthought of competition increases arousal in consumers and may lead to bidders being swept away in a bidding war they never intended.Q Dholakia et al. (2002) also support these concepts; they determine that potential bidders are susceptible to herding biases in online auctions. Specifically, they found that bidders gravitate toward listings with existing bids, while demonstrating no consideration for listings without any existing bids. Accordingly, we contrive previous bidding levels at the outset of our auction experiments to test if these findings hold true for online auctions. 2.5. Proposed outcome indicators The following explanations justify why we believe the potential dependent measures of final sale price and auction interest are important in increasing the general understanding of online auctions from a seller’s perspective. Accordingly, it is assumed that sellers are concerned with maximizing their financial gain in online auctions. 2.5.1. Price Massad and Tucker (2000), propose that as bidding increases during online auctions bbidders are able to access data regarding previous sales, and therefore the maximum price bidders are willing to pay will increase due to improved information.Q Specifically, they state that bthe existence of seller reserves in online auctions can be expected to moderate this effect,Q moreover, it is proposed that the inclusion of a reserve price in an online auction, will increase the final price obtained and the initial bidding price (Massad and Tucker, 2000). Smith (1989) argues that in most auctions, it is the price that is of paramount concern for sellers. 2.5.2. Auction interest This can be approximated by using the amount of hits received on a particular auction. Massad and Tucker (2000) propose that the increased presence of bidders in a particular online auction will result in higher sale prices being obtained, due to increased bidder competition. Previous research found strong evidence that download counts is used by consumers as surrogates of product quality to download the software. These impacts were nonlinear and dynamic. This result may be applicable to online auctions with regards to the number of hits, and it is hypothesized that more hits on a particular auction will stimulate bidder interest and increase the number of bids placed on the auction. 2.6. Potential influencing factors The factors of gender, previous reputation of individual bidders (labeled purchasing profile in this research), the
aggregate average profile of bidders (labeled average bidding profile), and the number of unique bidders may influence the outcome indicators. 2.6.1. Gender Traditionally, in auction research, it has been noted that there are perceived gender differences in purchasing behavior (Smith, 1989). Current research has found that men participate in online auctions more than women (bDemographics: demographics influence online spending,Q). However, the number of women participating is gradually increasing from 35% to 45% (Crocket, 1999). 2.6.2. Reputation of participants and purchaser Reputation is defined as the boverall quality or character as seen or judged by people in generalQ (Merriam-Webster, 2000). Fombrun (1996) describes how companies (through the public relations function) generate breputational capitalQ by developing strong and consistent images, and thereby compete for prestige and achieve celebrity. Another definition of reputation is brecognition by other people of some characteristic or abilityQ (Johnston, 2002). Reputation is a subjective concept that resides in the mind of the buyer. The inability to observe and quantify reputation makes it difficult to estimate its effect on price. However, it is believed that if two sellers offer the same product, the seller with the bbetterQ reputation should receive a higher price, all other things being equal. Klein (1997) views reputation as a mechanism for rewarding good conduct and punishing misbehavior, and hence is the bglueQ that keeps our relationships together and society functioning. Major auction sites use a feedback system that enables buyers and sellers to read previous auction comments about an individual’s feedback score, such as comments from buyers and/or sellers and the reasons for positive or negative ratings. The feedback system offers a benefit to researchers in quantifying the reputation of the seller (and buyer) when an auction is successfully concluded (Johnston, 2002). According to Chen and Wilson (2002), the feedback ratings also help to improve buyers’ decision making and increase the odds of participation in a successful auction, as buyers prefer to deal with known suppliers to reduce risk. 2.6.3. Unique bidders The number of individual bidders in an online auction stimulates increased interest in the auction, in the same manner as traditional auctions (Smith, 1989).
3. Hypotheses By synthesizing the findings of the literature review, a number of key hypotheses are proposed. The first is that the level of the reserve price in an online auction will exert an influence over the final sale price obtained and the level of
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
interest in the auction. It is hoped that if the theory presented by Waehrer et al. (1988) applies to online auctions then a decrease in the reserve price will result in increased buyer interest and final sale price. The following hypotheses are reflected in the conceptual model presented in Fig. 1. Hence, it is hypothesized that H1. A low reserve price results in a higher final sale price and overall interest in an online auction, when compared to a high reserve price (relative to the price of the product). The second is that disclosure of the reserve price will exert an influence over the final sale price obtained and the level of interest in the auction. We believe that if the theory of Horstman and LaCasse (1997) proves to be relevant for online auctions, then the disclosure of the reserve will result in an increase in overall auction interest and final sale price. Hence, it is posited that H2. The disclosure of a reserve price results in a higher overall level of consumer interest in an online auction, and in turn a higher final sale price obtained in the auction. Finally, the initial bidding history of the auction will exert an influence over the final sale price obtained and the amount of interest in the auction. Smith (1989) believes that under the conditions of a traditional auction, a degree of social inertia is created by increased bidding, stimulating higher levels of interest in the auction, and accordingly increasing the final sale price obtained in the auction. Therefore, if the theory of social inertia applies to online auctions, we hypothesize that an increase in the initial bidding history of the auction will result in an increase in
1413
auction interest and the final sale price obtained. Consequently, it is proposed that H3. An increase in the initial bidding history of an online auction will result in a higher level of auction interest and a higher final sale price obtained in the auction.
4. Method 4.1. Subjects and design To collect data for the experiment, a total of 160 individual auctions were posted on an Internet auction site. Participants in the auctions were not informed of nor selected for the experiment and were therefore behaving as expected under normal auction conditions. The experiment consisted of a 2 (reserve price: $1 or $15)2 (reserve price: disclosed or not disclosed)2 (bidding level: high or low) factorial design, resulting in eight experimental conditions. Each auction was set up to match one of the eight experimental conditions; otherwise, all other controllable auction variables remained constant. Specifically, the data were collected by observing auctions being posted for a 3-day period (Monday 9 p.m. through Wednesday 9 p.m.). This was to ensure that website traffic was relatively constant for all experiments. The product category chosen for all auctions was disk-based media, such as DVD movies and Sony Playstation games. No picture of the product was offered. The product was sold using an established trading account. The text description of the product was constant,
Fig. 1. Conceptual model.
1414
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
apart from title differences. Bidder questions regarding the auction were not answered. The auctions were all posted in the same DVD or gaming section of the auction site. Direct bank deposit payment was used to complete all auction transactions and was the only accepted form of payment identified in the auction text. The selection of DVDs and Playstation games as a category is consistent with Massad and Tucker’s (2000) statement that predominant goods traded in online auctions are commodities or products categorized as bsearchQ goods. Furthermore, Smith (1989) believes that due to the rudimentary nature of commodity goods, their final pricing in auctions is only determined by natural market conditions, in contrast to the influence exerted by forces of social classification and manipulation when auctioning collectible goods. Therefore, we decided to use DVDs and Playstation games due to their relatively low cost and high consumer appeal as a commodity. The exact choice of titles for sale was determined by reviewing the top-selling DVD and Playstation game list from a local retailer to ensure that all titles had approximately the same level of consumer desirability in the local market. The DVD and game titles selected all had similar characteristics, that is, familyoriented movies to provide a consistent demographic appeal between experimental conditions. The exact DVD or game title selected for each auction was rotated for each of the 20 experiment rounds to ensure the desirability of the exact DVD or game title was not confounding the results.
a DVD or Playstation game, is expected to be half of full retail price, the expected sale price of the auctions was predicted to be approximately $25. Therefore, the reserve levels of $1 and $15 can be deemed to represent both a low and high reserve price, relative to the expected sale price of the product. Reserve disclosure: The reserve price of the auction was either disclosed or not disclosed in the text of the auction. If the reserve was disclosed, the auction text merely stated breserve price is $. . ..Q In the case of the reserve being nondisclosed, the auction text remained silent in respect to reserve. However, the auction participants would have been aware that the auction did have a reserve, as site rules will not allow an auction to commence without a specified reserve. Bidding history: The bidding history of the auction was contrived using pseudonym bidding accounts established by the researchers. The bidding history was set to either a high or low level for each auction, depending on the relevant experimental conditions. A low bidding history was simulated by placing no bids on the auction, while a high bidding history was simulated by placing 10 bids early in the auction. The selection of the high bidding level was based around the researchers’ previous experience with online auctions suggesting that most auctions for DVD and Playstation game type commodities not often exceeding 15 bids during the entire auction period. 4.4. Dependent measures and covariates
4.2. Data collection procedure The auctions were simultaneously posted on the auction site at 9 p.m. on Monday evening and were set for a 3-day time limit. If a high bidding level was required for the auction’s experimental condition, 10 bids were immediately placed on the auction by pseudonym bidders. If the reserve was to be disclosed, this was specified in the text of the auction. Finally, the reserve price was set during the initial posting of the auction through the auction site registration. The auctions were observed at 3-hour intervals with the amount of interest in the auction being recorded for potential patterns. In addition, due to the anonymous nature of the Internet, online auction participants were not aware that they were being observed. Therefore, aside from observation, no other action was instigated in respect to the auction process by the researchers. At the conclusion of the auction, the winning bidder was contacted by the researcher, bank account details were provided, the goods were then sent to the purchasers delivery address completing the transaction. All information in respect to the auctions was then downloaded from the auction site for analysis. 4.3. Independent variables Reserve price was set at either a $1 or $15 level. Given that the final sale price of a second-hand commodity, such as
The dependent variables were final sale price and auction interest level. The final sale price was considered to be the final bidding level of the auction prior to any additions for postage and auction site fees. The auction interest was measured using an Internet hit counter placed on each auction; this could only be viewed by the researchers, and multiple hits from individual members were allowed. In addition, four covariates were also recorded, namely, gender of the purchaser, number of unique bidders on the auction, profile of the final purchaser, and average profile of bidders on the auction. The profile of the final purchaser and the average profile of all bidders were determined using the positive feedback ratings associated with all individuals who transact on the auction site. These profiles are developed based around the amount of transactions completed successfully by an individual trader.
5. Results All continuous variables were examined statistically to meet assumptions of normal distributions. Two variables appeared to be normally distributed, while three showed skewness and kurtosis levels outside the acceptable range of 2bxb2. To rectify the problems of skewed and highly kurtotic data distributions, these variables were transformed
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
1415
using a reciprocal reflex procedure (using the 1 (1/x) formula). The procedure rectified the problem for two out of three variables (final sale price: skewness=3.45 to 0.17, kurtosis=13.09 to 0.65; bidding profile: skewness=7.70 to 0.38, kurtosis=80.8 to 0.60) but only did a slight improvement for the purchaser profile variable (skewness=4.89 to 3.66, kurtosis=34.24 to 14.29). After completing the procedure to satisfy the assumption of normality, ANCOVA tests were conducted on both of the dependent measures, final sale price and auction interest. 5.1. Effects on the final sale price Table 1 includes a number of significant main effects. First, with regards to sale price as a dependent variable, main effects of reserve price and reserve price disclosure were significant at ( Pb.001). Two out of three independent variables appear to have some effect on the final price obtained in the sale of a product in an online auction (see Fig. 2 for a visual representation). Both main effects demonstrated large and moderate effect sizes, respectively, with reserve price disclosure (e 2=.29) showing the largest effect followed by reserve price level (e 2=.13). However, bidding history did not demonstrate any significant effect on the final price obtained by the seller. The descriptive statistics for means and standard deviations are displayed in Table 2. Both main effects observed were significantly adjusted by the average bidder profile ( Pb.05). Although the effect here is small, an examination of the means reveals that the final price obtained is lower as the average experience level of bidders increases. Thus, the final sale price is highest only under a $1 reserve price, which is disclosed ( Pb.001). In addition, if a $1 reserve is not disclosed, then the influence of a $1 auction is significantly eroded. A possible explanation for this effect could be that when the reserve is disclosed, there is typically an increased legitimate interest in the auction leading to an increase in
Fig. 2. Estimated marginal means of final sale price.
final sale price. However, when the reserve is not disclosed, there is a need to create interest in the auction, which can only occur through increased bidder participation. Yet, there is some evidence presented by Smith (1989) suggesting that under some circumstances of increased participation, risk- and competition-averse bidders will withdraw from placing bids at higher reserve price levels. This could possibly be associated with individual bidder’s price expectations being exceeded under the conditions of a $15 reserve auction. In sum, it appears that the optimal strategy to obtain the highest sale price is to use a disclosed $1 reserve price. 5.2. Effects on the overall auction interest Reserve price and reserve price disclosure also register as main effects significant at Pb.001 with respect to overall auction interest, measured here by total hits (Table 1). This means that two of the three independent variables significantly influence the level of hits, and thus the overall level of interest in a particular online auction (see Fig. 3). They
Table 1 ANCOVA results for final sale price and total hits Source of variation
F value, df=1
Dependent variable
Final sale price
Main effects Reserve price value Reserve price disclosure Bidding history
21.8 61.2 2.5
Significance
Effect size (e 2)
F value, df=1
Significance
Effect size (e 2)
Total hits .000 .000 .113
.13 .29 .02
18.9 48.0 2.83
.000 .000 .092
.11 .25 .02
Interaction effects Reserve PriceReserve Disclosure Reserve PriceBidding History Reserve DisclosureBidding History Reserve PriceReserve DisclosureBidding History
1.2 0.04 0.04 0.04
.270 .787 .847 .855
.01 .00 .00 .00
0.32 0.24 0.01 0.34
.538 .627 .975 .559
.00 .00 .00 .00
Covariates Gender Average bidder profile No. of unique bidders Purchaser’s profile
0.94 4.47 0.03 0.01
.335 .036 .760 .839
.01 .03 .00 .00
0.23 1.21 2.99 0.04
.653 .271 .086 .846
.00 .01 .02 .00
1416
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
Table 2 Dependent variables: final sale price and total hits Treatments, means (S.D.) Reserve price
$1
Reserve disclosure
No
$15
Bidding history
Low
High
Low
High
Low
High
Low
High
Sample size
n=20
n=20
n=20
n=20
n=20
n=20
n=20
n=20
Final price paid ($) Total hits
23.9 (11.61) 120.1 (37.50)
24.8 (10.37) 126.5 (36.75)
30.8 (14.24) 161.2 (45.15)
32.3 (13.84) 177.6 (48.72)
20.0 (10.73) 72.6 (38.15)
20.9 (10.02) 93.1 (40.59)
26.5 (12.50) 131.1 (44.35)
28.2 (13.22) 142.9 (43.37)
Yes
also showed moderate to large effect sizes with reserve price disclosure (e 2=.25) having the largest effect followed closely by reserve price (e 2=.11). However, bidding history again failed to impact on the number of visits to the online auction. Overall, there appears to be a large degree of consistency with respect to the influence of the independent variables over the dependent measures. The results suggest that total hits are maximized when there is a $1 reserve price that is disclosed. Under conditions of the reserve price being disclosed, the seller will be able to obtain high levels of interest in the auction, and while it is preferable for the auction to have a $1 reserve as well, the disclosure of the reserve remains the critical and dominant element in stimulating auction interest. For total hits, none of the covariates showed any significant effect ( PN.05). To summarize the online auction data thus far, it appears that having a $1 reserve price, which is disclosed, is the optimal strategy to adopt when selling goods in an online auction, if the goal of the seller is to maximize sale price and auction interest. However, generating some bidding history early in the auction did not impact on the outcome of the auction. With respect to the covariate of gender, more research needs to be conducted to determine if this effect was primarily due to the nature of the commodity used for the experiments. Finally, with respect to the covariate of the average bidders’ profile, there is minimal to no control on this covariate by the seller, therefore, the practical implications of being aware of its influence are somewhat limited.
Fig. 3. Estimated marginal means of total hits.
No
Yes
6. Discussion H1 states that a low reserve price ($1) will result in a higher final sale price and overall interest in an online auction, when compared to a high reserve price ($15). As discussed in the Results section, there was strong support for this hypothesis with the final sale price and overall auction interest being consistently higher for $1 reserve auctions, regardless of reserve disclosure or bidding history, when compared to $15 reserve auctions. Hence, we find that the findings of Gupta and Bapna (2001) apply to online auctions, and that a low reserve has the effect of attracting a larger spectrum of potential bidders to the auction. It would suggest that Smith’s (1989) theory of social inertia increases bidder interest and results in increased competition, which in turn results in higher sale prices. This was, however, contrary to the findings of Haubl and Popkowski Leszcyc (2000), who determined that higher seller-specified reserve prices led to fewer bidders per auction, but higher selling prices. It also contrasts with a recent study by Ariely and Simonson (2003) where starting prices for auction of football game tickets lead to higher winning bids but only when comparable items were not immediately available in the same context. Gilkeson and Reynolds (2003) also found some conflicting results for starting price as a relative measure to final winning bid for collectible items. H2 states that the disclosure of a reserve price will result in a higher overall level of consumer interest in an online auction, and in turn a higher final sale price obtained in the auction. The findings support this hypothesis with strong main effects being observed for both dependent measures of final sale price and overall auction interest. Because there is no significant body of literature discussing the effects of reserve price disclosure in online auctions, we have made inferences based around some of the discussions on reserve price by Horstman and LaCasse (1997), who believed the disclosure of a reserve price reveals the seller’s estimation of the good’s value to potential bidders. Consequently, the disclosure of reserve price will result in attracting individuals interested in purchasing the goods at the reserve price indicated. In addition, Horstman and LaCasse, who developed their ideas based around traditional in-person auctions, believed that the disclosure of the reserve price would result in an increase in the final sale price of the goods. Therefore, in light of the results of this research, it can be concluded
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
that the conclusions of Horstman and LaCasse also apply in the realm of online auctions. H3 states that an increase in the initial bidding history of an online auction will result in a higher level of auction interest and a higher final sale price obtained in the auction. Overall, the results do not support this hypothesis. Therefore, H3 is not supported for both dependent variables, but further research may be required. It would appear that initial bidding does accelerate the auction process, but does not alter the outcome. 6.1. Managerial implications Theoretically, this research helps to bridge the gap between the expansive literature and theories relating to traditional in person auctions and the newly emergent technology of online auctions. Predominantly, this research has proven that while there are rudimentary physical differences between traditional and online auctions, fundamentally, the underlying consumer psychology of online auctions does not differ greatly from the consumer behavior paradigms of traditional in person auctions. Practically, this research has begun to develop a model of how auction agents, hosts, and sellers can set up auctions to maximize potential sale price and overall auction interest. Clearly, these findings have obvious marketing benefits for either individual resellers or the new breed of small business entrepreneurs emerging from this new peer-to-peer channel of distribution. Summed up in a few words, this initial phase of the research has concluded that an auction with a disclosed, $1 reserve price is most likely to result in a higher final sale price and level of auction interest. 6.2. Directions for future research A number of factors can impact the outcome of an auction and, more importantly, different combinations of these can contribute to varying outcomes. Primarily, there is a need to investigate the influence of other factors related to the online auction process such as the inclusion of pictures, the reputation of the seller through the testing of differing feedback ratings, the impact of reserve price across other currencies, the varying effects of short- and long-term auctions, and the correlation between gender and product interest. As a direct consequence of questions emerging from this research, more research is required to fully understand the influence of bidding history as it accelerates the bidding process. As stated in the Limitations section, person-to-person commodity auctions are typically governed by a distinct set of norms and rules; therefore, further research is required in the area of auctioning with respect to B2B materials procurement and person-to-person collectible auctions. A study of other auction sites in the world would provide useful information about online auction behavior across cultures, especially cultures such as the Dutch who follow a
1417
vastly differing auction methodology and style compared with the traditional English model. In-depth qualitative analysis of the reasoning underlying the decisions of auction participants and purchasers would provide information as to why certain behaviors were exhibited, and more importantly, if they were consciously affected by the experimental stimulus. In addition, this may provide some insight into the social impacts of online auctions, and in particular how online auctions are changing the relationships and expectations between buyers and sellers. Finally, some consideration needs to be given to the contrasting findings of this research and Haubl and Popkowski Leszcyc’s (2000) research, with each paper implying vastly different approaches to maximizing sale price when manipulating the reserve price. 6.3. Limitations Due to the nature of the product (DVDs and Playstation games) being auctioned, the results of the auction can only be generalized to commodity goods or search goods. The reputation of a seller has also been acknowledged to have a direct influence over the amount of interest in a conventional auction (Smith, 1989). However, whether this claim applies to online auctions has not been established. Therefore, the presence of seller reputation should be considered not only as a limitation to the generalizability of this research, but also as a potential area for future research through explaining the exact influence of seller reputation over the dependent variables of interest.
References Ariely D, Simonson I. Buying, bidding, playing, or competing? Value assessment and decision dynamics in online auctions. J Consum Psychol 2003;13(1–2):113 – 23. Bajari P, Hortacasu A. Winner’s curse, reserve prices and endogenous entry: empirical insights from eBay auctions. Working paper, Department of Economics, Stanford University; 2000. Chen H., Wilson D. Online auctions: are relationships doomed? 2002. Available at: http://www.bath.ac.uk/imp/pdf/xx _ ChenWilson.pdf. Retrieved October 6, 2002. Cox DF. The measurement of information value: a study in consumer decision making. In: Decker WS, editor. Emerging concepts in marketing. Chicago (IL)7 American Marketing Association; 1963. p. 413 – 21. Crocket RO. Going, going. . .richer. Bus Week 1999 December 13;(3659). Demographics: demographics influence online spending; 2002. Available at: http://www.cyberatlas.internet.com/big_picture/demographics/article/ 0,1323,5901_344751,00.html. Retrieved October 10, 2002. Dholakia UM, Basuroy S, Soltysinski K. Auction or agent (or both)? A study of moderators of the herding bias in digital auctions. Int J Res Mark 2002 June;19(2). Available at: http://www.business2.com/articles/ 1999/06/content/editors.html. Retrieved October 8, 2002. Engelbrecht-Wiggans R, Shubik M, Stark RM. Auctions, bidding, and contracting: uses and theory. New York7 New York Univ. Press; 1983. Fombrun CJ. Reputation: realizing value from the corporate image. Boston (MA)7 Harvard Business School Press; 1996. Forrester Research. Available at: http://www.forrester.com. Retrieved October 10, 2002.
1418
M.J.C. Walley, D.R. Fortin / Journal of Business Research 58 (2005) 1409–1418
Gilkeson J, Reynolds K. Determinants of Internet auction success and closing price: an exploratory study. Psychol Mark 2003;20(6):537 – 66. Gupta A, Bapna R. Online auctions: a closer look. Handbook of electronic commerce in business and society. Boca Raton (FL)7 CRC Press; 2001 Available at: http://misrc.umn.edu/workingpapers/fullpapers/2001/ 0120_050101.pdf. Retrieved October 10, 2002. Haubl G, Popkowski Leszcyc P. Going, going, gone—determinants of bidding behavior and selling prices in Internet auctions. Paper presented to Division 23, APA Conference, San Francisco, CA, 2000. Herschlag M, Zwick R. Internet auctions—popular and professional literature review. Q J Electron Commerce 2000;1(1). Horstman I, LaCasse C. Secret reserve prices in a bidding model with a resale option. Am Econ Rev 1997;87:663 – 84. Internetauctionlist. Available at: http://www.internetauctionlist.com. Retrieved October 10, 2002. Johns C, Zaichkowsky J. Bidding behavior at the auction. Psychol Mark 2003;20(4):303 – 22. Johnson NA, Friedman HH. Online auctions for business to business transactions. Washington (DC)7 National Public Accountant; June 2002. Johnston T. Reputation price premiums in online auctions; 2002. Available at: http://www.sbaer.uca.edu/Research/2001/SMA/01sma178.html. Retrieved October 9, 2002. Kagel JH, Roth AE. The handbook of experimental economics. USA: Princeton Univ. Press. Kauffman R, Walden E. Economics and electronic commerce: survey and research directions. Int J Electron Commerce 2001;17. Klein DB. Reputation: studies in the voluntary elicitation of good conduct (economics, cognition, and society). Ann Arbor (MI)7 University of Michigan Press; 1997. Lucking-Reiley D. Using field experiments to test equivalence between auction formats: magic on the Internet. American Economic Review 1999 December;89(5):1063 – 80. Massad VJ, Tucker JM. Comparing bidding and pricing between in-person and online auctions. J Prod Brand Manage 2000;9(5):325 – 40. McAfee RP, McMillan J. Auctions and bidding. J Econ Lit 1987;25: 699 – 738.
Merriam-Webster collegiate dictionary. Springfield (MA)7 MerriamWebster; 2000. Milgrom P. Auctions and bidding: a primer. J Econ Perspect 1989;3:3 – 22. Myerson RB. Optimal auction design. Math Oper Res 1981;6:58 – 73. Rothkopf MH, Harstad RM. Modeling competitive bidding: a critical essay. Manage Sci 1994;40(3):364 – 84. Rumpe B. E-Business Experiences with On-line Auctions; 2003. Available at: http://www4.in.tum.de/~rumpe/ps/IRMARumpe.pdf. (2004, August 11). Smith CW. Auctions: the social construction of value. Berkeley (CA)7 University of California Press; 1989. Varadarjan PR, Yadav MS. Marketing strategy and the Internet: An organizing framework. J Acad of Mark Sci 2002;30(4):296 – 312. Vickrey W. Counter-speculation, auctions, and competitive sealed tenders. J Finance 1961;41:8 – 37. Vincent D. What reserve prices may be kept secret. J Econ Theory 1995; 65(2):575 – 84. Waehrer K, Harstad R, Rothkopf M. Auction form preferences of riskaverse bid takers. Rand J Econ 1988;29(1):179 – 92.
For Further Reading Amazon.com. Available at: http://www.amazon.com. Retrieved October 13, 2002. Bromley DB. Reputation, image, and impression management. West Sussex (England)7 Wiley; 1993. Cox DF. Risk taking information handling in consumer behavior. Boston (MA)7 Harvard Univ. Press; 1967. Daly J. Let’s make a deal. Business 2.0, June 1999. Patsuris P. One-to-one auction; 1999 Available at: http://www.forbes.com/ tool/html/98/jan/0116/side1.htm. Retrieved October 08, 2002. Weiss AM, Anderson E, MacInnis DJ. Reputation management as a motivation for sales structure decisions. J Mark 1999;63(4):74 – 89.
Journal of Business Research 58 (2005) 1419 – 1425
The role of affective expectations in memory for a service encounter Elizabeth Cowleya,*, Colin Farrellb, Michael Edwardsonb a
Discipline of Marketing, University of Sydney, Sydney, NSW 2006, Australia School of Marketing, University of New South Wales, Sydney NSW 2052, Australia
b
Received 1 December 2002; received in revised form 1 September 2003; accepted 1 October 2003
Abstract Affective expectations influence affective reactions at the time of an experience and when the individual reflects back on the episode. The study reported here investigates whether a consumer’s uncertainty orientation explains how they use affective expectations. The results reveal that people with an internal locus of control continue to use their expectations after the event. Luck-oriented dinternalsT selectively remember the facts to be expectation-consistent. Chance-oriented dinternalsT reinterpret the facts to be more expectation-consistent. Affective expectations do not influence affective reactions after they are formed for consumers with an external locus of control. In fact, affective expectations have very little impact on the affective reactions of luck-oriented dexternalsT, even when they are initially formed. Implications for service settings are discussed. D 2004 Elsevier Inc. All rights reserved. Keywords: Consumer behavior; Affective expectations; Memory; Uncertainty; Luck
1. Introduction Rob and Janet recently went to see a play; they both had great expectations for a pleasant evening out. They had choice seats in a wonderfully restored theatre. As the scenes progressed, Janet felt the disappointment set in. The acting and scripting were not as she had hoped. She thought it would be such a brilliant experience. Janet left having enjoyed certain aspects of the play, but overall not really feeling as satisfied as she had expected. Their discussion after was tempered, they both wanted to have enjoyed the play more. Over the next few days, Janet heard Rob recount the events of the evening. Each time the story was told it was more positive, the details were correct, but he now found humor in what had been disappointing. Janet began to wonder if his expectations had colored the way he was reflecting on the experience or whether her memory for the details of the evening was failing her. The affective expectation model (AEM: Wilson et al., 1989) asserts that how much a person thinks they will like * Corresponding author. Tel.: +61 2 9384 3384; fax: +61 2 9663 1985. E-mail address: [email protected] (E. Cowley). 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.10.011
an experience (affective expectation) is as important as what actually happens during the experience in the determination of how much the experience is enjoyed (affective reaction). The role of affective expectations (AEs) does not always end with the affective reaction (AR) during the experience. AEs may be used to reinterpret or selectively remember the experience (Klaaren et al., 1994). We propose that an individual’s approach to uncertain situations may determine how AEs influence his or her AR to the event and reflection on the experience. This study investigates the role of AEs in forming initial ARs, how the ARs may change over time, and the moderating effect of an individual’s uncertainty orientation.
2. Affective expectations Although AEs influence the evaluation of an experience (Geers and Lassiter, 1995, 2002; Hodges et al., 2000; Klaaren et al., 1994; Wilson et al., 1989), it is unclear whether AEs continue to play in an individual’s willingness to repeat an activity after an AR has been experienced (Klaaren et al., 1994). The initial effects hypothesis purports
1420
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
that AEs exert their influence only at the time that people experience an activity, by directing attention toward expectation-consistent information. If a consumer experiences a product or service that is inconsistent with their expectations, they may alter their AR to be more expectationconsistent. According to the initial effects hypothesis, the direct role of AEs ends with the initial AR, no additional effect occurs as time passes. For instance, a consumer’s expectation of enjoying a play will affect their perceptions of the experience, their evaluation of the experience, and therefore indirectly, their willingness to revisit the theatre or tell others to attend the play. However, the evaluation, once formed, will not be further affected by the initial belief that the experience would be pleasant. A second hypothesis posits that AEs continue to influence evaluations of the consumption experience. The reinterpretation hypothesis suggests that people continue to reinterpret aspects of the experience that were inconsistent with expectations. The event becomes more expectationconsistent by altering the meaning of features of the experience, the poor acting of a character becomes a humorous parody, or by adjusting the importance of a feature of the experience, the poor acting played only a small role in the overall experience. Memory of the features does not change, but the interpretation of the features does change. Hodges et al. (2000) and Klaaren et al. (1994) found results supporting this hypothesis. The third hypothesis also proposes that the role of AEs does not end with the initial AR. This proposition includes a change in memory of the features of the experienced episode. According to the selective memory hypothesis expectationinconsistent aspects of the experience are less accessible than expectation-consistent aspects of the experience. Since expectation-consistent memories are more accessible, ARs become more expectation-consistent over time. In this case, the poorly acted scenes become less accessible and therefore less influential on ARs, but the exquisite music and avantgarde sets are accessible, resulting in a more positive AR. In other words, the interpretation of a particular feature does not change, but the ability to remember a particular feature does change. An implicit assumption in the last two hypotheses is that when individuals are responding to the question, bHow much did you enjoy that experience?Q they reconstruct the experience and form an evaluation, as opposed to retrieving the AR stored during the experience. Each of the three explanations is plausible. We assert that the uncertainty orientation of the consumer may determine how, and to what extent, AEs influence ARs. Four uncertainty orientations are described in the next section.
3. Uncertainty orientation Consumption experiences always involve a degree of uncertainty because the outcome is at least partially dependent on the somewhat unpredictable behavior of
others. Not all individuals perceive and react in a similar manner to uncertain situations (Breen and Zuckerman, 1999; Zuckerman and Kuhlman, 2000). We propose an uncertainty orientation framework that uses an individual’s locus of control to further divide Friedland’s (1998) luck versus chance orientation, a tendency to attribute outcomes to either luck or chance. What is the difference between luck and chance? bThe notion of causelessness is so alien to us in the absence of a known because we tend to attribute events to imaginary causes like luck and chanceQ (Wagenaar and Keren, 1988). Luck is a causal category used to explain successes and failures that cannot be attributed to ability and effort, or task difficulty (Weiner et al., 1971). Chance distributes events fairly and evenly, producing all possible outcomes with equal frequencies in the short and long term (Wagenaar, 1989; Wagenaar and Keren, 1988; Keren, 1994; Friedland, 1998). Moreover, chance outcomes are perceived as selfcorrecting, whereby a deviation in one direction causes an equal deviation in the opposite direction to restore equilibrium (Tversky and Kahneman, 1971; Kahneman and Tversky, 1972). Therefore, people misunderstand chance outcomes believing the process displays a memory and sense of justice (Fischhoff, 1982), leading to the belief that outcomes are dependent on one another. 3.1. Luck orientation An individual’s orientation toward luck versus chance determines the use of probabilistic outcomes in forming expectations, particularly under uncertain circumstances (Friedland, 1992, 1998; Keren and Wagenaar, 1985, 1987; Wagenaar, 1989; Wagenaar and Keren, 1988). Luckoriented people pay little attention to probabilities that define the decision problem. They expect carryover from one random or independent event to another. However, all luck-oriented people are not alike; they can have either an internal or external locus of control. Some luck-oriented people, luck internals, believe that luck is a stable internal force that will influence events in their favor (Darke and Freedman, 1997). For example, some people believe they have an advantage betting at the track or playing poker machines because they are lucky. Other luck-oriented people, luck externals, believe that luck comes and goes, but it can be detected (Friedland, 1998; Wagenaar, 1989). For example, if something goes well in the morning, it is a sign of a lucky day, if something goes wrong they wait for two other negative outcomes because bad luck strikes in threes. 3.1.1. Luck internals Luck internals feel personally lucky, and as a result they have more positive expectations for the outcome of events (Darke and Freedman, 1997). This group perceives a strong outcome–person correspondence because they consider an outcome to be dependent on the involved person’s luck
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
(Friedland, 1998). This group has a high desire for control and believes that they can control events. Since luck internals have a personal stake in the consistency of expectation and outcome, they are particularly likely to assimilate their AEs into their AR. Over time, luck internals selectively remember positive outcomes in general, but particularly when they are consistent with their initial expectations because an ego defensive motivation directs them to convince themselves they understood their luck. 3.1.2. Luck externals Luck externals believe that when an unusually high frequency of similar outcomes occurs, luck caused the sequence (Wagenaar, 1989; Teigen, 1994; Friedland, 1992; 1998). They think that good and bad luck lasts for certain periods of time, and although they cannot control luck, they can detect the beginning or end of a lucky streak with signs in the environment (Friedland, 1998). Consequently, making the prediction of the outcome in uncertain settings is a task of detecting the presence of good or bad luck (Wagenaar, 1989). Luck externals are least susceptible to the hindsight bias (Farrell et al., 2002), as they are more focused on outcomes and whether the outcome can be used in the detection of the beginning or end of a lucky streak, with an eye to predicting future events. This group is least likely to use their AEs to reinterpret or selectively remember an experience. 3.2. Chance orientation Chance-oriented people see decision making in uncertain settings as a task of guessing what the outcome will be by looking at the decision problem and using information that might assist them in understanding the patterns in chance (Friedland, 1998). Chance-oriented people are likely to believe that the outcome will be the same regardless of the person involved. These people also expect that chance will distribute events evenly such that all possible outcomes will occur with equal frequency (Wagenaar, 1989; Wagenaar and Keren, 1988; Keren, 1994; Friedland, 1998). As is the case with luck-oriented people, not all chance-oriented people are the same. Some chance-oriented people, chance internals, believe that there is a pattern to uncertain outcomes: stability in external probabilistic events. Consequently, they can improve the likelihood of guessing correctly by dseeingT the pattern. Other chance-oriented people, chance externals, hold the more rational belief that there is no reliable process for predicting the outcome of uncertain events (Darke and Freedman, 1997). 3.2.1. Chance internals Chance internals believe they can gain some control by looking for information that will improve their ability to predict random events. They try to understand the environment in which they live and to gain confidence in their understanding using a kind of bcreeping determinismQ by
1421
Table 1 Summary of the role of affective expectations Internal locus of control
External locus of control
Luck orientation
Chance orientation
Selective memory 1) AEs assimilated into ARs. 2) Good memory for AE-consistent features. 3) AR moves toward AE over time. No effect 1) AEs not assimilated into ARs. 2) Good memory for the features. 3) No change to AR over time.
Reinterpretation 1) AEs assimilated into ARs. 2) Reinterpret events to be AE-consistent. 3) AR moves toward AE over time. Initial effects 1) AEs assimilated into ARs. 2) Good memory for the features. 3) No change to AR over time.
updating their knowledge of a phenomenon with observations of perceived cause and effect. Chance internals are motivated to reinterpret events to be expectation-consistent such that their skill at predicting outcomes is not challenged. 3.2.2. Chance externals Chance externals see little difference between chance and luck because they are most likely to understand the properties of randomness. Although their expectations may affect their perception and evaluation of a consumption experience, these rational externally oriented individuals are not motivated to reinterpret the events that occurred during the experience or to selectively retrieve expectation-consistent aspects of the episode. We hypothesize that the initial effects explanation best describes the behavior of the chance external. In summary, internally focused people are motivated to change the outcome of an experience either by reinterpreting the facts, or selectively remembering them. Luck internals selectively remember events because the outcome is attributable to their own luck status. Chance internals reinterpret events to defend their belief that they skilled at predicting uncertain events. Externally focused people do not believe they have control over events. Luck externals believe outcomes can be predicted by the detection of lucky streaks. Since the outcome does not reflect personally on them, AEs will have little effect during or after the experience. Chance externals view the outcome in an uncertain setting as not predictable. They may be motivated to pay more attention to expectation-consistent information because they hope their AEs are met, but that is where the role of AEs ends. The expected results are summarized in Table 1.
4. Empirical work 4.1. Design A 22 between-subjects design was used. The factors were expectations (positive, negative) and experienced
1422
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
outcome (positive, negative). The participants were divided into uncertainty orientation on the basis of their locus of causality scale (internal locus of control and external locus of control) and their responses to four luck/chance orientation scenarios adapted from Friedland (1998). 4.2. Procedure One hundred fifty-two undergraduate students received course credit for participation in the study. Participants were told they would be reading a story about going to a restaurant. They were asked to use their imaginations to picture the events as they were described in the story. Two paragraphs introduced the participant to the restaurant experience as an evening out with a friend. The episode began with them choosing a restaurant. They stepped into one restaurant to look at the menu. They were given an initial impression of the restaurant, the food, the service, and the way the evening might unfold (positive or negative). Thirty-five postgraduate marketing students participated in a pretest of the restaurant stories. They were asked to list events that might occur in a restaurant that were either positive or negative. Their responses were used in the development of the stories. The likelihood that good service, good food, and an enjoyable evening would transpire was used as initial expectations. Participants responded on three 120-mm continuous scales anchored with bnot very likelyQ and bvery likely.Q They were also asked for the likelihood that they would stay at the restaurant. The four variables were combined for the expectations measure with an alpha of .82. Three paragraphs followed providing the participant with further expectation information concerning the evening, their interaction with the waiter, the conversation with their friend, and their impression of the menu options (all were consistent with initial expectations). Expectations for service, food, and overall enjoyment were collected again. Since participants had not yet committed to staying at the restaurant, they were again asked about the likelihood that they would stay. The rest of the evening was described to them either as a positive or negative experience. Participants were asked three times to provide their evaluations of the events. After the description of the outcome, they were asked whether they would be likely to return to the restaurant and whether they would recommend the restaurant on 120-mm scales anchored with bnot very likelyQ and bvery likely.Q Finally, they were asked to provide their overall evaluation of the experience on a 120-mm continuous scale anchored with bnot very goodQ and bvery goodQ. The three scales were combined for the ARs measure with an alpha of .95. Participants then filled out an augmented locus of causality questionnaire (Valecha, 1972) based on Rotter’s (1966) original scale. Participants then completed a 15-min filler task. The scale is comprised of 29 pairs of statements
Exhibit 1.
including six filler items. In each pair of statements, one relates to an internal locus of control and the other an external locus of control. In its original format, participants were asked to indicate which one of the two statements best described their thoughts and behaviours. The augmented scale requires that participants state if the choice is much closer or slightly closer to their actual opinion. The first three questions following the filler task asked about their enjoyment of the evening in the restaurant, these responses will be used to test whether their ARs had changed over time. The first two questions asked the participants to rate the food and service in the restaurant on a 120-mm scale anchored with bnot very goodQ and bvery good.Q The third question asked them to indicate their experience in the restaurant on a 120-mm scale anchored with bnot very goodQ and bvery good.Q These questions were carefully worded to ask for their current evaluation, not their memory for a previous evaluation. The variables were combined for a measure of AR over time with an alpha of .94. Then participants indicated their recognition judgements and their confidence in their response for 17 statements. Seven of the 17 statements were taken from the version of the story they read. Seven of the 17 statements were taken from the other version of the story, in other words, exactly the opposite of what they read. Three of the questions could only be recognized if the participant had reinterpreted the facts. For instance, one statement mentioned how many waiters were working which was in the story, but also stated that the restaurant was either busy or not busy. This information was not provided in the story. Finally, participants completed the chance/luck scenarios which were adapted from Friedland (1992, 1998) to determine their tendency to attribute outcomes to either luck or chance. Participants indicated whether the outcome each of four scenarios was a result of luck or chance by dividing 100 points between the factors (see Exhibit for an example.
5. Results The scores on luck orientation and locus of control were approximately normal. As expected, the correlation between the measure for luck orientation and locus of control was not
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
significant (r=.08, p=.33). Where possible, regressions were run with the scale scores, however, for the majority of the analysis, a median split on both scales was used to divide the sample into four groups. The two median splits resulted in approximately equal cell sizes (n=37 for luck internals, n=41 for luck externals, n=38 for chance internals, n=36 for chance externals). The responses to the questions asking about the expectations for service, food, and overall enjoyment were averaged for each participant as a manipulation check. As expected, the average rating was more positive for the participants in the positive expectations condition (M pos=51.15 mm, M neg=29.22 mm, t=9.24, pb0.0001). 5.1. ARs immediately after the episode Assimilation of expectations into the ARs is a necessary, but not sufficient, condition for the initial effects, reinterpretation, and selective memory hypotheses. It was hypothesized that three of the uncertainty orientations would assimilate their AEs into their ARs to the episode, and that only luck externals would not use expectations in their ARs. Separate regressions were run for each of the uncertainty orientation groups on the overall evaluation with expectations and outcome included as factors. The regressions reveal that expectations are a significant determinant of ARs of the luck internals (t=2.15, pb.05), the chance externals (t=2.68, pb0.01), and the chance internals (t=2.33, pb0.05). Expectations did not influence the ARs of luck externals (t=1.18, p=.25). This result for the luck externals is consistent with the no effects hypothesis. The regressions also reveal a significant effect for outcome treatment for all groups: luck internal (t=13.17, pb.0001), luck external (t=10.71, pb.0001), chance internal (t=10.65, pb.0001), chance external (t=10.15, pb.0001). 5.2. Changes in ARs over time The initial effects hypothesis suggests that the role of expectations end at the time of the evaluation of the experience, therefore ARs should not change to be more expectation-consistent. Both the reinterpretation and selective memory hypotheses posit that ARs will change to be more expectation-consistent. To measure whether the AR changed over time, participants were asked to rate how much they had enjoyed the experience with three questions after the filler tasks. These three variables were averaged for each participant. The initial reaction was subtracted from the recent reaction to create a difference variable. A positive number reflects a more positive reaction over time; a negative number reflects a less positive reaction over time. As expected, regardless of initial expectations, AR did not change over time for either chance externals [(M pos=+0.40, n.s.), (M neg= 3.98, n.s.)] or luck externals [(M pos= 4.59, n.s.), (M neg= 4.73, n.s.)]. The pattern of results for chance externals was
1423
consistent with the initial effects hypothesis. The pattern of results for luck externals was consistent with the no effects hypothesis. As expected, ARs of chance internals did shift in the direction of the initial expectations [(M pos=+4.94, t=2.01, pb.05), (M neg= 7.10, t=2.60, pb0.01)]. As expected, the luck internals also altered their reactions over time, however, we hypothesized that they would alter their reactions to be more consistent with expectations. Instead, luck internals altered their AR to be more positive in both conditions [(M pos=+7.02, t=2.43, pb0.05), (M neg=+8.17, t=2.14, pb.05)]. Although it is counter to our expectations, it is consistent with our assertion that they believe the outcome depends on their luckiness. These people do not want to be unlucky. 5.3. Memory of the episode The selective memory hypothesis asserts that individuals forget those aspects of the experience that are inconsistent with their expectations. In two of the four conditions, expectations were manipulated to be inconsistent with the outcome, only those were included in this analysis. Ten recognition questions included statements about the experience: six were included in the story (three expectation-consistent, three expectant-inconsistent), four were not included in the story (two expectation-consistent, two expectation-inconsistent). The coding of their responses was such that a higher score indicates that the participant is not recognizing expectation-inconsistent information and is incorrectly remembering expectationconsistent information. An ANOVA run on the selective memory score with locus of control and luck groups as the factors reveals a significant effect for group [ F(3,66)=6.13, pb.001]. The luck internal group had the highest score on this test (M luck int=6.55, M luck ext=4.85, M chance int=4.11, M chance ext=4.21) which is significantly greater than chance (t=2.37, pb0.05) and is consistent with the selective memory hypothesis. 5.4. Reinterpreting the event The reinterpretation hypothesis suggests that individuals remember the events correctly, but reinterpret the events such that they are more consistent with their expectations. Three of the recognition questions tested this effect. In each of the questions the stated fact was correct, but a phrase was added that was outcome-inconsistent, but expectation-consistent. For instance, individuals in the positive expectation condition read that bthe music was very loud,Q but the recognition statement read bthe music was very loud, but lively.Q If they incorrectly recognized the statement, they were awarded a point. A higher score reveals a greater propensity to reinterpret events to be expectation-consistent. An ANOVA run on the reinterpretation score with luck orientation and locus of control
1424
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
as the independent factors revealed a significant interaction [ F(1,147)=14.73, pb.0001]. Chance internals and luck externals were most likely to reinterpret the experience (M luck int=0.52, M luck ext=0.78, M chance int=0.86, M chance ext=0.43).
6. Discussion The results indicate that luck internals assimilated expectations into the evaluation of an experience, the likelihood of returning and recommending the restaurant to friends. This group changed their ARs over time, but not to be expectation-consistent as hypothesized, instead they altered their reaction to the experience to be more positive as time passed. Correctly predicting the outcome was less important than defending the belief of having the good luck trait. This group selectively remembered expectation-consistent events occurring. The data support the selective memory hypothesis. This is important because the hypothesis has been tabled before, but sufficient evidence was not found to confirm the proposition (Klaaren et al., 1994). Further research is required to test for a relationship between the luck internal orientation and optimism, as optimism has been associated with a similar type of information processing distortion (Taylor and Brown, 1988). Chance internals altered their ARs to be more consistent with their initial expectations. This was expected because it allows them to assure themselves that they are skilled at using the signs to predict the pattern of chance outcomes. Further research should test whether the role of expectations on reactions becomes even greater if the consumer does not commit to an evaluation immediately after the consumption experience. Chance internals tended to reinterpret the events to be expectation-consistent. As chance internals are looking for information to reduce uncertainty and to try and predict outcomes in uncertain settings, it would be interesting to investigate what information they believe is diagnostic while forming expectations. Chance externals assimilated expectations into the evaluation of an experience and their intention to recommend the restaurant to friends, but this appears to be where the role of expectations ends. There was no evidence of selective memory or reinterpretation of the event. In fact, chance externals had the most accurate memory for the service encounter and memory for their evaluation of it. The data support the initial effects explanation. In this case, measuring a consumer’s AR at the end of an experience should predict the consumer’s future behavior as the evaluation appears to be stable over time. Finally, the luck externals group did not assimilate their expectations into their evaluation. They had accurate memory for the events involved in the experience. Unexpectedly, they did show some evidence of reinterpretation. Since this group is concerned with predicting when a good or bad streak of luck will begin or end, they may have
reinterpreted the events to be consistent with their initial detection of luck. Overall, internally focused individuals assimilated their expectations into the evaluation of the experience. In this case, contrary to conventional wisdom, some overpromising may not be a detrimental marketing strategy. For the luck internals, marketing communications, first impressions and WOM used to form expectations influences postpurchase behavior as well, and therefore are extremely important to the success of a service with these consumers. Previous research into the advertising for services, rather than goods, has found that a focus is often placed on factual informational content (Grove et al., 1995; Mortimer, 2000). AE assimilation may require marketing communications to be more specific in detailing the emotional benefits of the service rather than, or in additional to, factual content. This will provide more specific AEs that can be assimilated if performance fails to deliver. The effects of emotional labor for service staff and the possible dissonance between felt emotions and displayed emotions has been extensively researched (Hochschild, 2003). However, if a service worker has positive AEs of the service experience, expectation assimilation may moderate any emotional dissonance. Staff briefing sessions (Gamble and Kelliher, 1999) could therefore be used to reinforce the positive emotions that the service staff will feel as a result of their service provision. Internally focused people either selectively remembered facts or reinterpreted them, possibly to defend their beliefs about self. They also altered their ARs over time, either to be expectation-consistent if they are chance-oriented, or to reflect more positively on self if they are luck-oriented. In both cases, collecting ARs immediately after a service encounter would not provide an accurate indication of the AR used in the future decisions to return. This has implications for the timing of a satisfaction measurement because immediate evaluations may not be predictive of repurchase behavior. Externally oriented people do not have the same personal investment in the consistency between their expectations and the outcome. ARs were more stable over time for externally oriented consumers and therefore may be a more useful tool in predicting future behavior. In the study reported here, participants read a scenario describing an experience, they did not actually live through the episode. Further research is necessary to investigate whether people use their thoughts, feelings, and memories differently for real, as opposed to imagined, experiences. Also, participants’ ARs were measured with scales, future research should allow the respondents to provide information about why they are expecting certain things, what information they are using to interpret the events described, and why they are changing their expectations and reactions. Limitations aside, this research introduces an uncertainty orientation framework that appears to be useful in understanding why consumers react differently in uncertain settings.
E. Cowley et al. / Journal of Business Research 58 (2005) 1419–1425
References Breen RB, Zuckerman M. dChasingT in gambling behavior: personality and cognitive determinants. Pers Individ Differ 1999;27(6): 1097 – 111. Darke PR, Freedman JL. Lucky events and beliefs in luck: paradoxical effects on confidence and risk-taking. Pers Soc Psychol Bull 1997;23(4);378 – 89. Farrell C, Cowley E, Edwardson M. A new classification of uncertainty orientation: exploring the susceptibility to the hindsight bias in a gambling context. Working Paper Series, ISBN 1324-68IX. University of New South Wales, Sydney, Australia; 2002. Fischhoff B. For those condemned to study the past: heuristics and biases in hindsight. In: Kahneman D, Slovic P, Tversky A, editors. Judgment under uncertainty: heuristics and biases. Cambridge (UK)7 Cambridge University Press; 1982. p. 335 – 54. Friedland N. On luck and chance: need for control as a mediator of the attribution of events to luck. J Bus Res 1992;5:267 – 82. Friedland N. Games of luck and games of chance: the effect of luck-versuschance-orientation on gambling decisions. J Behav Decis Mak 1998;11:161 – 79. Gamble PR, Kelliher CE. Imparting information and influencing behavior: an examination of staff briefing sessions. J Bus Commun 1999; 36(3):261 – 79. Geers AL, Lassiter GD. Affective expectations and information gain: evidence for assimilation and contrast effects in affective experience. J Exp Soc Psychol 1995;35(4):394 – 413. Geers AL, Lassiter GD. Effects of affective expectations on affective experience: the moderating role of optimism–pessimism. Pers Soc Psychol 2002;28(8):1026 – 39. Grove SJ, Pickett GM, Laband DN. An empirical examination of factual information content amongst service advertisements. Serv Ind J 1995; 15:216 – 33. Hochschild AR. The managed heart: commercialisation of human feeling. Berkeley (CA)7 University of California Press; 2003. Hodges SD, Klaaren KJ, Wheatley T. Talking about safe sex: the role of expectations and experience. J Appl Psychol 2000;30(2):330 – 49. Kahneman D, Tversky A. Subjective probability: a judgment of representativeness. Cogn Psychol 1972;3:430 – 54.
1425
Keren G. The rationality of gambling: gamblers’ conceptions of probability, chance and luck. In: Wright G, Ayton P, editors. Subjective probability. New York7 John Wiley; 1994. p. 485 – 99. Keren G, Wagenaar WA. On the psychology of playing blackjack: normative and descriptive considerations with implications for decision theory. J Exp Psychol Gen 1985;114:133 – 58. Keren G, Wagenaar WA. Temporal aspects of probabilistic predictions. Bull Psychon Soc 1987;25:61 – 4. Klaaren KJ, Hodges SD, Wilson TD. The role of affective expectations in subjective experience and decision-making. Soc Cognit 1994;12(2): 77 – 101. Mortimer K. Are services advertised differently? An analysis of the relationship between product and services types and the informational content of advertisements. J Commun 2000;6:121 – 34. Rotter JB. Generalised expectancies for internal versus external control of reinforcement. Psychol Mon Gen Appl 1966;80(1):1 – 28. Taylor SE, Brown JD. Illusions and well-being: a social psychological perspective on mental health. Psychol Bull 1988;103(2):193 – 210. Teigen KH. Variants of subjective probabilities: concepts, norms and biases. In: Wright G, Ayton P, editors. Subjective probability. New York7 John Wiley; 1994. p. 211 – 38. Tversky A, Kahneman D. Belief in the law of small numbers. Psychol Bull 1971;2:105 – 10. Valecha GK. Construct validation of internal–external locus of control as measured as an abbreviated 11-item scale. Unpublished Doctoral Dissertation, The Ohio State University; 1972. Wagenaar WA. Paradoxes of gambling behavior. London7 Lawrence Erlbaum and Associates; 1989. Wagenaar WA, Keren GB. Chance and luck are not the same. J Behav Decis Mak 1988;1:65 – 75. Weiner B, Frieze I, Kukla A, Reed L, Rest S, Rosenbaum RM. Perceiving the causes of success and failure. In: Jones EE, Kanouse DE, Kelley RE, Nisbett RE, Valins S, Weiner B, editors. Attribution: perceiving the causes of behavior. Morristown (NJ)7 General Learning Press; 1987. p. 95 – 120. Wilson TD, Lisle DJ, Kraft D, Wetzel CG. Preferences as expectationdriven inferences: effects of affective expectations on affective experience. J Pers Soc Psychol 1989;56(4):519 – 30. Zuckerman M, Kuhlman MD. Personality and risk-taking: common biosocial factors. J Pers 2000;68(6):999 – 1029.
Journal of Business Research 58 (2005) 1426 – 1436
Distinguishing between feelings and emotions in understanding communication effects Flemming Hansen* Department of Marketing, Center for Marketing Communication, Copenhagen Business School, Solbjerg Plads 3, C.3, DK-2000 Frederiksberg, Denmark Received 1 December 2002; received in revised form 1 October 2003; accepted 1 October 2003
Abstract This article explores measures that may be particularly relevant in connection with peripheral or low-involvement information processing. For FMCG, peripheral information processing is dominant, but in terms of most measures used in communication research, central information processing is more efficient. The only exception relates to emotional responses. The more positive and strong effects are registered following peripheral information processing. This directs the attention towards contemporary neurophysiological research into memory and emotional processing. In addition to measuring emotional processes in terms of behavioural, glandular and autonomous responses, it is possible to infer something about underlying emotional dispositions from analyses of questions about feelings. D 2004 Elsevier Inc. All rights reserved. Keywords: Emotions; Low-involvement information processing; Peripheral information processing; Communication effects
1. Introduction In much research concerned with advertising effect, it being pretesting, posttesting or tracking, cognitive modelling of consumer behaviour has dominated the choice of measures and the models proposed. Awareness preceding interest, information search and evaluation has been assumed to be a fundamental sequence in the consumer’s information processing. Attention leading to brand perception, preferences, purchase intention and eventually buying behaviour are the elements in this assumed chain of events. At the same time, creative advertisers have presented campaigns relying upon irony, emotions, postmodern perception of human values, surprise and provocation. Not only have such campaigns frequently won creative awards, but in many instances, their sales effect has also been well documented. In parallel with presentations of the traditional, primarily cognitive-based model of thinking’s influence on the way in
* Tel.: +45 3815 2134; fax: +45 3815 2101. E-mail address: [email protected]. 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.10.012
which advertising works as communication, occasional findings have been presented, suggesting very different aspects of the communication process. Some, mostly theoretically oriented consumer behaviour researchers, have given thought to alternative ways of looking at the advertising communication process. Low involvement (Krugman, 1968; Zaichkowsky, 1985) and attitudes towards the ad (A-ad; Mitchell and Olson, 1981; MacKenzie and Lutz, 1982; Lutz, 1985) have been fundamental concepts, and a distinction between central and peripheral information processing has emphasised the multifaceted nature of advertising information processing (Petty et al., 1983). In more recent years, neuropsychologists, brain researchers and other behavioural scientists have strongly emphasised the importance of emotional response (Damasio, 2000; Le Doux, 1998). In this research, a distinction has emerged between feelings and emotions. Basically, emotions are thought of as very primitive, extremely fast, unconscious mechanisms controlling the individual responses to a wide variety of situations ranging from serious threats (for instance, from an approaching car) to trivial decisionmaking tasks (for instance, choosing a coffee brand in the supermarket; Heath, 2001; Franzen and Bouwman, 2001).
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
Feelings, on the contrary, are those conscious and cognitive perceptions we use to describe our more primitive noncognitive emotional control of what we do. We may talk about feelings of sadness, jealousy, happiness, etc. Such feelings are much more detailed in nature than emotions and they can be described verbally in more or less precise terms by the individual experiencing such feelings. In the paper, we shall review some of the findings relating to emotions as well as to feelings and look into the possibilities of gaining insight about emotional response potentials from measurements of feelings.
2. Alternative information processing and other ways of handling incoming information Cognitive psychology has dominated the consumer behaviour researchers’ study of information and information processing since the middle of the century. Early contributions are Hovland et al. (1953), Rogers (1962/1995), Hansen (1976), Fishbein (1966), McGuire (1976) and Bettman (1979). Fig. 1 summarizes some of the major models of this kind. The classical AIDA formulation goes back to the end of the 19th century and is explicitly formulated by Copeland (1925). The hierarchy of effects model is most convincingly presented by McGuire (1976) or W7rneryd (1959). The Defined Advertising Goal for Measured Advertising Results (DAGMAR) model is presented by Colley (1961) on behalf of the Advertising Research Foundation and the product adoption model represents Rogers’ (1962/1995) integration of a wealth of research to explain the way in which new products and ideas are accepted among consumers. Different researchers offer a number of observations that are difficult to interpret in terms of the cognitive effect hierarchy way of thinking (Hansen, 1976). Picture perception has been studied by psychologists and Nickerson
1427
(1968) presents highly relevant findings for our purpose. Imagine an experiment where you show 200 pictures for 2 s each to an audience of 200 (often undergraduate students). On the following day, you do the same; only this time, you include 200 new pictures to make a total of 400 pictures. Following the exposure of each picture, you ask a question to test if people can recognise the pictures from the day before. The overall observation in such an experiment is that 95–100% of the stimuli presented on the first day are recognised. The research has gone on to look into how such recognition persists over time, the role of the length of the exposure time, the nature of the pictures presented, and the motivation of the subjects in the audience. Provided 2 s are allowed for exposure (or 1 s for exposure and 1 s for undisturbed information processing before the next exposure), the recognition process is extremely efficient and documents an enormous capacity for storing of such information in the brain. This information storage is not limited to short- or medium-term memory and may persist for years. Another line of research has been reported by Zajonc (1968). Here, the purpose was to study how the evaluation of items, to which subjects is exposed, increase with the number of exposures. In a classic study, Chinese characters and nonsense words have been used. The experimental design is simple. On the first day, you show a sequence of, for instance, Chinese characters, but you control your stimuli so that some of the characters are shown 1, 2, 5, 10, or even 25 times. The following day, you ask the same respondents what they think the meaning is of the different Chinese characters in terms of being more or less positive. Findings from such studies have shown a clear dmere exposureT hypothesis. Mere exposure improves evaluation in many situations, but, of course, limits exist. If the words are not nonsense words, but rather meaningful ones, such as names of known
Fig. 1. Information-processing models used in testing of communication.
1428
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
persons, cities or brands, what you in advance know and think about the persons or cities may inhibit the mere exposure effect. Zajonc reports finding from students counting the frequency with which different words occurred in a sample of literature from Chicago University Library. With a different sample of students, they ranked the average positive, negative evaluation of the different words. The mere exposure relationship is evident in their data, but, remarkably, exceptions can be observed too. In the 1960s and 1970s, corporate image became a major issue in applied research. Some of the major findings from the Mori group in London are reported in Worcester (1979). The basic observation here is that if you measure how frequently a person has heard about a particular company (self-rated awareness) and how positively/negatively the same person evaluates that company (self-rated attitude), a classical image-positioning picture emerges. Here, it appears that the better known a company is, the more positively it is evaluated. In an extension of the mere exposure research, Zajonc and Markus (1982) studied what they labelled emotions without cognitions. In this line of research, the basic idea is very simple and can be illustrated with one of their classical studies. You ask two groups of respondents to evaluate Walkmans by listening to music and speech from these, which they have mounted on their head. The respondents are told that it is an experiment designed to evaluate the quality of a new type of Walkman. The two groups listen to the same content under the same conditions and with the same instructions. The only difference between the two groups is the following. In the test, it is explained that when you use a Walkman in real life, you will often do so when walking, running or in other ways be occupied with parallel activities. To simulate the effect of these other activities in the listening situation, the respondents are asked to nod their heads while listening, whereas the respondents in the other group are asked to shake their heads. Surprisingly enough, the group nodding their heads evaluate their Walkmans significantly better than those shaking their heads. Finally, research concerned with left- and right-brain information processing should be mentioned also. The whole issue is complicated and it is doubtful whether its applicability to the study of consumer information processing has been properly demonstrated. However, the following are some of the characteristics normally associated with leftbrain versus right-brain information processing. The left brain focuses on conscious, verbal, analytical, sequential and arithmetic information processing or with what psychologists call cognitive information processing. The right brain, in contrast, much more involves unconscious, nonverbal, synthetic, holistic, geometrical or spatial information processing (Sperry, 1982; Hansen, 1981; 1985). These findings suggest that things occur when we receive information in situations like those in which we are exposed to advertising that are very different from the processes that the cognitive psychologists assume.
3. Peripheral and low-involvement information processing In central and peripheral information processing, the concept of low involvement is essential. Emphasised by Krugman (1968) and Mitchell and Olson (1981), involvement has been made operational by Zaichkowsky (1985). In many studies of consumer information processing, it has been demonstrated that with low involvement, consumers are less accessible, have fewer defences and are little motivated to cope with complicated information. Probably one of the more integrated approaches to dealing with low-involvement information processing is found in the Rossiter and Percy (1999) informational grid. Here, two important distinctions are made. One of them has to do with the kind of motivation that drives consumers in connection with different purchases. This distinction relates to positive (such as obtaining good taste in food, feeling happiness by demonstrating good taste in clothes) and negative/avoidance motivation (such as avoiding a headache by using headache remedies, avoiding too much trouble in winter by having a good snow shovel, etc.). The other distinction relates to high and low involvement in the already mentioned terminology of Zaichkowsky (1985). From this platform, the authors propose four different kinds of communication situations shown in Fig. 2. Whereas the high involvement and particularly highinvolvement informational processing comes close to what we traditionally find in the effect hierarchy cognitive type of modelling, the low-involvement transformational kind of processing is of a much more emotional nature, occurs at a low level of consciousness, and is stored in terms of implicit memory (Heath, 2001). In this conceptual framework, a distinction between recognition versus recall is important. In many purchase situations where involvement is low and little factual information is available, recognition of the alternative to be chosen is of major concern. Here, a parallel to the subsequently discussed peripheral processing appears; in contrast, in other purchasing tasks of a more elaborate nature with planning and comparison of alternatives, the ability to recall relevant alternatives is extremely important. It is suggested that communication resulting in recall is structurally different from information solely providing recognition. As we shall see subsequently, this is very relevant for some of the aspects of contemporary thinking about emotional information processing. Another theoretical approach to the study of consumer information processing is presented by Petty and Cacioppo (1986). Here, the distinction between peripheral versus central information processing is fundamental. Central information processing is pretty much what we think of in terms of the cognitive hierarchical treatment of information, whereas peripheral relies on less elaborate, probably less conscious and probably also more emotional types of information processing. The authors themselves write:
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
1429
Fig. 2. The Rossiter–Percy grid (Rossiter and Percy, 1999).
bWe have outlined two basic routes to persuasion. One route is based on the thoughtful (though sometimes biased) consideration of arguments central to the issue, whereas the other is based on affective associations or simple inferences tied to the peripheral cues in the persuasion context. When variables in the persuasion situation render the elaboration likelihood high, the first kind of persuasion occurs (central route). When variables in the persuasion situation render the elaboration likelihood low, the second kind of persuasion occurs (peripheral route). Different consequences occur from the two routes to persuasion.
Attitude changes via the central route appear to be more persistent, resistant, and predictive of behaviour than changes induced via the peripheral routeQ. The Petty and Cacioppo (1986) Elaboration Likelihood Model (ELM) is modified by Hansen (1997) in terms of an Elaboration Likelihood Advertising Model (ELAM). This model is illustrated in Fig. 3. Here, central information processing focuses first and foremost on product and brand relevant information, which generates brand awareness, brand perception, image preferences and eventually buying intentions.
Fig. 3. The ELAM (Hansen, 1997).
1430
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
Peripheral information processing following exposure and attention concerns itself more with how the message looks, what the story is, what perceptual representations, such as, e.g., music, pictures, etc., appear in the advertisement or commercial. These, in turn, generate attitudes towards the ad (rather than towards the brand) and emotional responses all reflected in ad liking. Attitudes towards the ad and ad liking may influence the extent and nature of the parallel central information processing to the extent that such occur. It may also lead to implicitly stored memories that may eventually be generated as recognition and possible purchase influence in purchase situations. Variables used to reflect the central processing are often brand related in terms of brand recall, brand recognition, brand processing, attitude towards the brand preferences for the brand, purchase intentions and possibly changing purchase behaviour. Measurements more related to peripheral information processing are concerned with ad recall, ad recognition, ad processing, attitudes towards the ad emotions and ad liking. Particularly in connection with advertising processing, different studies have been conducted to identify the characteristics of peripheral versus central processing at the Center for Communication Research at the Copenhagen Business School.
4. The Copenhagen peripheral versus central information processing studies Center for Marketing Communication has had access to 18 standardised advertising pretests for fast-moving consumer goods, conducted by Gallup/TNS in Denmark 1998– 1999. Each of these included 120–150 respondents. They follow the lines described in the ELAM model. In each test, questions are partly formulated at the brand level (e.g., attitudes toward the brand) and partly at the advertising level (advertising recall and A-Ad). A more detailed account of this study is found in Hansen (1997) and Hansen and Hansen (2001). Here, we are particularly concerned with the extent of peripheral versus central information processing and the Table 1 Number of central and peripheral responses in a study of 18 ads (a respondent may be counted both as one with negative and as one with positive central responses, or as one with both negative and positive peripheral responses) Central positive Central negative Central Peripheral positive Peripheral negative Peripheral No. of respondents classified Unclassified respondents Total a
No. of respondents
No. of statements
353 231 540 1209 1068 1627 2167 310a 2477
706 320 1026 1603 1488 3091
4117
Respondents with no classifiable responses are excluded.
Table 2 Self-rated recall, liking and buying intention
Respondents with central information processing Respondents with peripheral information processing
Self-rated recall
Liking
Buying intention
3.9
3.9
2.4
2.3
2.2
1.1
On a 6 point scale, 5–0. All differences significant, Pb.01.
nature of the responses depending upon which data processing approach is dominating. Responses from three open questions asked in the course of the test, partly evaluating positive and negative aspects of the ads and partly asking for what the respondent thinks the advertising is supposed to communicate, were analysed. A standardised procedure for probing respondents was used and responses were coded by two independent coders to their basic elements; for instance a bgood-looking girl in beautiful carQ was divided into bgoodlooking girlQ and bbeautiful carQ. Subsequently, all such informational items were categorised depending on whether they were positive or negative, and depending on whether they primarily related to the product, its use, its advantages (central) or whether they primarily reflected the story in the ad, its pictures, its underlying music, and its execution (peripheral; see Table 1). The relatively few items that were unclassifiable along these dimensions were excluded. On average, around two informational elements were registered for each respondent. The total number of useful statements totalled 4117. Several observations can be made from these figures. First, more than three times as many peripheral statements can be identified relative to the number of central informational ones. Secondly, when central information processing dominated, respondents were mostly positive in their responses. With peripheral information processing however, negative statements near the number of positive ones, and here, 650 or more than 1/3 of the respondents reported positive as well as negative ad-related items. Only 44 respondents (353+ 231540) repeated both positive and negative, centrally (brand) related statements. Summary results regarding the nature of central versus peripheral communication effects are shown in Table 2. Table 3 Attitudes towards the ad in % of respondents with central or peripheral information processing Exciting Credible Sensitive Warm Entertaining Informative Stupid Irritating
Central, n=540 (%)
Peripheral, n=1627 (%)
22 42 31 22 36 32 19 17
19* 33* 27* 21 32* 29* 24* 24*
* Significant, Pb.01 t test.
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
Here, when central information processing dominates, the ad is recalled better, liking is higher, brand preference is higher and buying intention (here measured as positive selfrated changes in buying intention following the exposure) is higher. In addition, attitudes towards the ad show the same picture. The items here are borrowed from the Gallup– Robinson standardised advertising pretesting procedure (Metha, 1994). Two statements from each of the four factors underlying this instrument (informative, entertaining, evaluative and negative attitudes) are used (see Table 3). Five out of six positive attitudes score (significantly) higher when central information processing is involved, and the negative ones do the opposite. All measures suggest that better communication results are achieved when central information processing is generated. One exception occurs. The test included 12 feeling statements (Table 4). Here, we found that peripheral information processing resulted in more positive emotional responses and fewer negative emotional ones. Of the 12 items used, 9 show significant differences in the direction indicated here. Summarizing this second part: when predominantly concerned with fast-moving consumer goods advertising, 75% of the information processing is peripheral. When central information processing occurs, it is always more efficient as indicated by practically all commonly used measurements of the communication effect. When peripheral information dominates, only the emotional responses tend to be more positive. An overall conclusion may be that we may like to generate central information processing as much as we can; however, when we cannot achieve this, we must be concerned particularly with the emotional responses following the peripheral information processing. In designing campaigns, it is likely to be a basic issue whether one should emphasise central information processing by providing lots of relevant product and use information or whether one should focus on creating more attention and thereby possibly also ending up with more peripheral information
Table 4 Self-rated feelings/emotions associated with central and peripheral information processing (in % of respondents) Pleasure Hope Acceptance Happiness Dominate Enjoyment Inspiring Surprising Mistrust Sorrow Anger Fear
Central, n=540 (%)
Peripheral, n=1627 (%)
26 29 22 27 9 46 29 19 8 7 12 3
40* 33* 34* 41* 13* 58* 28 23* 9 5* 7* 5
* Significant, Pb.01 t test.
1431
processing. The more creative advertising executions may have a tendency to do the latter. That is, high attention is achieved at a price in terms of lesser informational content. Of course, the good, creative solution generates both central and peripheral information, but is difficult to get at.
5. The nature of emotional responses Presenting an extensive review of contemporary neuropsychological research is impossible here. Excellent discussions are available in Damasio (2000), Goode (2002) and Franzen and Bouwman (2001). It is, however, necessary for our discussion of emotional responses to repeat some basic observations from this line of research. Basically, the brain can be seen as composed of three elements. The so-called neo cortex, that is, the outer part of the brain and in humans by far the largest. Cognitive processes are primarily believed to take place here and this occurs with some specialisation in terms of the left and right side of the brain as discussed earlier. The second part, the so-called old cortex, is found in mammals and in all animals as low as at the reptilian stage of development. This system functions as the controlling brain system in most animals and plays an important role in interacting with the cortex in the human brain. Finally, the inner central or oldest part of the brain—the prereptilian brain is where the most basic, elementary, controlling processes occur. Particularly, the prereptilian part of the brain contains thalamus through which most sensory stimulation passes and amygdale, which is shaped like an almond and controls the most elementary responses, such as glandular behaviour and autonomous responses, and hippocampus—named after its shape, which resembles a sea horse (Fig. 4). In the hippocampus, very elementary information is stored, and in interaction with amygdale, it controls the simple emotional responses. It may be the seat of implicit memory. Contemporary neuropsychology refers to the processes controlled by thalamus, amygdale and hippocampus as emotional. They may occur before any cognitive activity is activated in responses to stimuli and they control very elementary approach and defensive, aggressive responses of importance to the survival of the individual. The entire prereptilian brain interacts with the cortex and information is transmitted, coded, edited and stored here. In addition, in this part of the brain, the occurrence of different kinds of feelings (different from emotions) can appear. Whereas the basic emotional responses are of a rather primitive avoidance/approach character, people report in much more detail when asked to describe aspects of the relationship between perceived situations, goals and more basic emotions (feelings). Emotional responses can first and foremost be identified with the use of physiological measurements. Their EEG response, eye movements, heart rate, voice and facial expressions are measures frequently used. Because the emotions seem to be products of the
1432
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
Fig. 4. The function of the emotional brain (Le Doux, 1998).
prereptilian brain, our verbal questions about feeling responses do not necessarily relate directly to underlying emotional states. If, for example, an individual in the middle of a road observes a car approaching with fast speed, the perception channelled through the thalamus may, through the amygdale, generate an increased heart rate (autonomous response), sweating in the hands (glandular response) and freezing or running away. All these may occur before any activity in cortex takes place. Only later, when information has been transmitted here, can the more precise nature of the danger be identified and labelled, and possibly, this may influence the further direction of the response. In a more straightforward consumer behaviour case, the individual may—faced with a row of coffee brands—identify the brand usually bought, pick it up and conclude the activity. This may happen before any coffee information processing occurs or prior to the activation of earlier stored information about quality or other aspects of different coffee brands. Emotional responses, mostly unconscious, automatic reactions, are controlled by information stored in implicit memory. Feeling responses, the associated, perceived, cognitive perceptions of what is going on emotionally, involve centrally stored information, cognitive processing and evaluations. In this connection, the distinction, emotions versus feelings, is not synonymous with the distinction, conscious versus unconscious processes. Feelings may be conscious as well as unconscious. Cognitive processes may be of a high- or low-involvement character and may also relate to emotional activities. We realise the automatic, very little demanding, fast and energy-saving emotional responses occur in many situations where cognitions, comparisons and information searches would have been unnecessarily time demanding and complicated.
In earlier psychological theory and in studies of consumer behaviour, emotions and feelings are often used interchangeably. In our terminology, feelings have been studied since the very early days of psychology and a large number of measurement instruments used for identifying different kinds of feelings based upon verbal responses have been developed (Franzen and Bouwman, 2001). The number of feelings identified with a different instrument varies from few (two to four) to as many as 20 or 30. In many of these studies, a factor analysis has been used based upon the respondents’ own rating of words that reflect different kinds of feelings. One such typical set is reported by Richins (1997). Departing from several of the more generally accepted batteries, such as Izard (1977) and Mehrabien and Russell (1974), she identified 16 plus 4 consumption-related feelings. In most cases, these can be reduced to two to three meaningful factors. One case of this kind is Shaver et al. (1987). Here, a hierarchical cluster analysis on a large number of feeling items results in 24 grouped feelings. However, when these are grouped at higher levels, they fall into two (or three) separate sets: a predominantly positive (approach dimension—love and joy); a predominantly negative set (anger, sadness and fear); and possibly an arousal-like dimension labelled surprise. Of course, the amount of variance explained with these lower order levels is less than when a larger number of factors are included, but the uniformity of the dimensions and their face validity is such that the solutions can still be judged to be operationally useful. Some authors have even focussed on these fewer more fundamental dimensions. For instance, Mehrabien and Russell (1974) talk about pleasure versus dissatisfaction. Similarly, in a commercially applied setting introduced by Vaughn (1980), a distinction is made between negative/ positive versus active/passive feelings.
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
1433
underlying dimensions identified. Only in the design test was the avoidance dimension less expressed. Here, one might better work with three or four factors. The data give no uniform answer to the question of whether one should ideally work with one, two, three, or maybe even four emotional dimensions. If a need for a more general, basic measurement instrument is felt, more research is needed before such standardised list of items can be identified with the purpose of measuring across very different consumption and communication situations. Still, based on the following case from one of the studies already completed, it is possible to illustrate the usefulness and the strengths of such an emotional battery.
The fundamental observation made here is that in the terminology of an SOR model, and with the use of selfadministered questionnaires, we may identify 10–20 different indicator variables describing aspects of feelings. However, in forcing solutions and fewer dimensions upon the data, more basic dimensions emerge. These dimensions we may rightfully label emotions. Thus, our major observation at this point is that we may still gain some insight into emotional responses based on measurement of feelings and by doing a bconcentratedQ analysis of the data (i.e., looking for fewer factors). Here, some of our own data we present from four different studies of communication effects (see also Hansen, 1997). The design test and the logo tests are reported in more detail in Kristensen et al. (2000), and the two sponsoring studies are reported in more detail in Hansen et al. (2002). All tests are based on the variables used in the ELAM test. In these studies, one statement corresponding to each of the underlying feeling dimensions in the Richins (1997) battery is included. In some of the experiments, the battery has been reduced further, but in each of the tests, at least the same 12 feeling items are used. In addition, there is good reason to look at findings across the different applications in the different contexts involving the short list of 12 items. In all cases, a two- to four-dimensional factor solution is meaningful. All solutions include an approach (positive) and an avoidance (negative) emotional dimension, and one or two dimensions reflecting involvement with the issue, strength of the emotion or arousal. To force comparability upon the data, two-factor solutions are shown for all data in Table 5. The amount of variance explained with the relatively limited number of factors is lower than ideally wished (26– 32%). However, the consistency of the dimensions across the different data sets confirms the general validity of the
6. The value of sponsorships explained by emotions We choose to look into the findings from the study concerned with how possible sponsoring objects (sponsees) are evaluated (Hansen et al., 2002) in a little more detail. Here, the respondents were 169 first-year undergraduate students at the Copenhagen Business School completing a self-administered questionnaire for which they received a book on advertising published by the Center for Communication Research as a reward. The authors involved in this project do not claim that the findings are in any way representative of the Danish population when it comes to evaluating possible sponsees. They do argue, however, that the relationships identified between the different measurements are much less sensitive to the sample bias involved here and thus suggest a more general usefulness of the measurement instrument with larger random samples. Such a project is presently underway. Basically, the 27 feeling items used are derived from Richins’ (1997) study. Some feeling statements obviously less related to consumer information processing are deleted
Table 5 Loadings on two factors in analyses of data on sponsoring, ad-test, design and colours of logo (I=Approach, II=Avoidance) Sponsoring Happiness Joy Pleasure Accept Inspiring Hope Surprise Anger Fear Mistrust Trust Sorrow Dominant Trustworthiness None of these x) indicates missing data.
Ad test
Design
Colours of logo
I
II
I
II
I
II
I
II
.39 .61 .48 .45 .38 .33 .02 .07 .03 x) .32 .07 .10 x) x)
.03 .11 .01 .56 .16 .02 .16 .61 .45 x) .36 .40 .08 x) x)
.70 .67 .58 .56 .54 .51 .40 .01 .17 .06 x) .07 .23 x) .50
.01 .04 .21 .08 .05 .12 .19 .67 .65 .60 x) .50 .44 x) .13
.72 .78 .75 .06 .62 .43 .16 .84 .73 .71 x) .76 .47 x) x)
.16 .06 .03 .55 .43 .30 .81 .11 .13 .07 x) .31 .30 x) x)
.63 .71 .54 .02 .50 .38 .22 .53 .49 .55 x) x) .25 .08 x)
.07 .23 .05 .68 .13 .13 .51 .47 .48 .53 x) x) .10 .72 x)
1434
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
and some more appropriate items taken from the Ray and Batra battery (1982) are included. The total list is shown in Table 6. The measurement of the feeling responses took place by having each respondent choose feeling statements they felt agreed with 27 possible sponsoring events divided into sports, culture, television programmes and social aid organisations. The sponsors studied are listed in the first column of Table 7. With the somewhat crude data collection procedure of respondents choosing only a few items to be associated with each possible sponsoring object and with a total of 27 items reduced to four factors, the amount of variance explained in the data is 29% of the total variance. The face validity of the four factors suggest that each of the factors reflect one important emotional dimension when dealing with sponsoring. The four-dimensional solution chosen centres on an avoidance (negative emotional) and approach (positive emotional) dimension (numbers two and three). Loneliness and sadness are the two highest loading items reflecting avoidance, whereas joy, romantic love, enjoyment, happiness are the four items that best represent the approach tendency. The fourth factor in the battery, labelled arousal, reflects feelings, such as excitement, surprise, lack of trust and lack of accept. Finally, the first
Table 6 Varimax factor analytical solution based on a four-factor solution on emotional responses to different sponsoring aspects
Hope Sorrow Worry Fear Inspiring Optimism Dominating Satisfaction Shame Loneliness Anger Sad Envy Desire Guilt Dissatisfaction Joy Romantic love Enjoyment Happiness Peacefulness Excitement Surprising Trust Accept Relief Pride
bUncertaintyQ
bAvoidanceQ
1
2
3
4
59 57 46 45 31 28 23 16 0 4 3 15 2 7 3 24 6 0 14 14 5 12 7 8 8 4 18
19 16 6 14 15 27 17 9 54 40 40 40 37 34 34 28 17 5 5 0 8 8 3 11 10 10 7
19 1 3 4 0 17 6 12 1 4 4 2 7 5 2 2 62 52 51 49 31 17 2 15 21 1 9
19 0 8 2 13 12 6 9 4 11 11 0 0 30 7 3 20 10 0 3 18 64 44 40 35 22 21
Decimals were omitted.
bApproachQ
bArousalQ
Table 7 Standardised scores for 27 different sponsoring objects on awareness, liking and self-rated purchase intention, together with total score for each (geometrical average score) N=169
Awareness
Liking
Purchase int.
Overall score
The Danish Cancer Society Danish Red Cross Save the Children Denmark The AIDS Foundation Danchurchaid The Danish Muscular Dystrophy Association The Danish Heart Foundation Average—social aid organisations Friends (TV2) The Weather Report (TV2) Ally McBeal (TV2) Rejseholdet (DR) Onside (TV3) Big Brother (TVDANMARK) The Hotel (TV2) The Great Mission (TV2) Average—tv programmes Tivoli Eurovision song contest The zoo The royal theatre The Roskilde Festival The Danish National Gallery of Art Arken Average—culture Men’s National Soccer Squad Women’s National Handball Squad Team Danmark FC Copenhagen Brbndby I.F. Average—sport
1.09
1.23
1.16
1.55
1.05 0.99
1.19 1.18
1.10 1.11
1.37 1.30
0.91
1.21
1.15
1.26
0.78 0.68
1.09 1.13
1.03 1.09
0.88 0.84
0.62
1.18
1.11
0.81
0.87
1.17
1.11
1.14
1.33 1.23
0.95 0.86
1.01 0.93
1.27 0.98
1.08 0.99 0.84 1.09
0.89 0.88 0.88 0.67
0.94 0.95 0.97 0.74
0.91 0.82 0.71 0.54
0.72 0.77
0.78 0.76
0.90 0.85
0.51 0.49
1.01
0.83
0.91
0.78
1.41 1.27
1.13 1.02
1.07 1.00
1.69 1.29
1.05 0.98 1.08 0.80
1.07 1.06 0.95 1.04
1.04 1.01 0.96 1.03
1.17 1.05 0.98 0.86
0.70 1.04 1.22
0.94 1.03 1.06
0.96 1.01 1.07
0.63 1.10 1.39
1.11
0.99
0.98
1.08
0.97 1.18 1.05 1.11
1.06 0.95 0.88 0.99
1.05 0.94 0.87 0.98
1.07 1.06 0.80 1.08
item, labelled uncertainty, relies on responses to feelings, such as hope, sorrow, worry and fear. The usefulness of such an emotional battery, of course, depends upon the extent to which it meaningfully describes the different items (sponsoring objects) rated, and the extent to which measurements relate meaningfully to traditional, well-established effect– measures used in the study of sponsorship effects. In the study reported here, different traditional effectrelated measures of sponsoring were included. Awareness of the sponsee was measured on a five-point self-rating scale
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
(bHow well would you say you know. . .Q), the same was done for liking, and persuasion was quantified as selfreported changes in purchase intentions. The question was bdo you think that it is more likely that you would purchase a product sponsoring dxxxxT after having learned about the sponsorship than beforeQ. The measurement, crude as it is, has been applied in most of the studies reported earlier, and on the whole, is sensible and meaningful. On average, 50% of the respondents report that they do not intend to change their purchase intentions. However, the rest do report positive as well as negative purchase intentional changes. These measures in ad testing are in accordance with other research believed to make good sense in evaluating sponsorship attitude based on these. A total score is computed as the geometric average of the evaluations of the sponsorships with the sponsoring objects. Overall effect score Awareness Linking ¼ Average Awareness Average Linking Self rated change in buying intention Av: Self rated change in buying int: The overall score for all of them is 1.00, suggesting that a sponsoring object scoring higher is evaluated as better than average and vice versa. That is, for instance, the liking score for the Danish Cancer Society was divided with the average liking score for all sponsoring objects rated and thereby a score larger than one reflects an above-average evaluation. The same was done with the awareness and purchase intention scores and to arrive at a single overall evaluative score, the three past scores were simply multiplied with each other. The critical question now, of course, is to what extent the emotional responses we have identified are able to explain variations in this overall effect score. This is done by computing a factor score for each sponsoring object on the four dimensions, shown in Table 6. With these as independent variables, and with the total effect score of Table 7 as a dependent variable, a multiple regression analysis was carried out. For the 27 rated sponsoring objects as dependent variables, the analysis provides an adjusted R 2 score of .60. It appears that negative emotions, positive emotions and uncertainty are significantly related to the overall evaluation, with the greatest explanatory power associated with negative emotions (b=.003, t=3.54, Pb.001), followed by uncertainty (b=.0024, t=4.34, Pb.001) and positive emotions (b=.0016, Pb.01). We shall not try to extract more far-reaching interpretations from these data, but we do want to emphasise that an adjusted R 2 value of .60 in an analysis with 27 observations, four independent variables and basic ratings simply made by having people choose whether specific feeling statements do or do not relate to sponsoring objects seems to be a strong
1435
confirmation of the usefulness of the approach. We look forward to work with larger, more representative samples with more sensitive rating procedures applied in other marketing communication contexts than sponsoring. So far, however, we dare conclude that energy spent in this direction seems worthwhile.
7. Conclusion Despite the somewhat tentative nature of the findings reported here, a number of conclusions emerge to which we would like to direct the attention of the reader. (1) More and different measures are available when studying marketing communication responses than what is suggested in AIDA and similar formulations. (2) A distinction between central/peripheral, higher/lower involvement or more or less cognitive information processing seems useful. (3) The central information processing is, when it occurs, by far more efficient in terms of effect scores normally used in marketing communication studies. (4) Only in terms of emotional responses does peripheral information processing seem to have a stronger impact. (5) In the real world, the advertiser may wish to generate strong concentrated central information processing. (6) However, competition from other communication, other advertising, low involvement on behalf of the receivers, etc. set limits to the extent to which this is feasible. (7) In reality, at least when talking about fast-moving consumer goods, peripheral information processing seems to be dominant in most instances. (8) To study emotional responses may be a useful approach, when peripheral responses are dominant. (9) With a distinction introduced between feelings and emotions, it is suggested that an operational measure of more basic emotions may be derived from scores based on statements of feelings of a more cognitive– conscious nature. In any event, emotional activities contribute significantly to the overall effectiveness of communication. This is not least the case when peripheral communication is at stake. It is not obvious from the data analysed here whether the emotional effect primarily makes other information processing more efficient, or whether it is a more direct effect with information being stored at an emotional (implicit) level and reactivated in purchasing and other situations at the same very low prereptilian brain level. In all events, we do not know enough about how emotions are formed or how they influence subsequent consumer behaviour. The area must be a central one for future research in consumer behaviour.
1436
F. Hansen / Journal of Business Research 58 (2005) 1426–1436
References Bettman JR. An information processing theory of consumer choice. Reading7 Addison-Wesley Publishing; 1979. Colley RH. Defining advertising goals for measured advertising results. New York7 Association of National Advertisers; 1961. Copeland MT. Principles of merchandising. New York7 Shaw; 1925. Damasio A. The feelings of what happens. London7 Vintage; 2000. Fishbein M. The relationship between beliefs, attitudes and behavior. In: Feldman S, editor. Cognitive consistency. New York7 Academic Press; 1966. p. 199 – 223. Franzen Giep, Bouwman Margot. The mental world of brands. Reading7 ACR; 2001. Goode Alistair. The value of implicit memory. Hanley-on-Thames7 ADMAP; 2002 [December]. Hansen Flemming. Psychological theories of consumer choice. J Consum Res 1976;3(1):117 – 42. Hansen Flemming. Hemispheral lateralization: implications for understanding consumer behaviour. J Consum Res 1981;8(1):23 – 7. Hansen Flemming. Studies of communication effects—methodological and theoretical papers on left/right lateralization. Copenhagen7 Civilbkonomernes Forlag; 1985. Hansen Flemming. Quantifying creative contributions: advertising pretesting’s new generation. ESOMAR Conference Proceedings, Edinburgh, September, 1997. Hansen, Flemming and Lotte Yssing Hansen. The Nature of Central and Peripheral Advertising Information Processing, Research Paper, Department of Marketing, Copenhagen Business School, 2001. Hansen, Flemming, Halling, Jens, Christensen, Lars Bech. Choosing Among Alternative Sponsoring Objects for Supporting Brand Strategies, based upon Emotional Responses. Research Paper no. 12/2002, Dept. of Marketing, Copenhagen Business School, 2002. Heath Robert. The hidden power of advertising. London7 Admap Publications; 2001. Hovland CI, Janis IL, Kelly HH. Communication and persuasion. New Haven7 Yale University Press; 1953. Izard Carroll E. Human emotions. New York7 Plenum; 1977. Kristensen, Tore et al., The Meaning of Colours in Design, Research Paper, Department of Marketing, Copenhagen Business School, 2000. Krugman HE. The impact of television advertising: learning without involvement. Public Opin 1968;29. Le Doux Joseph. The emotional brain. New York7 Phoenix; 1998. Lutz Richard J. Affective and cognitive antecedents of attitude toward the ad: a conceptual framework. In: Alwritt L, Mitchell Andrew A., editors. Psychological processes and advertising effects. Hillsdale, NJ7 Erlbaum; 1985. p. 45 – 63.
MacKenzie SB, Lutz RJ. Monitoring advertising effectiveness: a structural equation analysis of the mediating role of attitude towards the ad. Working paper No. 117, Center for Marketing Studies, University of California, Los Angeles, 1982. McGuire WJ. Psychological factors influencing consumer choice. In: Ferber B, editor. Selected aspects of consumer behavior. Washington7 University Press; 1976. p. 319 – 60. Mehrabien Albert, Russell James A. An approach to environmental psychology. Cambridge (MA)7 MIT Press; 1974. Metha Abdilasha. How Advertising Response Modelling (ARM) can increase ad-effectiveness. J Advert Res 1994; 34(3),1994;62 – 74 [May/June]. Mitchell AA, Olson JC. Are product attribute beliefs the only mediator of advertising effects on brand attitude? J Mark Res 1981;18:318 – 32. Nickerson RS. A note on long-term recognition memory for pictorial material. Psychon Sci 1968;11:58. Petty JT, Cacioppo RE. Communication and persuasion. New York7 Springler-Verlag; 1986. Petty RE, Cacioppo JT, Schumann David. Central and peripheral routes to advertising effectiveness: the moderating role of involvement. J Consum Res 1983;10(2):135 – 46 [September]. Richins ML. Measuring emotions in the consumption experience. J Consum Res 1997;24;127 – 42. Rogers EM. Diffusion of innovations. New York7 The Free Press; 1962/ 1995. Rossiter John, Percy Larry. Advertising and promotion management. New York7 McGraw-Hill; 1999. Shaver Philip, Schwartz Judith, Kirson Donald, O’connor Cary. Emotion knowledge: further exploration of a prototype approach. J Pers Soc Psychol 1987;52:1061 – 86 [June]. Sperry R. Some effects of disconnecting the cerebral hemispheres. Science 1982;217:1223 – 6. Vaughn R. How advertising works: a planning model. J Advert Res 1980;20(5):27 – 33 [New York, Oct]. W7rneryd KE. Ekonomisk psykology. Stockholm7 Institute for Economic Research, Stockholm School of Economics; 1959. Worcester R. Strategic decision research: its communications and use at ICI. Paper presented at the ESOMAR Congress, Brussels, September, 1979. Zaichkowsky JL. Measuring the involvement construct. J Consum Res 1985;12(3):341 – 52. Zajonc RB. Attitudinal effects of mere exposure. J Pers Soc Psychol 1968;9(2). Zajonc RB, Markus H. Affective and cognitive reasons factors in preferences. J Consum Res 1982;9(2):123 – 31.
Journal of Business Research 58 (2005) 1437 – 1445
Emotions in consumer behavior: a hierarchical approach Fleur J.M. Laros*, Jan-Benedict E.M. Steenkamp Marketing Department, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands Received 1 December 2002; received in revised form 1 September 2003; accepted 1 September 2003
Abstract A growing body of consumer research studies emotions evoked by marketing stimuli, products and brands. Yet, there has been a wide divergence in the content and structure of emotions used in these studies. In this paper, we will show that the seemingly diverging research streams can be integrated in a hierarchical consumer emotions model. The superordinate level consists of the frequently encountered general dimensions positive and negative affect. The subordinate level consists of specific emotions, based on Richins’ (Richins, Marsha L. Measuring Emotions in the Consumption Experience. J. Consum. Res. 24 (2) (1997) 127–146) Consumption Emotion Set (CES), and as an intermediate level, we propose four negative and four positive basic emotions. We successfully conducted a preliminary test of this secondorder model, and compare the superordinate and basic level emotion means for different types of food. The results suggest that basic emotions provide more information about the feelings of the consumer over and above positive and negative affect. D 2004 Elsevier Inc. All rights reserved. Keywords: Consumer emotions; Hierarchy of emotions; Positive and negative affect; Basic emotions; Specific emotions
1. Introduction After a long period in which consumers were assumed to make largely rational decisions based on utilitarian product attributes and benefits, in the last two decades, marketing scholars have started to study emotions evoked by marketing stimuli, products and brands (Holbrook and Hirschman, 1982). Many studies involving consumer emotions have focused on consumers’ emotional responses to advertising (e.g., Derbaix, 1995), and the mediating role of emotions on the satisfaction of consumers (e.g., Phillips and Baumgartner, 2002). Emotions have been shown to play an important role in other contexts, such as complaining (Stephens and Gwinner, 1998), service failures (Zeelenberg and Pieters, 1999) and product attitudes (Dube et al., 2003). Emotions are often conceptualized as general dimensions, like positive and negative affect, but there has also been an interest in more specific emotions. Within the latter stream of research, some researchers use a comprehensive set of
* Corresponding author. Tel.: +31 13 4668212; fax: +31 13 4662875. E-mail address: [email protected] (F.J.M. Laros). 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.09.013
specific emotions (Richins, 1997; Ruth et al., 2002). Other researchers concentrate on one or several specific emotions, such as surprise (e.g., Derbaix and Vanhamme, 2003), regret (e.g., Inman and Zeelenberg, 2002; Tsiros and Mittal, 2000), sympathy and empathy (Edson Escalas and Stern, 2003), embarrassment (Verbeke and Bagozzi, 2003) and anger (Bougie et al., 2003; Taylor, 1994). Despite this emerging body of research, progress on the use of emotions in consumer behavior has been hampered by ambiguity about two interrelated issues, viz., the structure and content of emotions (Bagozzi et al., 1999). First, with regard to structure, some researchers examine all emotions at the same level of generality (e.g., Izard, 1977), whereas others specify a hierarchical structure in which specific emotions are particular instances of more general underlying basic emotions (Shaver et al., 1987; Storm and Storm, 1987). Second, and relatedly, there is debate concerning the content of emotions. Should emotions be most fruitfully conceived as very broad general factors, such as pleasure/arousal (Russell, 1980) or positive/negative affect (Watson and Tellegen, 1985)? Alternatively, appraisal theorists (see, e.g., Frijda et al., 1989; Roseman et al., 1996; Smith and Lazarus, 1993) argue that specific emotions
1438
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
should not be combined in broad emotional factors, because each emotion has a distinct set of appraisals. The confusion concerning structure and content of emotions has hindered the full interpretation and use of emotions in consumer behavior theory and empirical research (Bagozzi et al., 1999). The purpose of our paper is twofold. First, we integrate seemingly opposing research streams in psychology and consumer behavior by developing a hierarchical model of consumer emotions. We will show that the general dimensions with positive and negative affect are the superordinate and most abstract level at which emotions can be defined. The subordinate level consists of specific consumer emotions. We will develop an intermediate level with basic emotions that links these two levels. Second, we conduct a preliminary test of this proposed structure and compare the means for positive and negative affect with those of the basic emotions for four different food types.
2. Emotions in consumer research This section will briefly discuss an illustrative set of consumer studies on emotions (see Table 1 for an overview). Several studies focused on the emotional responses to ads. Holbrook and Batra (1987) developed their own emotional scale based on an in-depth review of the literature. They uncovered a pleasure, arousal and domination dimension in their data, and showed that these emotions mediate consumer responses to advertising. Edell and Burke (1987) also created their own emotion list and found that feelings play an important role in the prediction of the ad’s effectiveness. They proposed three factors: an upbeat, negative, and warmth factor. Olney et al. (1991)
showed that the emotional dimensions pleasure and arousal mediate the relation between ad content and attitudinal components, and consequently viewing time of an ad. They used part of Mehrabian and Russell’s (1974) scale. Derbaix (1995) replicated the research of Edell and Burke (1987) in a natural setting. His emotion words were based on a prestudy, and uncovered a positive and negative factor. Steenkamp et al. (1996) investigated the relations between arousal potential, arousal, and ad evaluation, with need for stimulation as a moderator. They based their arousal dimension on the scale of Mehrabian and Russell (1974). In the satisfaction literature, Westbrook (1987) was one of the first to investigate consumer emotional responses to product/consumption experiences and their relationship to several central aspects of postpurchase processes. Oliver (1993) extended this work by showing that emotional responses mediate the effects of product attributes on satisfaction. Both studies relied on Izard’s (1977) taxonomy of fundamental affects, and found positive and negative affect as underlying emotion dimensions. Mano and Oliver (1993) investigated the structural interrelationship among evaluations, feelings, and satisfaction in the postconsumption experience. They combined Watson et al.’s (1988) PANAS scale and Mano’s (1991) circumplex scale. Both three dimensions—similar to the upbeat, negative, and warmth factors of Edell and Burke (1987)—and two dimensions—positive and negative affect—were uncovered, but only the latter dimensions were used in the studies. Dube and Morgan (1998) modeled trends in consumption emotions and satisfaction in order to predict retrospective global judgments of services. They used the PANAS scale (Watson et al., 1988) and uncovered positive and negative affect. Phillips and Baumgartner (2002) confirmed the
Table 1 Overview of consumer research using emotions as a main variable Reference
Emotion measure used
Resulting structure
Edell and Burke (1987) Holbrook and Batra (1987) Westbrook (1987) Olney et al. (1991) Holbrook and Gardner (1993) Mano and Oliver (1993)
Edell and Burke (1987) Holbrook and Batra (1987) Izard (1977) Mehrabian and Russell (1974) Russell et al. (1989) Watson et al. (1988); Mano (1991)
Oliver (1993) Derbaix (1995) Steenkamp et al. (1996) Nyer (1997) Richins (1997)
Izard (1977) Derbaix (1995) Mehrabian and Russell (1974) Shaver et al. (1987) Richins (1997)
Dube and Morgan (1998) Phillips and Baumgartner (2002) Ruth et al. (2002)
Watson et al. (1988) Edell and Burke (1987) Shaver et al. (1987)
Smith and Bolton (2002)
Smith and Bolton (2002)
Upbeat, negative, and warm Pleasure, arousal, and domination Positive and negative affect Pleasure and arousal Pleasure and arousal Upbeat, negative and warm Positive and negative Positive and negative affect Positive and negative affect Arousal Anger, joy/satisfaction, and sadness Anger, discontent, worry, sadness, fear, shame, envy, loneliness, romantic love, love, peacefulness, contentment, optimism, joy, excitement, and surprise Positive and negative affect Positive and negative affect Love, happiness, pride, gratitude, fear, anger, sadness, guilt, uneasiness, and embarrassment Anger, discontent, disappointment, self-pity, and anxiety
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
importance of including positive and negative affect in explaining satisfaction. Smith and Bolton (2002) investigated the role of consumer emotions in the context of service failure and recovery encounters. They used content analysis for the responses of the participants and grouped the (negative) emotion words of consumers in five categories. Holbrook and Gardner (1993) investigated the relation between the emotional dimensions pleasure and arousal and the duration of a consumption experience, which was in their case, listening to music. They used Russell et al.’s (1989) Affect Grid to measure pleasure and arousal of the musical stimuli. Nyer (1997) and Ruth et al. (2002) focused on defining the antecedents rather than the consequences of emotions.
1439
Nyer (1997) showed that the appraisals of goal relevance, goal congruence, and coping potential are determinants of several basic consumption emotions. These emotions were mainly based on Shaver et al. (1987). Ruth et al. (2002) explored the cognitive appraisals of situations and their correspondence to 10 experienced emotions. They also used emotions based on the hierarchical structure of Shaver et al. (1987). In summary, this overview shows that there is wide divergence in the content of emotions studied in consumer research. Studies often use different scales to measure emotions and focus on different emotions. In spite of this, consumer researchers frequently use, or exploratory data analysis yields, a small number of dimensions (Bagozzi et al., 1999). Among these, the classification of emotions in
Table 2 Emotion words Negative emotion words
Positive emotion words
Aggravationa,b,c, Agitationa,b,c, Agonyb,c, Alarmb,c,d, Alienationb, Anger a,b,c,d,e,f,g, Anguisha,b,c, Annoyancea,b,c,d,e,f,h, Anxietya,b,c,e, Apologeticc, Apprehensiona,b,c, Aversione, Awfulc, Badc, Bashfulc, Betrayalc, Bitternessa,b,c, Bluea,c,i, Botheredc, Cheerlessa, Confusedh, Consternationc, Contemptb,c,e,g, Crankyc, Crossc, Crushedh, Cryc, Defeatb, Deflateda,b, Defensivec, Dejectiona,b,c, Demoralizedc, Depression a,b,c,d,h, Despairb,c, Devastationc, Differentc, Disappointmenta,b,c,e,f, Discomfortc, Discontent a,c, Discouragedc, Disenchantmentc, Disgusta,b,c,e,g,h, Dislikeb,c,g, Dismayb,c, Displeasurea,b,c, Dissatisfieda,c, Distressa,b,c,d,g,i,j, Distrustc,e, Disturbedc, Downa,c, Dreadb,c, Dumbc, Edgyc, Embarrassment a,b,c, Emptya,c, Envy a,b,c, Exasperationb, Fear b,c,d,e,f,g,h,i,j, Fed-upa, Ferocityb, Flustereda, Forlornc, Foolishc, Franticc, Frighta,b,c,h, Frustration a,b,c,d,f,g, Furya,b,c, Gloomb,c,d,h, Glumnessb, Griefa,b,c,f, Grouchinessb,c,i, Grumpinessb,c,i, Guilt b,c,e,g,j, Heart-brokena,c, Hateb,c,Hollowc, Homesickness a,b,c, Hopelessnessb,c, Horriblec, Horrora,b,c,f, Hostility b,c,h,i,j, Humiliation b,c, Hurta,b,c, Hysteriab, Impatienta,c, Indignantc, Inferiorc, Insecurityb, Insultb,c, Intimidatedh, Iratea,c, Irkeda, Irritation a,b,c,h,j, Isolationb,c, Jealousy a,b,c,e, Jitteryi,j, Joylessa, Jumpyc, Loathingb, Loneliness a,b,c,i, Longingc, Lossc, Lovesicka, Lowa,c, Mada,c, Melancholyb,c, Misery a,b,c,d, Misunderstoodc, Mopingc, Mortificationa,b, Mournfulc, Neglectb,c, Nervousness a,b,c,i,j, Nostalgiac, Offendedh, Oppressedc, Outragea,b,c, Overwhelmeda, Painc, Panic b,c, Petrifieda,c, Pitya,b,c, Puzzledh, Rageb,c,e, Regreta,b,c,e,g, Rejectionb,c, Remorsea,b,c, Reproachfulc, Resentmenta,b,c, Revulsionb, Ridiculousc, Rottenc, Sadness a,b,c,d,e,f,g,h,i, Scared a,c,h,j, Scornb,c,i, Self-consciousc, Shame ,a,b,c,e,g,j, Sheepishc, Shocka,b,c, Shyc, Sickeneda,c, Smallc, Sorrowa,b,c,e,i, Spiteb, Startlede,h, Strainedc, Stupidc, Subduedc, Sufferingb,c, Suspensec, Sympathyb, Tenseness b,c,h, Terriblec, Terrora,b,c, Threatenedh, Tormenta,b,c, Troubledc, Tremulousc, Uglyc, Uneasinessa,b,c, Unfulfilled, Unhappinessa,b,c,i, Unpleasanth, Unsatisfiedc, Unwantedc, Upseta,c,e,j, Vengefulnessb,c, Wantc, Wistfulc, Woeb,c, Worry b,c, Wrathb,c, Yearningc
Acceptancec,h, Accomplishedc, Activei,j, Admirationc, Adorationb,c, Affectionb,c, Agreementc, Alerth,j, Amazementb, Amusementa,b,c, Anticipationb,c, Appreciationc, Ardentc, Arousala,b,d, Astonishmentb,d,i, At easea,d, Attentiveh,j, Attractionb,c, Avidc, Blissb, Bravec, Calm a,d, Caringb,c, Charmeda, Cheerfulnessa,b,c,h, Comfortablec, Compassionb,c, Consideratec, Concernc, Contentment a,b,c,d,I, Courageousc, Curioush, Delighta,b,c,d,h, Desireb,c, Determinedj, Devotionc, Eagernessb,c, Ecstasya,b,c, Elationa,b,c,i, Empathyc, Enchantedc, Encouraging c, Energeticf, Enjoymentb,c,f, Entertainedc, Enthrallmentb, Enthusiasm b,c,e,f,i,j, Euphoriab,c, Excellentc, Excitement a,b,c,d,f,i,j, Exhilarationb,f Expectantc, Exuberantc, Fantasticc, Fascinatede, Finec, Fondnessb,c, Forgivingc, Friendlyc, Fulfillment c, Gaietyb,c, Generousc, Gigglyc, Givingc, Gladnessa,b,c,d, Gleeb,c, Goodc, Gratitudec, Greatc, Happiness a,b,c,d,e,f,h,i, Harmonyc, Helpfulc,h, Highc, Hope b,c,g, Hornyc, Impressedc, Incrediblec, Infatuationb,c, Inspiredj, Interestedf,j, Jollinessb, Jovialityb, Joy a,b,c,e,f,g, Jubilationb,c, Kindlyc,i, Lightheartedc, Likingb,c,g, Longingb, Love a,b,c,e, Lustb,c, Merrimentc, Moveda, Nicec, Optimism b, Overjoyeda,c, Passion a,b,c, Peaceful c,f, Peppyi, Perfectc, Pityc, Playfulc, Pleasure a,c,d,f,i, Pridea,b,c,e,f,g,j, Protectivec, Raptureb, Reassuredc, Regardc, Rejoicec, Relaxedc,d,f, Releasec, Relief a,b,c,e,f,g, Respectc, Reverencec, Romantic c, Satisfactiona,b,c,d,f,i, Securec, Sensationalc, Sensitivec, Sensualc, Sentimentality b,c, Serened,c, Sexy c, Sincerec, Strongi,j, Superc, Surpriseb,e,f,i, Tendernessb,c, Terrificc, Thoughtfulc, Thrill a,b,c, Toucheda, Tranquilityc, Triumphb, Trustc,h, Victoriousc, Warm-hearted c,i, Wonderfulc, Worshipc, Zealb, Zestb
Note: The emotion words of Richins’ CES (1997) are in italics. a Morgan and Heise (1988). b Shaver et al. (1987). c Storm and Storm (1987). d Russell (1980). e Frijda et al. (1989). f Havlena et al. (1989). g Roseman et al. (1996). h Plutchik (1980). i Watson and Tellegen (1985). j Watson et al. (1988).
1440
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
positive and negative affect appears to be the most popular conceptualization (see Table 1).
3. Positive and negative affect Many papers acknowledge that positive and negative affect are bever present in the experience of emotionsQ (Diener, 1999, p. 804; see also Berkowitz, 2000; Watson et al., 1999). We have content-analyzed 10 seminal studies in psychology on emotions and emotion words (Frijda et al., 1989; Havlena et al., 1989; Morgan and Heise, 1988; Plutchik, 1980; Roseman et al., 1996; Russell, 1980; Shaver et al., 1987; Storm and Storm, 1987; Watson and Tellegen, 1985; Watson et al., 1988). We were able to classify all emotion words as either a positive or negative emotion (see Table 2). Table 2 shows the emotion words and indicates which studies included a particular word as a positive or negative emotion word in their structure. The number of references for each emotion word illustrates to what degree researchers agree that this is an emotion word. For example, the emotion words fear, sadness, and happiness appear almost in every emotion structure, whereas others, like mournful, forlorn, and zeal, are only mentioned occasionally. In addition, Table 2 supports the notion that there are more negative than positive emotion words (Morgan and Heise, 1988). Yet, which of these many emotion words should be used to measure consumer emotions? To address this issue, we can use the important study by Richins (1997). Based on extensive research, she constructed the Consumption Emotion Set (CES). This scale includes most, if not all, emotions that can emerge in consumption situations and was developed to distinguish the varieties of emotion associated with different product classes. Table 2 reveals that the words included in the CES (in italics) are among the most frequently encountered words in the psychological emotion literature, and can be easily divided in positive and negative affect. Advantages of the division in positive and negative affect are that (1) the model can be kept simple and (2) the combination of a person’s positive and negative affect is indicative of his/her attitude. The disadvantage is that important distinctions among different positive and negative emotions disappear (Lerner and Keltner, 2000; 2001). Thus, more precise information about the feelings of the consumer is lost (Bagozzi et al., 1999). Because different emotions can have different behavioral consequences, it is important to know, for example, whether a failure in a product or service elicits feelings of anger or sadness. Both angry and sad people feel that something wrong has been done to them, but whereas sad people become inactive and withdrawn, the angry person becomes more energized to fight against the cause of anger (Shaver et al., 1987). Several studies have shown how important it is to take into account differences across emotions of the same valence (Lerner and Keltner, 2000; 2001; Zeelenberg and Pieters, 1999).
4. A hierarchy of consumer emotions The research streams supporting the different emotion structures (positive/negative vs. specific emotions) seem opposing, but can in fact be seen as complementing. Shaver et al. (1987) and Storm and Storm (1987) have suggested that emotions can be grouped into clusters, yielding a hierarchical structure. The most general, superordinate, level consists of positive and negative affect. The next level is considered as the basic emotion level, and the lowest, subordinate, level consists of groups of individual emotions that form a category named after the most typical emotion of that category. Along the lines of the hierarchical structures of Shaver et al. (1987) and Storm and Storm (1987), we thus propose that consumer emotions can be considered at different levels of abstractness. Our hierarchy of consumer emotions distinguishes between positive and negative affect at the superordinate level. The specific consumer emotions based on Richins’ (1997) CES encompass the subordinate level. Which basic emotions should constitute the intermediate level, however, is less clear. Basic emotions are believed to be innate and universal, but because there are different ways to conceive emotions (facial, e.g., Ekman, 1992; biosocial, e.g., Izard, 1992; brain, e.g., Panksepp, 1992), there is also disagreement about which emotions are basic (Turner and Ortony, 1992). Ortony and Turner (1990) have shown that 14 different emotion theorists proposed 14 different sets of basic emotions. Table 3 shows the usage frequency of the basic emotions in the different structures reviewed by Ortony and Turner (1990). With few exceptions, the basic
Table 3 Basic emotions in the psychological literature (adapted from Ortony and Turner, 1990) Acceptancea, Angera,b,c,d,e,f,g,h,i, Anticipationa, Anxietyf,h,j, Aversionb, Contemptd,i, Contentmenth, Courageb, Dejectionb, Desireb,k, Despairb, Disgusta,c,d,e,f,h,i, Distressd,i, Elatione, Expectancyl, Feara,b,c,d,e,g,h,i,l,m,n, Grief m, Guiltd, Happinessf,h,k,o, Hateb, Hopeb, Hostilityh, Interestd,k, Joya,c,d,g,i,j, Likingh, Loveb,g,h,m,n, Painh,p, Panicl, Pleasurep, Prideh, Ragej,l,m,n, Sadnessa,b,c,f,g,h,o, Shamed,h,i, Sorrowk, Subjectione, Surprisea,c,d,i,k, Tendere, Wondere,k a
Plutchik (1980). Arnold (1960). c Ekman et al. (1982). d Izard (1971). e McDougal (1926). f Oatley and Johnson-Laird (1987) g Shaver et al. (1987). h Storm and Storm (1987). i Tomkins (1984). j Gray (1982). k Frijda (1986). l Panksepp (1982). m James (1884). n Watson (1930). o Weiner and Graham (1984). p Mowrer (1960). b
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
emotions from Table 3 are among the most frequently mentioned emotion words in Table 2. To develop a set of basic consumer emotions, we draw on the hierarchical structures of Shaver et al. (1987) and Storm and Storm (1987), and Table 3. Some basic emotion words are mentioned in most of the structures (see Table 3). These are anger, fear, love, sadness, disgust, joy, and surprise. Anger, fear, love and sadness are basic emotions in both the structures of Shaver et al. (1987) and Storm and Storm (1987), and will be retained in our structure. Disgust is not included in the structure of Richins (1997) and therefore excluded as a basic consumption emotion. Surprise was excluded for several reasons. First, it is a neutral emotion (Storm and Storm, 1987) and therefore impossible to classify as a positive or negative emotion. Second, when participants were required to list emotions, surprise was hardly mentioned (Fehr and Russell, 1984). Following Storm and Storm (1987), we added the emotion shame to the basic negative emotions. Anger, sadness, and fear are all emotions elicited by situations caused by others or circumstances, whereas shame is caused by a negative action of consumers themselves (Roseman et al., 1996). The positive emotions can be roughly divided in interpersonal emotions and emotions without interpersonal reference (Storm and Storm, 1987). The interpersonal emotions are covered by love and its specific emotion words, but there are distinct differences between the emotions that are not interpersonal. Following Storm and Storm (1987), we therefore replaced the more general term joy by the basic emotions contentment, happiness, and pride. Contentment is low in arousal and passive, whereas happiness is higher in activity and a reactive positive emotion. Pride, on the other hand, concerns feelings of superiority. Due to these differences, we argue that it is better to include these basic emotions separately rather than all under one large basic emotion of joy. Our proposed hierarchy thus consists of three levels: the superordinate level with positive and negative affect, the basic level with four positive and four negative emotions, and the subordinate level with specific emotions. The final
1441
result can be seen in Fig. 1. Next, we will conduct a preliminary test of our hypothesized structure.
5. Method 5.1. Sample and procedure Data were collected in a nationally representative sample among 645 Dutch consumers using a questionnaire. The market research agency GfK carried out the data collection. Of the respondents, 53.6% were women, 58.3% were responsible for the daily grocery shopping, and 69.1% were the main wage earner of the household. The average household size was 2.39 persons and all levels of education and income were represented. The average age was 48 years and ranged between 16 and 91 with a fairly normal spread. Respondents were asked to indicate to what extent they experience 33 specific emotions for one (randomly assigned) type of food (genetically modified food, functional food, organic food, or regular food). Thus, we measure emotions at a general, product-type level of categorization. In The Netherlands, these types of foods are widely known, the exception being functional foods (this was confirmed in discussions with industry experts). Therefore, respondents who rated their emotions for functional foods received additional explanation: bFunctional foods are food products that have been enriched or modified. The reason for this is to make the product healthier or to prevent diseases (e.g., milk with extra calcium, margarine with additives to lower the cholesterol level)Q. 5.2. Measures With some exceptions, the emotion words shown in Fig. 1 were used. Emotions were rated on a five-point Likert scale ranging from I feel this emotion not at all (1) to I feel this emotion very strongly (5). In our empirical test, we omitted the basic emotions bloveQ and bprideQ, and the
Fig. 1. Hierarchy of consumer emotions.
1442
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
emotion words benviousQ and bjealousQ. bLoveQ is demonstrated to be mainly experienced in the case of sentimental products, like mementos and gifts (Richins, 1997). The latter three emotions are interpersonal and less applicable in the case of widely available food. The emotion bprideQ generally occurs when a consumer feels superior compared to another person, whereas the emotions benvyQ and bjealousyQ occur when consumers feel that another person has something more or better than them. Thus, the basic emotions in our analyses are as follows: anger, fear, sadness, shame, contentment, and happiness, measured in total by 33 specific emotion words. 5.2. Stability of the emotions structure across food types Before we can test our second-order hierarchical model of consumer emotions, we have to establish whether we can pool the data across the four food types. We do this in two ways. First, we assess whether principal component analysis yields the same factor structure in each of the four food groups. The Bartlett’s test of sphericity is significant for all four foods, and the measure of sampling adequacy ranges between .86 (organic food) and .92 (genetically modified food), which means that principal component analysis can be applied. The scree test indicated two factors in all four groups, explaining between 48% (regular food) and 60% (genetically modified food) of the variance. The factor structures (after rotation) were highly similar, Tucker’s congruence coefficient always being greater than .95 ( P b.01; Cattell, 1978). A second way to assess the similarity of the four food groups is to test for the invariance of the covariance matrices across the four groups using LISREL 8.50 (Steenkamp and Baumgartner, 1998). The fit was good, given the large sample and high number of degrees of freedom (Baumgartner and Homburg, 1996): v 2(1683)=3845.90 ( Pb.001); CFI=.86; TLI=.82. Hence, we can pool the data across the different food types.
6. Results 6.1. Testing the proposed model We used LISREL 8.50 to test the proposed hierarchical emotions model. The standardized parameter estimates of the second-order factor analysis are reported in Fig. 2. Model fit is acceptable: v 2(490)=3036.79 ( Pb.001), CFI=.84, TLI=.83. Although the v 2 was highly significant (not unexpected given the large sample size; Anderson and Gerbing, 1988), other indicators suggest reasonable model fit, especially considering that fit is adversely affected by model complexity (Baumgartner and Homburg, 1996; Bollen, 1989; Bone et al., 1989). In addition, the fit measures are in line with simulation results (see Gerbing and Anderson, 1993 for a review) and compare favorably to
other models with similar degrees of freedom (e.g., Netemeyer et al., 1991; Richins and Dawson, 1992; Wong et al., 2003). All factor loadings were significant at P b.001, the average loading being .73. Only the factor loading of the emotion nostalgia on the basic emotion sadness was below .40. A possible explanation for this is that nostalgia involves complex emotional responses and can have both a positive and a negative connotation (Holak and Havlena, 1998). The correlation between the second-order factors positive and negative affect was significant (r= .35, Pb.01), confirming earlier results found in consumer research (e.g., Westbrook, 1987; Phillips and Baumgartner, 2002). These results support the convergent and discriminant validity of our model (Steenkamp and Van Trijp, 1991). The reliability of our measures was high. Cronbach alphas were a=.94 and a=.95 for the dimensions positive and negative affect, respectively. The basic emotions yielded the following reliabilities: anger (a=.88), fear (a=.88), sadness (a=.76), shame (a=.74), contentment (a=.86), and happiness (a=.92). 6.2. Comparison of the superordinate level with the basic emotions Although the emotion structure is similar for the four food groups, that does not imply that the various foods evoke the same emotional intensity. Table 4 provides the mean scores for the superordinate dimensions positive and negative affect and for the basic emotions. ANOVA with multiple comparisons (LSD) was used to investigate whether the mean values across food groups are significantly different. Participants experience significantly more negative affect and less positive affect for genetically modified foods than for the other food groups. Yet, the basic emotions show differences among the food types that would have been lost if only positive and negative affect had been considered. Both the basic emotions fear and contentment contain additional subtle distinctions across the food groups. The negative affect experienced by consumers is similar for functional, organic, and regular food. Yet, consumers feel a lot more fearful concerning functional food than for organic and regular food. Concerning the positive emotions, contentment has very low values for organic food compared to functional and regular food. These nuances, however, are wiped away for positive affect. To demonstrate the usefulness of basic emotions for understanding the consumer’s feelings, we will take a closer look at one of the food groups. Genetically modified food represents a controversial topic in contemporary society, and previous research (e.g., Bredahl, 2001) has shown that consumers have a rather negative attitude towards this type of food. The scores on negative and positive affect support this, but the basic emotions indicate more clearly how consumers feel. Participants do not feel sad or ashamed, but
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
1443
Fig. 2. Results of second-order factor analysis.
are very angry and afraid. This means that they feel energized and powerful rather than inactive, and feel that they themselves are not to be blamed, but someone else is.
In addition, genetically modified food elicits strong associations of risk and uncertainty leading to feelings of fear.
Table 4 Differences in the intensity of the superordinate and basic emotions for the food groups Emotion
GMF
Functional
Organic
Regular
F
P value
Negative affect Anger Sadness Fear Shame Positive affect Contentment Happiness
1.99a 2.19a 1.79a 2.16a 1.65a 1.68a 1.82a 1.64a
1.45b 1.51b 1.46b 1.57b 1.32b 2.41b,c 2.69b 2.32b
1.43b 1.47b 1.47b 1.40c 1.29b 2.32c 2.40c 2.29b
1.46b 1.55b 1.47b 1.43c 1.31b 2.48b 2.81b 2.37b
31.25 34.49 11.99 46.06 11.30 40.09 47.38 33.64
b.001 b.001 b.001 b.001 b.001 b.001 b.001 b.001
Note: Different supercripts reflect a significant difference of the intensity at a p-value b0.05.
1444
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445
7. Conclusion
References
Based on our literature review, we concluded that despite the different ways to measure emotions, positive and negative affect are frequently employed as general emotion dimensions. Important nuances, however, are lost if emotions of the same valence are collapsed together. This paper therefore proposed a hierarchical model of consumer emotions (Fig. 1) to integrate the different research streams concerning emotion content and structure. This model specifies emotions at three levels of generality. At the superordinate level, it distinguishes between positive and negative affect. This is generally considered to be the most abstract level at which emotions can be experienced (e.g., Berkowitz, 2000; Diener, 1999). At the level of basic emotions, we specify four positive (contentment, happiness, love, and pride) and four negative (sadness, fear, anger, and shame). At the subordinate level, we distinguish between 42 specific emotions based on Richins’ (1997) CES. Our empirical study provides support for the proposed model and suggests that the basic emotions allow for a better understanding of the consumers’ feelings concerning certain food products compared to only positive and negative affect. Note that not in all situations this model need be used as a whole. Dependent on the research question, only part of the model may be used. However, even in such cases, the researcher can still relate his/her specific results to the broader structure of our emotions. This makes it easier for emotions research to cumulatively build on each other and to identify gaps in our knowledge. Our study has several limitations, which offer avenues for future research. First, we excluded two basic emotions (love and pride) from our empirical analysis. Future research is needed to validate the whole hierarchy of emotions, and to test our model on other products and services. Second, future research can expand the set of specific consumer emotions. Possible candidates include the negative emotions regret and disappointment that recently received a great deal of attention in consumer research (e.g., Inman and Zeelenberg, 2002; Tsiros and Mittal, 2000; Zeelenberg and Pieters, 1999). Regret stems from bad decisions, whereas disappointment originates from disconfirmed expectancies (Zeelenberg and Pieters, 1999). We thus propose that regret can be positioned under the basic emotion shame, and disappointment under the basic emotion sadness (Zeelenberg et al., 1998), but future research should investigate this. Third, future research can investigate whether the set of basic emotions has greater explanatory power than positive and negative affect. Our exploratory analysis indicates this, but future research should test this hypothesis. Fourth, we tested our emotions model in The Netherlands. The further advancement of consumer research as an academic discipline requires that the validity of our theories and measures and their degree of general validity and boundary conditions be tested in different countries (Steenkamp and Burgess, 2002).
Anderson James C, Gerbing David W. Structural equation modeling in practice: a review and recommended two-step approach. Psychol Bull 1988;103(3):411 – 23. Bagozzi Richard P, Gopinath Mahesh, Nyer Prashanth U. The role of emotions in marketing. J Acad Mark Sci 1999;27(2):184 – 206. Baumgartner Hans, Homburg Christian. Applications of structural equation modeling in marketing and consumer research: a review. Int J Res Mark 1996;13(2):139 – 61. Berkowitz Leonard. Causes and consequences of feelings. New York7 Cambridge University Press; 2000. Bollen Kenneth A. Structural equations with latent variables. New York7 Wiley; 1989. Bone Paula Fitzgerald, Sharma Subhash, Shimp Terence. A bootstrap procedure for evaluating goodness-of-fit indices of structural equation and confirmatory factor models. J Mark Res 2003;26(1):105 – 11. Bougie Roger, Pieters Rik, Zeelenberg Marcel. Angry customers don’t come back, they get back: the experience and behavioral implications of anger and dissatisfaction in services. Journal of the Academy of Marketing Science; 2003;31(4):377 – 393. Bredahl Lone. Determinants of consumer attitudes and purchase intentions with regard to genetically modified foods—results of a cross-national survey. J Consum Policy 2001;24:23 – 61. Cattell RB. The scientific use of factor analysis in behavioral and life sciences. New York7 Plenum; 1978. Derbaix Christian M. The impact of affective reactions on attitudes toward the advertisement and the brand: a step toward ecological validity. J Mark Res 1995;32(4):470 – 9. Derbaix Christian M, Vanhamme Joelle. Inducing word-of-mouth by eliciting surprise—a pilot investigation. J Econ Psychol 2003;24: 99 – 116. Diener Ed. Introduction to the special section on the structure of emotion. J Pers Soc Psychol 1999;76(5):803 – 4. Dube Laurette, Morgan Michael S. Capturing the dynamics of in-process consumption emotions and satisfaction in extended service transactions. Int J Res Mark 1998;15:309 – 20. Dube Laurette, Cervellon Marie-Cecile, Jingyuan Han. Should consumer attitudes be reduced to their affective and cognitive bases? Validation of a hierarchical model. Int J Res Mark 2003;20:259 – 72. Edell Julie A, Burke Marian C. The power of feelings in understanding advertising effects. J Consum Res 1987;14(4):421 – 33. Edson Escalas Jennifer, Stern Barbara B. Sympathy and empathy: emotional responses to advertising dramas. J Consum Res 2003;29(4):566 – 78. Ekman Paul. Are there basic emotions? Psychol Rev 1992;99(3):550 – 3. Fehr Beverly, Russell James A. Concept of emotion viewed from a prototype perspective. J Exp Psychol Gen 1984;113:464 – 86. Frijda Nico H, Kuipers Peter, Ter Schure Elisabeth. Relations among emotion, appraisal, and emotional action readiness. J Pers Soc Psychol 1989;57(2):212 – 28. Gerbing David W, Anderson James C. Monte Carlo evaluations of goodness-of-fit indices for structural equation models. In: Bollen Kenneth A, Long J. Scott, editors. Testing structural equation models. Newbury Park, CA7 Sage; 1993. p. 40 – 65. Havlena William J, Holbrook Morris B, Lehmann Donald R. Assessing the validity of emotional typologies. Psychol Market 1989;6:97 – 112. Holak Susan L, Havlena William J. Feelings, fantasies, and memories: an examination of the emotional components of nostalgia. J Bus Res 1998;42(3):217 – 27. Holbrook Morris B, Batra Rajeev. Assessing the role of emotions as mediators of consumer responses to advertising. J Consum Res 1987; 14(3):404 – 20. Holbrook Morris B, Gardner Meryl P. An approach to investigating the emotional determinants of consumption durations: why do people consume what they consume for as long as they consume it? J Consum Psychol 1993;2(2):123 – 42.
F.J.M. Laros, J.-B.E.M. Steenkamp / Journal of Business Research 58 (2005) 1437–1445 Holbrook Morris B, Hirschman Elizabeth C. The experiential aspects of consumption: consumer fantasies, feelings, and fun. J Consum Res 1982;9(2):132 – 40. Inman Jeffrey J, Zeelenberg Marcel. Regret in repeat purchase versus switching decisions: the attenuating role of decision justifiability. J Consum Res 2002;29(1):116 – 28. Izard Carroll. Human emotions. New York7 Plenum; 1977. Izard Carroll. Basic emotions, relations among emotions, and emotion– cognition relations. Psychol Rev 1992;99(3):561 – 5. Lerner Jennifer S, Keltner Dacher. Beyond valence: toward a model of emotion-specific influences on judgment and choice. Cogn Emot 2000;14(4):473 – 93. Lerner Jennifer S, Keltner Dacher. Fear, anger and risk. J Pers Soc Psychol 2001;81(1):146 – 59. Mano Haim. The structure and intensity of emotional experiences: method and context convergence. Multivariate Behav Res 1991;26(3):389 – 411. Mano Haim, Oliver Richard L. Assessing the dimensionality and structure of the consumption experience: evaluation, feeling, and satisfaction. J Consum Res 1993;20(3):451 – 66. Mehrabian Albert, Russell James A. An approach to environmental psychology. Cambridge, MA7 MIT Press; 1974. Morgan Rich L., Heise David. Structure of emotions. Soc Psychol Q 1988;51(1):19 – 31. Netemeyer Richard G, Durvasula Srinivas, Lichtenstein Donald. A crossnational assessment of the reliability and validity of the CETSCALE. J Mark Res 1991;28(3):320 – 7. Nyer Prashanth U. A study of the relationships between cognitive appraisals and consumption emotions. J Acad Mark Sci 1997;25(4): 296 – 304. Oliver Richard L. Cognitive, affective, and attribute bases of the satisfaction response. J Consum Res 1993;20(3):418 – 30. Olney Thomas J, Holbrook Morris B., Batra Rajeev. Consumer responses to advertising: the effects of ad content, emotions, and attitude toward the ad on viewing time. J Consum Res 1991;17(4):440 – 53. Ortony Andrew, Turner Terence J. What’s basic about basic emotions? Psychol Rev 1990;97(3):315 – 31. Panksepp Jaak. A crititical role for baffective neuroscienceQ in resolving what is basic about basic emotions. Psychol Rev 1992;99(3):554 – 60. Phillips Diane M, Baumgartner Hans. The role of consumption emotions in the satisfaction response. J Consum Psychol 2002;12(3):243 – 52. Plutchik Robert. Emotion: a psychoevolutionary synthesis. New York7 Harper and Row; 1980. Richins Marsha L. Measuring emotions in the consumption experience. J Consum Res 1997;24(2):127 – 46. Richins Marsha L, Dawson Scott. A consumer values orientation for materialism and its measurement: measure development and validation. J Consum Res 1992;19(3):303 – 16. Roseman Ira J, Antoniou Ann A, Jose Paul J. Appraisal determinants of emotions: constructing a more accurate and comprehensive theory. Cogn Emot 1996;10(3):241 – 77. Russell James A., Weiss A., Mendelsohn G.A.. Affect grid - A single-item scale of pleasure and arousal. J Pers Soc Psychol 1989;57(3):493 – 502. Russell James A. A circumplex model of affect. J Pers Soc Psychol 1980; 39(6):1161 – 78. Ruth Julie A, Brunel Frederic F, Otnes Cele C. Linking thoughts to feelings: investigating cognitive appraisals and consumption emotions in a mixed-emotions context. J Acad Mark Sci 2002;30(1):44 – 58.
1445
Shaver Philip, Schwartz Judith, Kirson Donald, O’Conner Cary. Emotion knowledge: further exploration of a prototype approach. J Pers Soc Psychol 1987;52:1061 – 86. Smith Amy K, Bolton Ruth N. The effects of customers’ emotional responses to service failures on their recovery effort evaluations and satisfaction judgments. J Acad Mark Sci 2002;30(1):5 – 23. Smith Craig A, Lazarus Richard S. Appraisal components, core relational themes, and emotions. Cogn Emot 1993;7:295 – 323. Steenkamp Jan-Benedict EM, Baumgartner Hans. Assessing measurement invariance in cross-national consumer research. J Consum Res 1998; 25(1):78 – 90. Steenkamp Jan-Benedict E.M., Burgess Steven M.. Optimum stimulation level and exploratory consumer behavior in an emerging consumer market. Int J Res Mark 2002;19:131 – 50. Steenkamp Jan-Benedict EM, Van Trijp Hans CM. The use of LISREL in validating marketing constructs. Int J Res Mark 1991;8(4):283 – 99. Steenkamp Jan-Benedict EM, Baumgartner Hans, Van der Wulp Elise. The relationships among arousal potential, arousal and stimulus evaluation, and the moderating role of need for stimulation. Int J Res Mark 1996;13:319 – 29. Stephens Nancy, Gwinner Kevin P. Why don’t some people complain? A cognitive–emotive process model of consumer complaint behavior. J Acad Mark Sci 1998;26(4):172 – 89. Storm Christine, Storm Tom. A taxonomic study of the vocabulary of emotions. J Pers Soc Psychol 1987;53:805 – 16. Taylor Shirley. Waiting for service: the relationship between delays and evaluations of service. J Mark 1994;58(2):56 – 69. Tsiros Michael, Mittal Vikas. Regret: a model of its antecedents and consequences in consumer decision making. J Consum Res 2000;26(4):401 – 17. Turner Terence J, Ortony Andrew. Basic emotions: can conflicting criteria converge? Psychol Rev 1992;99(3):566 – 71. Verbeke Willem, Bagozzi Richard P. Exploring the role of self- and customer-provoked embarrassment in personal selling. Int J Res Mark 2003;20:233 – 58. Watson David, Tellegen Auke. Toward a consensual structure of mood. Psychol Bull 1985;98:219 – 35. Watson David, Clark Lee Anna, Tellegen Auke. Development and validation of brief measures of positive and negative affect: the PANAS scales. J Pers Soc Psychol 1988;54:1063 – 70. Watson David, Wiese David, Vaidya Jatin, Tellegen Auke. The two general activation systems of affect: structural findings, evolutionary considerations, and psychobiological evidence. J Pers Soc Psychol 1999; 76(5):820 – 38. Westbrook Robert A. Product/consumption-based affective responses and post-purchase processes. J Mark Res 1987;24:258 – 70. Wong Nancy, Rindfleisch Aric, Burroughs James E. Do reverse-worded items confound measures in cross-cultural consumer research: the case of the material values scale. J Consum Res 2003;30(1):72 – 91. Zeelenberg Marcel, Pieters Rik. Comparing service delivery to what might have been. J Serv Res 1999;2(1):86 – 97. Zeelenberg Marcel, Van Dijk Wilco W, Manstead Antony SR, Van der Pligt Joop. The experience of regret and disappointment. Cogn Emot 1998;12(2):221 – 30.
Journal of Business Research 58 (2005) 1446 – 1455
Four characters on the stage playing three games: performing arts consumption in Spain ´ lvarezb Jordi Lo´pez Sintasa,*, Ercilia Garcı´a A a
Dept. d’Economia de lVEmpresa, Facultat d’Economia, Universitat Auto`noma de Barcelona, Edifici B, 08193 Bellaterra, Barcelona, Spain b IESE Graduate Business School, Universidad de Navarra, Spain Received 1 November 2002; received in revised form 1 September 2003; accepted 1 October 2003
Abstract There are presently three competing hypotheses about the symbolic role played by consumption: the distinction, the boundary-effacement, and omnivore effects. Our research design follows another recently proposed alternate view in which all three effects are working simultaneously, but to differing degrees, and introduces several innovations. We found that in the Spanish performing arts symbolic space, the stage has four characters: sporadic, popular, snob, and omnivorous consumers, playing three symbolic games, namely, boundary-effacement, omnivore, and distinction, but to differing degrees. That is, we found (1) weak evidence for a few popular performing arts events liked by everyone, but not necessarily all lowbrow performances, (2) strong evidence favoring the hypothesis that consumers from the highest social class consume all arts at higher rates than everyone else do, and (3) moderate evidence that social classes use the performing arts’ space of consumption to symbolize status differences. D 2004 Elsevier Inc. All rights reserved. Keywords: Performing arts; Sociology of consumption; Distinction effect; Symbolic roles; Omnivore effect
1. Segmentation of consumers according to lifestyle The symbolic properties of products have been widely acknowledged since Martineau’s pioneer study of the sociology of marketing (Martineau, 1958). This has led to a great deal of interest in marketing to assess the ability of social class indicators to discriminate among patterns of consumption (Curtis, 1972; Schaninger, 1981; Wasson, 1969), with this interest persisting to this day (Dawson et al., 1990; Sivadas et al., 1997). Nevertheless, these findings seem to offer conflicting results, depending on the indicators used to judge consumers’ social class and the product categories researched. In this study, we draw some conclusions from these varied results by theoretically grounding our research design. There are currently three competing hypotheses
about the role played by consumption: the distinction, boundary-effacement, and omnivore effects. Our research design, nevertheless, follows the alternative view of Holbrook et al. (2002), which proposes that all three effects are working simultaneously but to differing degrees and introduces several innovations. In contrast with previous research, ours is the first that (a) is based on individuals’ rate of performing arts consumption rather than geodemographic clusters, (b) uses additional cultural categories, as well as different social class indicators, (c) simultaneously finds patterns of consumption and identifies the relationship to consumer’s social position, and lastly, (d) evaluates whether the three competing hypotheses are operating simultaneously in the performing arts symbolic space.
2. Theoretical framework * Corresponding author. E-mail address: [email protected] (J. Lo´pez Sintas). 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.10.013
Social classes can be characterized as groups of agents who are subject to similar conditions of existence (habitus)
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
and conditioning factors and, as a result, are endowed with similar preferences and, therefore, similar practices or lifestyles (Bourdieu, 1987; O’Shaughnessy, 1987). These similar positions in social space are produced by different levels and types of capital, both economic and cultural (Bourdieu, 1998[1979]). Thus, consumers who occupy neighboring positions in this social space are therefore subject to similar conditioning factors: They have many chances of having similar dispositions and interests, which leads to similar patterns of consumption. Here, we make the distinction between consumers’ social space position and consumers’ social class. The latter is indicated by the levels and types of economic and cultural capital, as suggested by the sociology of consumption (Bourdieu, 1998[1979]; DiMaggio and Useem, 1978), whereas the former focuses on the consumer’s social class and other cultural categories (McCracken, 1986). By judging the symbolic properties of consumption spaces (of the performing arts, in this case), we ought to search for consumers’ consumption patterns and see how they are associated to the consumers’ social class and other cultural categories (Boter and Wedel, 1999; van Rees et al., 1999). Below, we take a look at those symbolic properties. 2.1. The distinction effect The sociological analysis of consumption proposes that lifestyles be stratified according to social class (DiMaggio and Useem, 1978; Bourdieu, 1998[1979], 1987). According to Bourdieu, social hierarchy is translated into, and misrecognized as, cultural symbols and lifestyles inherent in individuals through the mediating structure of habitus. The habitus space includes schemes of perception and the evaluation of daily habits, cognitive structures, and evaluation procedures that are acquired from repetitive experience with the consumer’s conditions and position in the social space (see Levy’s pioneer studies on the symbolic properties of consuming the arts, compiled by Roock, 1999). The habitus space is also called the corpus of cultural principles (McCracken, 1986), as they explain the ideas or values that clarify the way in which social classes discriminate between them. People internalize the habitus space, determining cultural, as well as material, choices that reproduce the very class structure itself. choices that reproduce the very class structure itself. The main proposition derived from Bourdieu’s theory, as well as that of DiMaggio and Useem (1978), is that the symbolic space of consumption will be segmented in a homologous way as the society is stratified (Bourdieu, 1998[1979]; DiMaggio and Useem, 1978), i.e., that consumption symbolizes status. Nearly all research conducted until now claiming support for or casting doubts on the distinction hypothesis has used binary indicators of consumption. To our knowledge, the current dispute about the ability of social class indicators to discriminate among
1447
patterns of consumption is based on the use of binary indicators. For the sake of clarity, we have produced Panel A in Fig. 1, where L1 and L2 stand for two different lowbrow cultural products and H1 and H2 for two highbrow cultural products, D and U for downscale and upscale social class, respectively, and the vertical axis for consumption rate. If we analyze the consumption of lowbrow (L) and highbrow (H) products, basing our judgment on whether the different social classes (D and U) consume each product at least once a year, it would appear that no one uses consumption to differentiate themselves from other consumers because everyone consumes each product at least once. If we take into account how much they consume each product, however, we would see the opposite pattern shown in Panel A. Both lowbrow and highbrow products have a high consumption marginal mean, but with an opposite pattern: Downscale (upscale) consumers really do prefer lowbrow (highbrow) products to highbrow (lowbrow). Thus, intuitively, we see that, although everyone consumes everything, they do so to differing degrees. 2.2. The boundary-effacement or homogeneity effect Both the cultural theorists—particularly those associated with the Frankfurt School—and the postmodernists have foreseen an increase in the consumption of popular cultural products. The former were preoccupied with the rise in the consumption of cultural products offered by the so-called cultural industries (Adorno, 1991; Morin, 1967). These authors pointed out the possibility that the rise of market could greatly erode upscale social classes’ preferences for highbrow cultural products, particularly opera, classical music, ballet/dance, etc., as cultural industries intentionally integrate its consumers, both from high culture to popular (Adorno, 1991, p. 85). The end result is that the status of popular culture is raised, and the status of high culture is lowered (Morin, 1967, p. 65), obscuring social differences through the objective homogeneity of popular mass culture (everyone consumes it), thereby making social class differences invisible. Postmodernists, nevertheless, have a positive view of cultural disorder. The blurring distinctions between levels of culture (high vs. popular) give way to a glutinous mass that stimulates and plays with the overproduction of signs (Featherstone, 1992, p. 271). Consumption, then, is embedded within systems of signification, of making and maintaining distinctions (Firat and Venkatesh, 1995, p. 249). Consequently, it is expected that the salience of social class factors is declining and that consumption patterns are becoming more fragmented (Clark and Lipset, 1991; Featherstone, 1992; Firat and Venkatesh, 1995; Gartman, 1991). These authors, then, suggest that we look for other cultural categories to understand the way in which consumption is stratified—here, we will turn our attention towards gender, age, and marital status to control its effect.
1448
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
Fig. 1. Boundary-effacement, omnivore, and distinction effects only and all simultaneously. (A) Distinction effect only. (B) Boundary-effacement effect only. (C) Omnivore effect only. (D) All effects operating simultaneously.
The problem we now encounter is how to test the proposition that the consumption of highbrow cultural products decreases as the consumption of mass culture products increases. Again, we have to distinguish between how many products are consumed and how much they are consumed. In this research, we take mass culture products to mean popular cultural products, pop music, flamenco shows, folk music, and jazz concerts (also potentially including films, recorded music, video productions, and, in general, all products classified as part of the cultural industries). If the homology effect works, we should find that everyone does indeed prefer lowbrow or popular cultural products to highbrow, as illustrated in Fig. 1, Panel B: In fact, upscale, as well as downscale, consumers do consume lowbrow cultural products at higher rates than highbrow products. Nevertheless, upscale social class consumers’ rate of consumption is higher than of the downscale. 2.3. The omnivore effect Another more recent line of research, however, has also cast doubts on the homology thesis between high position in social hierarchy and the consumption of high culture products. DiMaggio (1987, p. 444) has proposed
that the variety of cultural products that a person consumes is a function of his or her socioeconomic status (SES). Therefore, although the upper class clearly has more knowledge of, and participates more frequently in, high culture, research has also consistently shown that its members participate in the popular culture as well, and often at levels equivalent to the lower classes (Peterson and Kern, 1996; Peterson, 1992). That is, upper classes show an omnivorous pattern of consumption known as an omnivore effect (Holbrook et al., 2002). Social class status is gained not only by consuming prestigious forms of art, but also by showing off one’s cultural knowledge in a wide variety of genres. As a result, no rigid class boundaries appear to exist in the consumption of popular and high culture. Therefore, the omnivorous consumption thesis proposes that all persons (or at least a group of consumers) consume everything, although at different levels. If this hypothesis is working in isolation, we should find (1) no differences between the level of consumption of lowbrow and highbrow products and (2) differences between downscale and upscale rate of consumption, as in Panel C in Fig. 1; that is, we should see that upscale social classes consumption is always higher than that of downscale consumers.
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
2.4. All effects simultaneously operating but at different degrees Until the work of Holbrook et al. (2002), these three effects on consumption had been viewed as in competition. In contrast, these authors proposed a research design in which all three effects are working simultaneously, as shown in Panel D of Fig. 1: The distinction effect operates through the interaction between the type of consumer and the type of product (the lines are not parallel); there is a boundary-effacement effect as all consumers prefer lowbrow to highbrow cultural products (consumed at higher rates by all consumers); and, finally, the omnivore effect also works as everyone consumes every product but at different rates. Consequently, researchers should focus on measuring the extent to which each effect is simultaneously working with the other two.
3. Research Design To measure the extent to which all three effects are operating simultaneously, we should proceed in two steps: First, we ought to look for consumers’ patterns of consumption and see how they are related to their social class’s indicators and other cultural categories as well; and second, given the estimated patterns of consumption, we should test the degree to which the distinction, boundaryeffacement, and omnivore effects work simultaneously in the consumption of performing arts. 3.1. Sample Data were obtained from the bHabits of Cultural ConsumptionQ survey requested by the Sociedad General de Autores Espan˜ oles (SGAE) in 1998. This survey conducted home interviews of over 12,000 individuals of men and women, 14 years old and over, living in Spain, through four quarter waves of about 3000 interviewees, comprising a random sample, stratified by autonomous regions and municipalities according to size. Further technical characteristics of the sample are described in the survey report (SGAE, 2000). 3.2. Indicators of performing arts attendance Nine performing arts indicators were analyzed, with some of them classified as popular or lowbrow cultural products (pop music, flamenco shows, folk music, and jazz concerts) and others as highbrow cultural products (classical music, opera, light opera, ballet/dance, and theater). Interviewees were asked how often they went to each performing arts category within the past 12 months. Attendance to classical music, opera, light opera, ballet, and theater was originally recorded as a six-level ordered factor, and going to pop music concerts, flamenco shows,
1449
folk music events, and jazz concerts, as a seven-level ordered factor (see SGAE, 2000, for the details). To find the patterns of consumption, we recoded the answers and reduced them to only two levels: attendance and nonattendance during the last 12 months. Nevertheless, to disentangle the functioning of all three effects working simultaneously, we were also interested in the rate of consumption (how much), not only the pattern of consumption (how many). Consequently, the ordered factors were transformed into ratio metric variables using the central value of the interval in the case of closed intervals, or the lower limit for the rest. 3.3. Indicators of social class To approximate consumers’ social class, we used three indicators: SES and educational and income level. In the first case, we employed the Erikson–Goldthorpe procedure (EGP; Evans, 1992), as it is being used to analyze the relation between social class and cultural consumption (Katz-Gerro and Shavit, 1998). We transformed the original SGAE codification into an EGP class classification with five SES categories plus two additional categories that account for students and the unemployed. For the educational and income levels, we created five categories for each factor (see the Findings section for details). 3.4. Indicators of other cultural categories Many authors have found a generational effect in the consumption of cultural products (Holbrook, 1993). To control for this effect, we formed five age categories. Finally, two categories were used for consumers’ gender, and four for marital status (see results for further details). 3.5. Analytical procedure 3.5.1. Step 1 To define our latent class cluster model (Lazarsfeld and Henry, 1968), we used y 1, y 2,. . ., y 9 to denote our nine indicators of cultural activities, and the bold symbol y for the entire set of indicators. S was any possible combination that our nine indicators could produce, each one symbolized as y s . The six independent cultural categories were denoted as z 1, z 2, . . ., z 6, with z representing the entire set of covariates. A latent class cluster model with active covariates looks for a latent variable, x, that splits the original sample in T clusters or classes, such that the original association observed in the whole sample between the indicators of performing arts attendance, in this case, is removed from the clusters estimated, and, at the same time, the T clusters are also homogeneous in their independent cultural categories (see Magidson and Vermunt, 2001, for further details). Additionally, to make us sure that the same patterns of consumption characterize the T latent classes in the four quarter waves, we introduced a grouping variable
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
1450
for the quarter waves, v, with four levels and tested this reduced model (measurement homogeneity) against the general one (measurement heterogeneity), with the evidence favoring the reduced model (see McCutcheon, 1987. Due to space restrictions, the appendix for the model selection process will be sent upon request). We then tested the distributional homogeneity of the T latent classes among the four waves, but found no support. As a result, the reduced model estimated was the following:
p ys jv; z ¼
T X t¼1
pð xjv; zÞ
9 Y
pðyis jx; zÞ:
i¼1
3.5.2. Step 2 To analyze the distinction, boundary-effacement, and omnivore effects, we used a mixed-effects general linear model to analyze a repeated-measures design (Rutherford, 2001) to explain the rate of consumption. We modeled the performing arts rate of consumption depending on (1) a within-subjects fixed effect, a j , which is the factor assessing the boundary-effacement effect, (2) an external consumption pattern factor that analyzes the omnivore effect, b k , and (3) an interaction between both factors that evaluates the distinction effect, ab jk . Finally, as we wanted to generalize our result to the Spanish population, we modeled the consumer’s interaction with the types of performing arts as a factor with random effects, a ij . Our final model was the following: yijk ¼ l þ aij þ aj þ bk þ abjk þ eijk :
4. Findings 4.1. Step 1: Lifestyles and consumers’ social space position 4.1.1. Patterns of performing arts attendance Table 1 shows the estimates of the parameters for the four-cluster model with concomitant variables, but only displays the association of the performing arts indicators with each cluster—later, we show the relationship of clusters with cultural categories (the model parameters have been estimated with the LatentGold program, see Vermunt and Magidson, 2000). The first row presents how many attendees have been classified in each cluster, i.e., its relative size, and the following rows indicate the probability of belonging to a particular performing arts indicator, given one’s classification in that cluster; all values are expressed as percentages. For instance, if an interviewee has been assigned to Cluster 1, he or she has a probability of almost 99% of not going to any concert of classical music, and therefore, a 1% chance of having gone to one live performance of classical music. On the other hand, if the interviewee has been allocated to Cluster 4, his or her probability of going to classical music concerts is expected
Table 1 Patterns of consumption according to performing arts attendance (probabilities of attendance are in percentage) Indicators Cluster size Classical music Attend Opera Attend Light opera Attend Ballet/dance Attend Theater Attend Pop music Attend Flamenco Attend Folk music Attend Jazz Attend
Cluster 1
Cluster 2
Cluster 3
Cluster 4
Sporadic
Popular
Snobs
Omnivorous
69
20
8
3
1
5
58
63
0
0
18
23
0
0
21
17
0
0
13
37
10
25
70
85
0
83
0
85
2
9
5
32
2
7
10
41
0
6
5
46
25 25 25 25
25 25 25 25
34 27 22 17
Distribution of quarter waves 1 24 2 25 3 25 4 26
to be higher, 37%, and a 63% possibility of not going. Based on these conditional probabilities, we can characterize the probabilistic behavior of the Spaniards regarding this set of performing arts indicators. The estimated model suggests that in the Spanish population, on average, there are four patterns of consumption or lifestyles as far as performing arts attendance is concerned. There is a very large segment, Cluster 1 (which accounts for 69% of the population), that spends very little of its leisure time attending or going to any performance, whether it is high or low culture. For this reason, this group has been labeled the sporadic class of consumers. The second cluster has been designated the popular class of consumers due to their lifestyle: They have a consistent pattern of attending popular performances, but almost no attendance at highbrow performances. The popular class of consumers accounts for 20% of the Spanish population. Cluster 3 has been labeled the snob class of consumers to facilitate the discussion of our results with those published by other researchers. This group of consumers accounts for almost 8% of Spaniards who attend the performing arts. Its pattern of consumption is consistent with attendance at the highbrow performances, particularly classical music concerts, light opera (the highest probability among the four clusters), opera, and ballet/dance. Finally, consumers of the omnivorous class, Cluster 4, display an insatiable demand for a variety of cultural genres, as much for high culture types as for popular ones. Nevertheless, the group is small
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455 Table 2 Patterns of consumption and their cultural categories (probabilities are in percentage)
Mean probability Variables Social class Service 1a Service 2b Nonmanual Entrepreneurs SMEs Manual Not employed Students Age Less than 25 years old 25–34 35–44 45–54 More than 54 years old Gender Male Female Educational level Primary or less Low secondary High secondary College University degree Monthly income Less than 600o ]600o, 900o] ]900o, 1200o] ]1200o, 1800o] More than 1800o Marital status Single Married Widowed Divorced
Cluster 1
Cluster 2
Cluster 3
Cluster 4
Sporadic
Popular
Snobs
Omnivorous
69
20
8
3
35 52 62 69
21 25 28 17
29 14 5 10
15 9 5 4
70 84 41
24 7 53
3 8 2
3 1 4
43
52
1
4
56 72 80 88
32 13 4 1
5 10 14 11
7 5 2 0
66 71
23 17
7 9
4 3
95 77 52 41 33
2 17 37 30 22
3 5 7 19 28
0 1 5 10 17
88 74 65 53 39
7 18 24 29 30
4 6 7 12 22
1 2 4 6 9
46 80 90 72
43 9 1 11
5 9 9 12
6 2 0 5
a
Service 1 includes employers of large firms and employees with a service relationship, like professional, higher technical, administrative and managerial, with a higher grade. b Service 2 only includes employees with a service relationship, like Service 1, but of a lower grade.
and accounts for slightly more than 3%. If we look at the probability of going to every type of performance, consumers of the omnivorous class have the highest chance of going to the high culture performances, except for light opera. Nevertheless, when compared with consumers of the popular class, omnivores also have the highest probability, by far, of going to popular performing arts. 4.1.2. Social class indicators and other cultural categories associated to lifestyles Here, we present the clusters’ profile for the levels of each active concomitant variable once the simultaneous
1451
model of classification and identification is estimated as is usual (Magidson and Vermunt, 2001). This procedure is equivalent to estimating the row profiles in a correspondence analysis when there is no restriction in the estimated latent class cluster model. For easier reading of Table 2 for each class of consumers, we have underlined probabilities greater than their mean profile (cluster size). The social class indicators—SES and level of education and income—are clear in the construction of a social hierarchy. All indicators of the lower social class are associated with the group of sporadic consumers: the lowest SES and the lowest levels of income and education as well. The indicators of middle class are associated with the popular class of consumers, especially nonmanual workers and a relatively minor proportion of employees from the middle and high managerial hierarchy (SES Services 1 and 2), with a college or university degree and good levels of income. The snob class of consumers is slightly higher on the social ladder, upper–middle class, with members from the highest managerial hierarchies and with university degrees and high levels of income. Finally, omnivorous consumers are associated with the highest social rank: They come from the highest managerial hierarchy, educational level, and income in a greater proportion than snobs do.
Table 3 General linear model with repeated measures with within-consumer performing arts fixed effects Effect
Boundary-effacement effect (within subjects and across activities)a Omnivore effect (between-consumer patterns of consumption)b Distinction effect (performing arts by patterns of consumption)c
Significance tests
Effect size
df
F
P level
g2
8, 94424
1598.49
b.0001
0.119
3, 11803
4324.54
b.0001
0.524
24, 94424
1201.10
b.0001
0.234
a The test reported here assumes bsphericityQ; that is, that the error covariance matrix of the orthonormalized transformed dependent variables is proportional to an identity matrix. The data did not pass Mauchly’s test of sphericity (v 2=29013.81,df=35, Pb.0001). However, the corrected Greenhouse-Geisser (GG) and Huynh-Feldt (HF) tests produce results similar to those shown in the main body of the table; that is, F GG=1598.49, df GG=5.39, 63631.31, P GGb.0001; F HF=1598,49, df HF=5.39, 63679.96, P HFb.0001. b The four multivariate tests, Pillai’s trace, Wilk’s lambda, Hotelling’s trace, and Roy’s largest root, produced the same statistics: F=1117,1, df=8, 11796, P=b.0001; g 2=0.431. c The corrected test again produced results similar to those shown in the table, namely, F GG=19425.86, df GG=16.17, 63631.31, P GGb.0001; F HF=19440.71, df HF=16.19, 63679.96, P HFb.0001.
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
1452
As far as the other cultural categories are concerned, sporadic consumers are the oldest and include slightly more males than females, typically with a marital status of widowed and divorced. The class of popular consumers consists of mainly single individuals who are younger, with more male than female consumers. The cluster of snob consumers, on the other hand, is made up of consumers over 35 years old, especially between 45 and 54 years old, and generally more females. In terms of marital status, the class is composed of just as many married as widowed persons, with a greater proportion of divorced individuals. Lastly, the class of omnivorous consumers is not as young as the popular consumers are, but are younger than the snobs are, and consequently composed of more singles and divorced persons. As far as gender is concerned, there are more males than females, although the difference is small.
Therefore, there is weak evidence for the boundaryeffacement effect.
4.2. Step 2: Evaluation of distinction, boundary-effacement, and omnivore effects
4.2.2. Omnivore effect When working with the omnivore effect, once we average the consumers’ rate of consumption for all types of performing arts, we should find differences among the types of consumers. This is what the between-subjects factor tries to capture. The omnivore effect is meaningful and the strongest of all three effects (g 2=0.524). Certainly, there is a strong tendency for upscale consumers to show higher consumption rates across all types of performing arts (see Table 4). The omnivorous pattern of consumption has the highest combined rate of consumption, and the combined rate of consumption has also the highest difference measured in standard errors and highest g 2 value. Consequently, we may conclude that our research does provide strong evidence that some people like everything.
4.2.1. Boundary-effacement effect In Table 3, we present the results of testing the hypothesis in a condensed format (SPSS Program, v11, was used to estimate the model). We see that the boundary-effacement effect is meaningful (there are differences in participation rates among the performing arts), although not strong (g 2=0.119). A detailed look indicates that the overall rate of the participation column in Table 4 shows that the highest participation rate corresponds to pop music concerts and theater, two popular performing arts events (in fact, empirically, theatrical performances cannot be classified as either lowbrow or highbrow product, but as between them), followed by classical music, folk events, flamenco shows, jazz, ballet, light opera, and opera. Hence, there are a few lowbrow performing arts categories in which the rate of consumption is consistently greater than the participation rate of some highbrow performing arts for all types of consumers, but not all of them, because classical music is the third highest category in terms of consumption.
4.2.3. Distinction effect The distinction effect works as an interaction among the performing arts factor (the boundary-effacement effect) and the patterns of consumption (the omnivorous effect) on the level of consumption. As far as the performing arts are concerned, it seems to be moderately strong (g 2=0.234) but statistically meaningful. Table 4, where performing arts types are ordered according to the magnitude of the difference in participation rate between sporadic and omnivorous class of consumers (measured in standard errors), shows why the distinction effect is modest. The differences grow as one progresses from the categories of light opera to flamenco, folk, opera, pop music, classical music, ballet, theater, and jazz. Although upscale social classes consume highbrow performing arts events more often than do low class consumers, differences are greater in some popular performing arts, like jazz and theater, and the between-group variability explained (g 2) is greater in the case of attendance to pop music concerts. There is a clear order in the
Table 4 Univariate tests for effects of the class split on combined activity level and on participation rates Category
Combined Light opera Flamenco Folk Opera Pop music Classical music Ballet Theater Jazz
Overall participation rates and by consumer class, along with standard errors (S.E.)
Effect size 2
Overall (S.E.)
Sporadic (S.E.)
Popular (S.E.)
Snob (S.E.)
Omnivorous (S.E.)
g
0.19 0.06 0.12 0.13 0.05 0.53 0.20 0.06 0.51 0.08
0.05 0.01 0.06 0.05 0.00 0.02 0.03 0.00 0.25 0.00
0.43 0.02 0.23 0.20 0.01 2.49 0.10 0.03 0.63 0.18
0.63 0.57 0.14 0.36 0.46 0.03 1.65 0.36 2.01 0.16
1.38 0.40 0.86 1.15 0.54 2.81 1.75 0.87 2.65 1.36
0.52 0.14 0.04 0.08 0.15 0.50 0.30 0.18 0.25 0.15
(0.004) (0.004) (0.006) (0.007) (0.003) (0.013) (0.008) (0.004) (0.011) (0.006)
(0.001) (0.001) (0.05) (0.006) (.001) (0.001) (0.003) (0.001) (0.008) (0.001)
(0.009) (0.005) (0.021) (0.019) (0.002) (0.047) (0.013) (0.006) (0.027) (0.019)
(0.018) (0.042) (0.026) (0.045) (0.037) (0.004) (0.073) (0.034) (0.063) (0.030)
(0.041) (0.056) (0.087) (0.102) (0.060) (0.141) (0.117) (0.072) (0.098) (0.115)
Significance
Differences in S.E.
F(3,11803)
P level
368.89 97.5 114.29 157.14 180 214.62 215 217.5 218.18 226.67
4324.54 631.50 178.96 324.90 711.31 3999.33 1690.93 833.45 1307.13 688.93
b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
attendance rate among the four classes of consumers in the overall rate of performing arts consumption (see the combined means) that breaks down in some highbrow and popular performing arts categories. Only snob consumers differentiate themselves by just consuming highbrow performing arts; nevertheless, the upscale social class (omnivores) consumes highbrow performing arts not only at higher rates than snobs do but also attend popular performing arts at a higher rate than popular consumers do.
1453
ordered, that play three different games, but with different intensity: the game of obscuring social class differences through homogenization of consumption, the game of showing off social standing by omnivorous consumption, and the game of building up boundaries between social classes through consumption that symbolizes status.
6. Conclusion and implications 6.1. Theoretical Implications
5. Discussion We have found evidence that consumers buy patterns of consumption products that mold a specific lifestyle. That is, we have found four patterns of consumption or classes of consumers: sporadic, popular, snob, and omnivorous consumers. These patterns of consumption have also been found in different symbolic spaces (van Rees et al., 1999; Lo´pez and Garcı´a, 2002), differing only in their proportions. All studies found an occasional pattern of consumption that accounted for most of the population, the sporadic class of consumers, another segment with a popular taste (between 8% and 20%), and the snob and omnivorous classes of consumers, which had a varying size depending on the symbolic space researched (snobs are a smaller segment than omnivorous in the study of van Rees et al., 1999, of Dutch reading habits). The four patterns of consumption are hierarchically associated to the social class indicators in our findings and in those reported by van Rees et al. (1999) and Lo´pez and Garcı´a (2002). Nevertheless, the association with the other cultural indicators varies, depending on the symbolic space researched. Not only in this research but also in Lo´pez and Garcı´a (2002), omnivores are usually younger than 35, whereas in van Rees et al. (1999), they are older. Nevertheless, in every case, they are mainly men. Hence, we can conclude that although the research as a whole lends support to the omnivorous consumption hypothesis about higher social classes showing off their status through an omnivorous lifestyle (DiMaggio, 1987), the distinction and boundary-effacement effects are also at work. Even so, we do not know if differences found in other research are caused by differences in the symbolic space analyzed or in the social space. When comparing our findings with the ones found by Holbrook et al. (2002), we see that the boundary effacement is not as strong, but the omnivore effect is also the strongest, and the distinction effect is almost of the same size. Taking into account that the research of Holbrook et al. (2002) and ours differ in the procedure used to find social classes’ patterns of consumption, it is curious that the results are convergent. In the symbolic space of performing arts consumption in Spain, there seem to be four types of consumers, hierarchically
In the preceding discussion, we have shown that the four patterns of consumption differ in size according to the symbolic space researched and the country. We can infer that a better understanding of the omnivore phenomenon and its association with the highest social status requires the use of two research strategies. First, an analysis of the consumption of a given symbolic space in different countries. Second, a study of the consumption of the full symbolic spaces for a given country. Additionally, to understand the social development of the omnivore phenomenon (i.e., how omnivore consumption has replaced univore-snob consumption as a marker of the highest social standing), we must research the same symbolic space and the same country over time. 6.2. Implications for marketing strategies The varied symbolic meanings of Spaniards’ performing arts consumption will help marketers to design a proper marketing strategy. On the one hand, the class of consumers labeled univore-snob tries to set themselves apart from the middle and lower social classes (the distinction effect we found). This desire to differentiate themselves means that every initiative targeting this group for the purpose of promoting the popular performing arts will be prone to failure, as reported by Gainer (1995). Fortunately, the other three classes of consumers will respond positively to certain marketing strategies. 6.3. Nonlinear pricing strategy: pure highbrow performing arts subscriptions As we have already seen, omnivore and snob consumers belong to the highest social classes and have the highest rate of attendance to highbrow performances (see Table 4). Therefore, we can expect that they will not only have the highest rate of consuming highbrow performances but also pay the highest reservation price for every additional performance. Moreover, as they are from the highest social classes, these consumers will have the right characteristics for buying a subscription series. Subscriptions allow us to segment the market according to the rate of consumption and marginal value of attending additional performances.
1454
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455
6.4. Nonlinear pricing strategy: pure popular performing arts subscriptions Popular and omnivore consumers are the targeted group of this strategy: omnivore consumers because they mark their social standing by consuming everything (the omnivorous effect), including popular performances, at rates even higher than popular consumers, and the latter because they are more interested in popular performances (the boundary-effacement effect) due to their age. Consequently, these two groups of consumers will have the highest rate of consumption and reservation price for every additional performance (see Table 4) and will create a stable demand and a target group for subscriptions. Because of their age, however, popular consumers have a more spontaneous behavior than omnivores do, suggesting that a minisubscription series should also be offered. 6.5. Price bundling strategy: mixed subscriptions Given the symbolic properties of snobs, these consumers will not respond positively to an offer of popular performances. However, because of their age and social class, popular consumers will probably be relatively unaware of highbrow performances (although not opposed to them) and, therefore, will pay comparatively low reservation prices. The strategy of mixed subscriptions consists in bundling a reduced number of highbrow performances with a set of popular performances (here, we are fostering the omnivorous effect). The key factor to success is to form a mixed subscription with a price lower than that of popular consumers’ reservation price, which is, nevertheless, not attractive to omnivore consumers. 6.6. Single ticket strategy: massive marketing Finally, although sporadic consumers have a lower rate of attendance, they are important to mass performing arts consumption because they account for up to 70% of potential goers. They are especially important for marketers of light opera, theatre, folk, and other popular performances (the boundary-effacement effect). Sporadic consumers have a low rate of consumption, although their reservation price for these few tickets is higher than the ticket’s mean price in a subscription.
Acknowledgement The authors express their appreciation to two anonymous reviewers, the editors of this special issue, Rube´n Gutie´rrez del Castillo (SGAE’s Research Department and Assistance) and the attendees to the 30th International Research Seminar in Marketing (La Londe les Maures, France, June 11–13, 2003) for their kind and helpful
comments on a previous version. This research was funded by Grant Number BEC2003-04462 of the Spanish Ministry of Science and Technology, Technical Department of the Humanities and Social Sciences.
References Adorno Theodor W. The culture industry: selected essays on mass culture. London (UK)7 Routledge; 1991. Boter Jaap, Wedel Michael. Segmentation of hedonic consumption: an application of latent class analysis to consumer transaction databases. J Mark Focus Manage 1999;3:295 – 311. Bourdieu Pierre. What makes a social class? On the theoretical and practical existence of groups. Berkeley J Sociol 1987;23:1 – 17. Bourdieu Pierre. La distincio´n: criterio y bases sociales del gusto. Madrid (Spain)7 Taurus; 1998. Original work published 1979. Clark Terry Nichols, Lipset Seymour Martin. Are social classes dying? Int Sociol 1991;6(4):397 – 410. Curtis William W. Social class or income? J Mark 1972;36(1):67 – 70. Dawson Scott, Stern Bruce, Gillpatrick Tom. An empirical update extension of patronage behaviors across the social class hierarchy. In: Marvin Goldberg, Gerald Gorn, Richard Pollay, editors. Advances in consumer research, vol. 17. Provo (UT)7 Association for Consumer Research; 1990. p. 833 – 8. DiMaggio Paul. Classification in art. Am Sociol Rev 1987 August;52: 440 – 55. DiMaggio Paul, Useem Michael. Social class and arts consumption: the origins and consequences of class differences in exposure to the arts in America. Theory Soc 1978;5:141 – 61. Evans Geoffrey. Testing the validity of the Goldthorpe class schema. Eur Sociol Rev 1992;8(3):211 – 32. Featherstone Mike. Cultural production, consumption, and the development of the cultural sphere. In: Mqnch Richard, Smelser Neil J, editors. Theory of culture. Berkeley (CA)7 University of California Press; 1992. p. 265 – 89. Firat A. Fuat, Venkatesh Alladi. Liberatory Postmodernism and the reenchantment of consumption. J Consum Res 1995 December;22:239 – 67. Gainer Brenda. Ritual and relationships: interpersonal influences on shared consumption. J Bus Res 1995;32(3):253 – 63. Gartman David. Culture as class symbolization or mass reification? A critique of Bourdieu’s distinction. Am J Sociol 1991;97(2):421 – 47. Holbrook Morris B. Nostalgia and consumption preferences: some emerging patterns of consumer tastes. J Consum Res 1993 September;20:245 – 56. Holbrook Morris B, Weiss Michael J, Habick John. Disentangling effacement, omnivore, and distinction effects on the consumption. Mark Lett 2002;13(4):345 – 3357. Katz-Gerro Tally, Shavit Yossi. The stratification of leisure and taste: classes and lifestyles in Israel. Eur Sociol Rev 1998; 14(4):369 – 86. Lazarsfeld PF, Henry NW. Latent structure analysis. Boston (MA)7 Houghton Mifflin; 1968. Lo´pez Jordi, Garcı´a Ercilia. Omnivores show up again: the segmentation of cultural consumers in the Spanish social space. Eur Sociol Rev 2002 September 3;18:353 – 68. Magidson Jay, Vermunt Jeroen K. Latent class factor and cluster models, biplots, and related graphical displays. In: Sober Michael, Becker Mark, editors. Sociological methodology, vol. 31. Boston (MA)7 Blackwell Publishers; 2001. p. 223 – 64. Martineau Pierre. Social classes and spending behavior. J Mark 1958 October;23:121 – 30. McCracken Grant. Culture and consumption: a theoretical account of the structure and movement of the cultural meaning of consumer goods. J Consum Res 1986 June;13:71 – 84.
J. Lo´pez Sintas, E. Garcı´a A´lvarez / Journal of Business Research 58 (2005) 1446–1455 McCutcheon Allan L. Latent class analysis. Sage university papers series on QASS, vol. 07-64. Thousand Oaks (CA)7 Sage; 1987. Morin Edgar. La industria cultural. In: Morin Edgar, Adorno Theodore W., editors. La industria cultural. Buenos Aires7 Argentina Editorial Galerna; 1967. p. 27 – 67. O’Shaughnessy John. Why people buy. New York7 Oxford University Press; 1987. Peterson Richard A. Understanding audience segmentation: from elite and mass to omnivore and univore. Poetics 1992;21:243 – 58. Peterson Richard A., Kern Roger M. Changing highbrow taste: from snob to omnivore. Am Sociol Rev 1996 (October);61:900 – 7. Roock Dennis W. Brands, consumers, symbols, and research, Sidney J Levy on marketing. Thousand Oaks (CA)7 Sage; 1999. Rutherford Andrew. Introducing Anova and Ancova—a GLM approach. Thousand Oaks (CA)7 Sage; 2001.
1455
Schaninger Charles M. Social class versus income revisited: an empirical investigation. J Mark Res 1981 May;18:192 – 208. SGAE Charles M. Informe SGAE sobre ha´bitos de consumo cultural. Madrid7 Fundacio´n Autor; 2000. Sivadas Eugene, Mathew George, Curry David J. A preliminary examination of the continuing significance of social class to marketing: a geodemographic replication. J Consum Mark 1997;14(6):463 – 79. Wasson Chester R.. Is it time to quit thinking of income classes? J Mark 1969 April 2;33:54 – 69. van Rees Kees, Vermunt Jeroen, Verboord Marc. Cultural classification under discussion—latent class analysis of highbrow and lowbrow reading. Poetics 1999;26:349 – 65. Vermunt Jeroen, Magidson Jay. LatentGold: user’s guide. Belmunt (MA)7 Statistical Innovations; 2000.
Journal of Business Research 58 (2005) 1456 – 1463
Researching cultural metaphors in action: metaphors of computing technology in contemporary U.S. life Rita M. Denny*, Patricia L. Sunderland Practica Group, LLC, 207 E. Ohio, #370, Chicago, IL 60611, USA Received August 2003; received in revised form October 2003; accepted February 2004
Abstract This article presents the utility of cultural metaphor as an analytic tool and theoretical construct in consumer research. Specifically, we extend current applications within consumer research by situating metaphor as a linguistic and cultural practice in consumers’ socially embedded talk and actions (vs. text and self-report) and by foregrounding an anthropological, cultural analytic framework. Based on ethnographic research, we take as our prime example the ways that computer and internet metaphors have had an impact on U.S. consumer meanings and actions in offices and other realms of daily life. We suggest that researching cultural metaphors in situ, in the everyday, naturally occurring context of consumer lives, has the potential to demonstrate highly relevant categories of meaning among consumers and thus represents an important strategy for market research. D 2004 Elsevier Inc. All rights reserved. Keywords: Metaphor; Culture; Ethnography; Computing technology; Computers
1. What is an office? In late 1999, we were asked to speak to a Fortune 100 company about the impact of technology on the ways people organized information in offices, at work or home. As the anthropological voice among the handful of assembled business experts, our assignment was to address bthe human need to organize.Q This company, in the business of paper, file folders (bdividersQ) and labels, was seeing the effects of electronic communication and storage. If the office becomes paperless, then what? Drawing on our experiences as ethnographic researchers in consumer environments, our talk covered the multidimensional aspects of organization—the aesthetic, the functional and the symbolic. We emphasized the fact that organization occurs in a living space—whether office or home. As a living space, an office would be varyingly * Corresponding author. 5538 S. Kenwood, Chicago, IL 60637, USA. Tel.: +1 773 752 2665. E-mail addresses: [email protected] (R.M. Denny)8 [email protected] (P.L. Sunderland). 0148-2963/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2003.10.015
landscaped by symbols of the self (Belk and Watson, 1998) (Figs. 1 and 2). The underlying question the talk raised was dWhat constitutes an office?T. Across much of the last century, the answer would have foregrounded a space that included a desk with drawers, a chair, paper, pens, maybe pencils, a phone, file cabinets or other form of paper storage, a wastebasket; in the last few decades, with the addition of the desktop computer. But today, we argued, the office is the computer—wherever it may sit (on the kitchen counter, park bench, in a backpack, hotel room, airplane, Starbuck’s counter—wherever). We suggested to the company that it was no longer in the bdividerQ business, rather it was in the bcomputerQ business because people looked to the computer not only for functionality but as a model for organization. In short, the computer had become the generative or source metaphor, and in order to develop organizer ideas that would resonate with consumers, one would need to start with the computer as the metaphoric model. It was not clear that the Fortune 100 company found our suggestion to mine the implications of the computing
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
1457
Fig. 1. A desktop computer in a home office, adorned with Post-its (in some ways akin to refrigerator art) on a desk. The office as living space and life space; surrounding photos speak of family.
metaphor as promising as we did. But after that time, we observed the generative power of computer and internet metaphors in other quarters. BellSouth, for example, recast their Yellow Pages as a virtual store (Linnett, 2000). In the words of the chief creative officer at WestWayne, the agency responsible for the new Yellow Pages advertising campaign, bWe are calling it dthe big yellow databaseTQ and bThe final tagline that we came up with is dThe original search engine.T It’s a neat way to update the book with an deT word.Q Similarly, a Toronto TV station adopted a Web-like format in its news broadcasting (McClain, 1999). Today, virtually all U.S. news programs have an dinternetT look. Jaron Lanier, a chief architect of internet Two, noted that reality TV in fact is more like a Web page than a television show (Schrage, 2000). This recasting of traditional book or broadcast media in computer terms is just the sort of
Fig. 2. Laptop computer, opened windows on the screen, surrounded by dwindowsT of family, friends, personal life.
Fig. 3. Computer as metaphor and model: bThe original search engineQ.
metaphor-mining we had suggested that our client should be thinking about for the paper-and-desk office (Figs. 3 and 4).
2. Metaphoric constructions and consumer behavior In recent years, the metaphor concept has gained attention in academic as well as applied marketing and consumer research circles. A prime catalyst for this attention can be located in the work of Zaltman and Coulter (1995), Zaltman (1997) and Coulter and Zaltman (2000). Their ZMET technique, a multi-step analytic procedure which uncovers consumers’ metaphors through guided conversation, storytelling, collage building, explorations of visual and other sensory images, leads to the extraction of consumers’ conscious and unconscious (bhidden or deepQ) mental models and reasoning processes (see Zaltman and Coulter, 1995 for a more detailed description of the technique). The concept of metaphor has been utilized by Hirschman (2002) who, in a cultural reading of dog breeds, illustrated forms of metaphoric transference and by Belk (1996), who has analyzed metaphorical tropes for pets as well as metaphors of the body with implications for organ donation practices (Belk, 1990). McQuarrie and Mick (1996) consider metaphor in an analysis of literary tropes in advertising, Hanby (1999) considers the metaphoric constructions of brands, and Dodd (2002) uses metaphor as a tool for constructing a cultural model of entrepreneur-
1458
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
Fig. 4. Toronto station news format mimicking a Web page.
ship in the United States. Holt (1995) has also used the notion of metaphor in the development of a typology of consumption practices. (See also Piller, 1999 who analyses metaphoric meanings of U.S. car brands.) In our own ethnographic consumer research practice, we have frequently drawn on the concept of metaphor as one of our analytic and strategic tools. For instance, in 1999, we conducted ethnographic research on the symbolic meanings of drugs and drug practices among American Tweens, ages 10 to 13 (Sunderland and Denny, 2003). Implicit in the Tweens’ talk were five distinct categories of drugs: scary (including LSD, Special K, Ectasy), death (heroin), addiction (cocaine), stupidity (inhalants) and the normative, unmarked category of drugs, which referenced alcohol, cigarettes and marijuana. While marijuana was articulated in terms of plant and organic substance metaphors, other categories were surrounded and permeated with chemical metaphors, framed as much more direct and harmful to bodily function. Inhalants, for example, as prototypical chemicals, bcut off oxygen to the brain.Q But marijuana, as a plant, became dangerous only when blacedQ or combined with a drug from another category. In the unmarked drug category, the effects of pollution on physical appearance, not chemical damage per se, was the worry (e.g., yellow teeth and bad breath from cigarette smoke). Among our recommendations to the client, responsible for creating anti-drug use advertising, we pointed to the need to play on the pollution and chemical metaphors that resonate so powerfully for Tweens. We also pointed to ways the natural (safe, organic, plant) metaphors associated with marijuana posed a challenge in dissuading Tweens of its use. The use of metaphor as a theoretical construct in consumer research (Dodd, 2002; Coulter and Zaltman, 2000; Hanby, 1999; McQuarrie and Mick, 1996; Piller, 1999; Stern, 1989; Zaltman and Coulter, 1995) has relied primarily on cognitive linguistic, psycholinguistic and literary views (Cameron and Low, 1999; Gibbs, 1994;
Ko¨vecses, 2002; Lakoff, 1987; Lakoff and Johnson, 1980; Ortony, 1993), which position metaphor as an organizing principle used by individuals to construct conceptual understandings of the world. Our approach diverges from most others in the consumer research arena by our interest and focus on the cultural (or dsupraindividualT) realm over that of the individual and individual thought processes (but see Hirschman, 2002; Belk, 1990, 1996). Anthropologically trained, we do not stress the understanding of individual thought processes—or the mechanisms of the mind, conscious or unconscious— but rather the cultural level meanings and metaphors that animate consumer lives (Fernandez, 1991; Martin, 1987, 1994). Our work also extends the work of others in that we conduct the research ethnographically, stressing the importance of socially embedded and contextualized consumer action. Other consumer research on metaphor has largely concentrated on consumers’ self-reports, representations and/or texts of various kinds (written, images, advertisements, etc.). We research consumer behavior in situ, and analyze metaphors as they are instantiated in social action through behavior, speech, organization, artifacts, and thoughts. Our research often includes texts (especially photos, images, and video), as well as interview, introspective, and self-report data, but we consistently relate this to real action in real social worlds. Thus, we are primarily interested in understanding the way consumers live out culturally relevant metaphors in their everyday lives, believing that this arena has the potential to demonstrate highly salient meanings for consumers as well as realms of opportunity for marketers. When we search to understand the metaphors that organize behavior (or vice versa), we search for the cultural level meanings—the metaphoric meanings that resonate for consumers and marketers—because these meanings are important threads and patterns in the cultural fabric that is the context for product and brand development and consumption.
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
During the last decade, there has been much debate on the privileged position of metaphor in the construction of thought, in which the positions of cognitive linguists and more culturally grounded theorists were in opposition (see Fernandez, 1991; Quinn, 1991; Alverson, 1991). More recently, the role of culture in mediating metaphoric constructions has been increasingly recognized by cognitive psychology (see Gibbs and Steen, 1999). And though the debate continues as to whether metaphors constitute or reflect cultural models (Ko¨vecses, 1999), it seems clear that uses of metaphor in everyday talk are contextualized by cultural realities, whether by discursive conventions and ideology (Eubanks, 1999), historical time (Ohnuki-Tierney, 1991) or socio-cultural experiences (Gibbs, 1999; Emanatian, 1999). Our emphasis on metaphors as they are articulated in conversation and social action assumes language is a social resource and is itself a culturally defined and specified practice (Schieffelin, 1990; Silverstein, 1976). In the end, we take as a premise that language not only reflects cognitive processes but, in its use, is a creative practice of culture, in which artifacts and environment are catalysts for new expression and cultural meanings. Metaphor as a trope of everyday language thus becomes a prism through which to observe and refract consumer behavior. To further illustrate, we return to the example of computing technology.
3. Computing as a cultural metaphor Early design of desktop computers pointed to the potency of paper, filing cabinet, and desk metaphors. Desks and what surrounded them were the source metaphor for computers. ! Starting with Xerox PARC and mainstreamed by Apple and Microsoft, was the bdesktop,Q a two-dimensional space on which file folders (bdividersQ) neatly sat. We could name them and put them wherever we liked. There were even wastebaskets (now brecycle binsQ) and Mac trash included visual and sound effects.
Fig. 5. Mimicking the computer’s desktop: Folders on the floor, opened and closed throughout the day, as needed, then put into the backpack.
1459
Fig. 6. Actual desktop: information less immediately needed. It might not be touched day-to-day.
! Monitors were designed on a paper icon so that when we opened a word processing application it was as though we were writing a letter. ! Keyboards were like typewriters. But with computers came other organizational opportunities. For instance: ! Endless embedding of file folders within file folders. ! Embedding could be done with impunity because we had search functions for finding things really quickly. ! Visual icons for file types; sorting of bfoldersQ by name listing, icon, or date. ! We could use multiple sorting possibilities. We got instant sorting capability that mixed and matched in any number of ways so that eventually we could spin incoming e-mail by date, author, recipient, etc. With the proliferation of personal computers on desks by the 1990s, the source and target of the desk-computer metaphor could become reversed: it became possible to mimic our computers mimicking our desks. In this scenario, real desktops could be piled high with folders because of an increasing dependence on a visual cue for what is there (as our computer desktops gave us). In 2001, we observed a tech-savvy individual doing just this. On coming into the office he set up a model of his computer ddesktopT on the floor by his desk: File folders were laid out from his backpack, opened up when in use, then dclosedT when tasks were completed or the day was done. His actual desktop was piled with folders of less temporal immediacy (Figs. 5–7). The way of doing things in non-computer realms had begun to metaphorically mimic what we did with computers. Hence, retrieval systems in the non-computer world could suffer in comparison to what had been learned was possible via the computer’s desktop. As one of our research respondents, Travis, commented, bTechnology makes information management easier. You can save virtually forever. You have multiple search tools and search through the hard drive real quick, unlike file cabinets where everything is just kind of stacked there.Q For a client in the filing and bdividerQ
1460
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
Fig. 7. Desktop computer screen: Visually parsed with file folders and pictures of son, which once might have been framed and displayed on the desk.
business, it was time to consider these other, metaphorically inspired ways of doing things. What could be invented for floor-based filing systems that mimicked the computer desktop filing systems? What could be invented as computer-inspired ways of embedding, embedding, and embedding within folders? What could be communicated about the specific retrieval properties of paper products, akin to what one does with computers? These are the kinds of questions that we believed the company’s strategists needed to consider as they re-thought their business in 1999.
4. Computing technology: from humanly negative to personally positive In the United States, a metaphoric migration occurred with the proliferation of personal computing technology— from conceptualizing computing technology as humanly destructive to seeing it as personally empowering. Traditionally in the U.S., there reigned an ambivalence vis-a`-vis technology. Technology was seen as a force that drove society forward but, contained within this power to alter the nature of social life, was an equally powerful anti-human force—the one that made technology dcoldT—something that deprived the world of humanity, emotions and feeling. Earlier on in the advent of personal computers and particularly the internet in the U.S., research respondents would recite urban legends (as did the mass media) of people who would stay at their computers and online, no longer caring about their family or friends—for instance, mothers who spent so much time online that the children were neglected and went unfed. But, by the beginning of 2000, when we interviewed consumers we heard about computers and technology as bgood,Q bfriends,Q bhelpers.Q Metaphors of friendship and assistance had become the order of the day. At the base of the positive embrace of computing technology lay a change in the conceptualization of its power. People moved from understanding the technology of computers as a social force (outside of an individual’s power) to understanding it, and
experiencing it, primarily in personal (part of a person’s power) terms. On the cover of a 1972 book, The Computer: How it’s Changing our Lives, we find a man alone in a room with his hands reaching toward the controls of one big (the size of a room) computer (Snyder, 1972). The image, and its attendant associations, is reminiscent of HAL of 2001, A Space Odyssey, produced in about the same cultural time (1968). The (mainframe) computer of these times was crude and inscrutable, something that took over human life (Fig. 8). In the more recent conception, computing technology has moved from being conceived as a large social force outside of us to power on an individual scale, virtually inside of us. We now observe its articulation not as something that masters us, but rather as something that makes us masters— we help ourselves, and even help others, with our own personal computing technology. Thus, an additional array of elements has been foregrounded in talk about computers, as seen below: ! Customizing a homepage, showing specific stock quotes and news means, bI manage the flow of information. I get what I need.Q ! Having a hand-held computer can make me feel: bI’m supremely in control of my destiny.Q ! Using the searchable OED means, bI manipulate the information. I get what I want.Q ! bVia the internet I can find the information myself. I can communicate directly with people all over the world. I can find the medical information I need to assure I get— or stay—well. I can help others with their problems.Q This more celebratory sense of computers and the internet, a positive evaluation fostered by personal use of these personal machines is built on—and builds—the metaphor of computer-as-personal-power. Computer technology does help people in the world today. For the
Fig. 8. Early 1970s book cover image—mainframe computer of that time as inscrutable, and with the potential to take over human life.
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
everyday person, computing technology really can expand memory (e.g., a hand-held device takes bthe fear of forgetting out of lifeQ) and it can facilitate communication; with e-mail, respondents maintain that they have increased their contact with others across space and time. It would also seem that the embrace of the technology and its associated metaphors has upped performance antes. Expectations for products, our selves, our worlds have been increased. Gleick (1999) has argued that with computers speed has become the assumed standard for calibrating anything from intelligence to productivity. Faster is better. This standardization of speed paralleled the development of microprocessors—computers offered an ever finer ability to parse time. And so, as a society, we are now in the rather absurd position of finding relevance in chunks of time that are both imperceptible and measurable, e.g., the milliseconds by which races are won, or the seconds by which appliances save time. We would argue that it is not just the computer’s speed that has infiltrated notions of work, productivity and intelligence and that has put us in a perpetual state of immediacy. It is the surfing, the e-mailing, the embedding, the instant messaging and chatting, the finding, the inserting, the linking that give the cues for comprehending and interpreting what is around us, and for enacting and reenacting that reality. Our tropes for how we communicate with one another have also been influenced—we now have online and offline conversations. In our encounters with people, are we becoming more like our e-mail—brief, casual, punctuated and immediate? Are we surfing places much as we do websites? It is not just the idea of speed; it is the native model of what the computer does and the conventions derived from their use of them that makes a computer a potent metaphor and a native model for productivity. A tech-savvy respondent in his 20s tells us that he gets impatient and irritated at red lights not because the lights are slow, but because the lights are not bsmartQ— they should be able to dreadT traffic needs—and therefore be dgreenT when there is no traffic. While a bfastQ processing computer is a bgoodQ computer, a bfastQ computer is also one that allows us to carry out multiple processes, to simultaneously open up multiple windows, to do things like quadruple boot.
5. Conclusion: computing metaphors are reconfiguring the sense of self and society By forging their way into our notions of personal power, computing metaphors are affecting our cultural sense of what it means to be a person. These metaphors have also affected our sense of how our minds work. If people now say that something was not saved on their bhard drive,Q they might not literally mean in their computer, but rather in their own memories. One respondent told us that his wife has accused him of being a bbinary thinker,Q using a computer-
1461
inspired metaphor to express her exasperation over what, in other times, she might have described as his inability to think in shades of gray or his dblack and whiteT thinking. Since today the internet is increasingly synonymous with the computer, it has become an important source for contemporary mental processing metaphors. bBandwidthQ is now part of the figurative vocabulary, as in bdo you have the bandwidth for that?Q roughly glossed as scope and ability. And if once we bwrapped our heads around an idea,Q we now bdownloadQ and blinkQ to it, as we download and link to sites on the internet. The use of these tropes illustrate, again, the generative or source status of the computer metaphor—which now foregrounds organization, powers of retrieval and productivity of the mind versus more creative, figurative dimensions (even if in programming and artificial intelligence circles, articulations of computer processes have gone far beyond mere retrieval and calculation, see Turkle, 1995). Notably, computing metaphors have also entered professional models of how the brain and mind works (Gleitman et al., 1999; Turkle, 1997). In the well-known work of Turkle (1984, 1995), we also find the argument that people’s sense of self has been affected in light of computer familiarity and use. For instance, she maintains that people have a Windows-influenced sense of themselves as both multiply refracted and flexible (Turkle, 1995). We would add that the current portability of the computer and related technological devices must also be considered. Beyond the sense of self, the portability of contemporary technological devices has allowed us to reconfigure our sense of social space and place. Harvey (1990) has discussed the concept of btime–space compressionQ that has arisen in light of interconnected capital, transportation and communication technologies of the modern and postmodern worlds. As he observed, many have noted—with varying degrees of alarm—the seeming collapse of social space and time in light of these developments. We have observed that portable devices and wireless communications have given people the ability to collapse temporal and physical space between home and office, private and public, work and play. The kitchen counter, park bench, hotel room, airplane and the spot at Starbuck’s do become workspaces when one pulls out the laptop. Cell phones, call forwarding, internet access and dial-up networking allow for the dislocation of white-collar work from set work places. Play at work is also normative. Respondents take personal phone calls on their cell or separate lines, write personal e-mails or privately go to websites of their choice, buy and sell on favorite internet sites. Instant messaging with friends, watching sports events (streaming or with portable TVs) or listening to music with headphones—from radio over the computer or with portable CD players—is also common. Because the spatial divide can be collapsed, and one can work where once one only played or play where once one worked, it does sometimes seem that traditional boundaries between cultural domains such as work and home have
1462
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463
collapsed. Technology provides the potential for increasing the integration between these domains functionally (Venkatesh et al., 2001) and subsequently, conceptually (see Nippert-Eng, 1996). But, just as Harvey (1990, p. 232) noted that spatial barriers are only overcome through the production of other spaces (e.g., air travel’s reduction of spatial distance is achieved via airport dspacesT) and that even as spatial distance has become less relevant for capital, the search for the most advantageous geographic location remains highly relevant, we would suggest that the seeming elision of boundaries between cultural categories is structured. With Windows as the culturally articulated metaphor, the cultural categories of work and play that were once defined by particular senses of place are maintained in computer windows. Truly interesting are the ways in which the very same technologies that afford the space–time collapse between cultural spheres are simultaneously the means by which their functional and symbolic separation is assured. People’s actions maintain the separation of cultural categories. Cell phones are taken to work for making and receiving personal calls. People have separate e-mail accounts—one for personal, one for business. The very wired have several—the junk mail account, the friends and family account, the chat account, plus the business accounts. But even the not-so-wired with one account know which messages are personal and which are business. People play games, pull up stock quotes, peruse banking sites, download music and watch sports events during work hours, but these are in different windows—a click away from the windows of work. In essence, what we see people doing is not merging, but rather holding up the longstanding cultural boundaries between work and play, private and public, home and office, but doing so with the aid of their technological devices. Taking us full circle, the computer and related technological devices really have become our office dividers. Today, the well-established cultural categories are kaleidoscopic, finely interwoven and parsed in windows. The switch of attention between windows is mediated only by the perceived reaction time of our hand on the mouse, the speed of the computer’s processor, the time it takes to answer a ringing phone. Cultural categories and metaphors have historical roots. The physical separations necessitated by early industrial processes had a huge hand in creating the current symbolic divisions between public and private, work and play. Metaphors, as we have noted, are lenses which refract current cultural beliefs and values. They not only provide a prism through which to understand consumption behavior but, in their use/instantiations by individuals, are creative of ways of seeing. Metaphors change. For marketers, metaphorical representations must be fathomed in order to persuade, speak or resonate with target audiences (see Belk, 1990 in reference to organ donation). Culture and metaphors are lived or practiced, observable in the details of daily life—as new technologies are dconsumedT or inte-
grated into daily life, we can expect new refractions (see Venkatesh, 1998; Venkatesh et al., 2001). Today, Windows give us the current model for a kaleidoscopic lens, in which multi-tasking and a recalibration of linear time are currently framed. We might expect, tomorrow, that what it means to work or to play or what counts as private and public will fundamentally change as post-Windows technology and metaphors offer us a new framework for interpreting the world around us. We do not know what will happen when wireless internet screens are everywhere, everyday computers have the power to simultaneously process in multitudes of windows, broadband allows us to become a seamless part of the show or wireless networks allow us to interface seamlessly between phone, computer, PDAs (all technologies that exist today, but that are not yet fully integrated into everyday life). But we suggest that drawing out the cultural metaphors by turning an acute ear and sensitive eye to how consumers talk and act in everyday life will provide a significant frame for marketers’ attention be it theoretical or practical.
References Alverson H. Metaphor and experience: looking over the notion of image schema. In: Fernandez JW, editor. Beyond metaphor: the theory of tropes in anthropology. Stanford7 Stanford Univ. Press; 1991. p. 94 – 117. Belk R. Metaphoric relationships with pets. Soc Anim 1996;4(2):121 – 46. Belk R. Me and thee versus mine and thine: how perceptions of the body influence organ donation and transplantation. In: Shanteau J, Harris RJ, editors. Organ donation and transplantation. Washington (DC)7 American Psychological Association; 1990. p. 139 – 49. Belk R, Watson J. Material culture and the extended and unextended self in our university offices. Adv Consum Res 1998;25:305 – 10. Cameron L, Low G, editors. Researching and applying metaphor. New York7 Cambridge Univ. Press; 1999. Coulter R, Zaltman G. The power of metaphor. In: Ratneshwar S, Mick DJ, Huffman C, editors. The why of consumption: contemporary perspectives on consumer motives, goals, and desires. London7 Routledge; 2000. p. 259 – 81. Dodd SD. Metaphors and meaning—a grounded cultural model of US entrepreneurship. J Bus Venturing 2002;17:519 – 35. Emanatian M. Congruence by degree: on the relation between metaphor and cultural models. In: Gibbs RW, Steen G, editors. Metaphor in cognitive linguistics. Amsterdam7 John Benjamin; 1999. p. 205 – 18. Eubanks P. The story of conceptual metaphor: what motivates metaphoric mappings? Poetics Today 1999;20(3):419 – 42. Fernandez JW, editor. Beyond metaphor: the theory of tropes in anthropology. Stanford7 Stanford Univ. Press; 1991. Gibbs RW. Taking metaphor out of our heads and putting it into the cultural world. In: Gibbs RW, Steen G, editors. Metaphor in cognitive linguistics. Amsterdam7 John Benjamin; 1999. p. 145 – 66. Gibbs RW. The poetics of mind: figurative thought, language, and understanding. Cambridge7 Cambridge Univ. Press; 1994. Gibbs RW, Steen G, editors. Metaphor in cognitive linguistics. Amsterdam7 John Benjamin; 1999. Gleick J. Faster. New York7 Pantheon Press; 1999. Gleitman H, Fridlund A, Reisberg D. Psychology. 5th ed. New York7 W.W. Norton; 1999. Hanby T. Brands - Dead or alive? J Mark Res Soc 1999;41(1):7 – 18.
R.M. Denny, P.L. Sunderland / Journal of Business Research 58 (2005) 1456–1463 Harvey D. The condition of postmodernity. Cambridge (MA)7 Blackwell; 1990. Hirschman EC. Dogs as metaphors: meaning transfer in a complex product set. Semiotica 2002;139(1–4):125 – 59. Holt DB. How consumers consume: a typology of consumption practices. J Consum Res 1995;22:1 – 16. Kfvecses Z. Metaphor: does it constitute or reflect cultural models? In: Gibbs RW, Steen G, editors. Metaphor in cognitive linguistics. Amsterdam7 John Benjamin; 1999. p. 167 – 88. Kfvecses Z. Metaphor: a practical introduction. Oxford7 Oxford Univ. Press; 2002. Lakoff G. Women, fire, and dangerous things: what categories reveal about the mind. Chicago7 Chicago Univ. Press; 1987. Lakoff G, Johnson M. Metaphors we live by. Chicago7 University of Chicago Press; 1980. Linnett R. BellSouth Yellow Pages recast as virtual store. Advertising Age 2000;1: [June 5]. Martin E. Tracking immunity in American culture—from the days of polio to the age of AIDS. Boston7 Beacon Press; 1994. Martin E. The woman in the body: a cultural analysis of reproduction. Boston7 Beacon Press; 1987. McClain D. Toronto TV station adopts web-page format. N Y Times; 1999. C8 [December 27]. McQuarrie EF, Mick DG. Figures of rhetoric in advertising research. J Consum Res 1996;22:424 – 38. Nippert-Eng C. Home and work. Chicago7 University of Chicago Press; 1996. Ohnuki-Tierney E. Embedding and transforming polytrope: the monkey as self in Japanese culture. In: Fernandez JW, editor. Beyond metaphor: the theory of tropes in anthropology. Stanford7 Stanford Univ. Press; 1991. p. 159 – 89. Ortony A, editor. Metaphor and thought, 2nd ed. Cambridge7 Cambridge Univ. Press; 1993. Piller I. Extended metaphor in automobile fan discourse. Poetics Today 1999;20(3):483 – 98.
1463
Quinn N. The cultural basis of metaphor. In: Fernandez JW, editor. Beyond metaphor: the theory of tropes in anthropology. Stanford7 Stanford Univ. Press; 1991. p. 56 – 93. Schieffelin BB. The give and take of everyday life: language socialization among Kaluli children. New York7 Cambridge Univ. Press; 1990. Schrage M. Jaron Lanier (Interview). Adweek 2000; 41(40): IQ44, Eastern Edition. Silverstein M. Shifters, linguistic categories and cultural description. In: Basso K, Selby H, editors. Meaning in anthropology. Albuquerque7 Univ. of New Mexico Press; 1976. p. 11 – 55. Sunderland PL, Denny R. Psychology vs anthropology: where is culture in marketplace ethnography? In: Malefyt TD, Moeran B, editors. Advertising cultures. London7 Berg; 2003. p. 187 – 202. Snyder GS. The computer: how it’s changing our lives. Washington (DC)7 U.S. News & World Report; 1972. Stern B. Literary criticism and consumer research: overview and illustrative analysis. J Consum Res 1989;16:322 – 34. Turkle S. Computational technologies and images of the self. Soc Res 1997;64:1093 – 111. Turkle S. Life on the screen: identity in the age of the internet. New York7 Simon & Schuster; 1995. Turkle S. The second self: computers and the human spirit. New York7 Simon & Schuster; 1984. Venkatesh A, Stolzoff N, Shih E, Mazumdar S. The home of the future: an ethnographic study of new information technologies in the home. Adv Consum Res 2001;28:188 – 97. Venkatesh A. Cyberculture: consumers and cybermarketscapes. In: Sherry JF, editor. Servicescapes. Chicago7 NTC/Contemporary Publishing; 1998. p. 343 – 76. Zaltman G. Rethinking market research: putting people back in. J Mark Res 1997;34:424 – 37. Zaltman G, Coulter RH. Seeing the voice of the customer: metaphor-based advertising research. J Advert Res 1995;35(4):35 – 51.
Journal of Business Research 59 (2006) 726 – 727
Editorial
Marketing communications and consumer behavior: Introduction to the special issue from the 2005 La Londe seminar
This special issue of the Journal of Business Research includes a selection of papers presented at the 32nd International Research Seminar in Marketing organized by the Aix Graduate School of Management (I.A.E. Aix-en-Provence), a part of the University Paul Cézanne in Aix-Marseilles (France). This seminar, now better known as the “La Londe Seminar”, is devoted to Marketing Communications and Consumer Behavior on a biennial basis. The La Londe seminar encourages intense discussions of specialized topics in consumer behavior and communications among a relatively small group of interested scholars in a relaxed and informal atmosphere. In order to facilitate fruitful exchanges, there are only two tracks, and each presentation lasts 30 min followed by 15 min discussion. Those who have been to La Londe also know the exchange value of the coffee breaks on the terrace overlooking the Mediterranean Sea and the Porquerolle islands. The La Londe seminar is truly internationally oriented. It is always chaired by both a European chair and an American chair. Luk Warlop (Katolieke Universiteit Leuven) and Curtis P. Haugtvedt (Ohio State University) chaired the 2005 edition. Participants came from all over the world, with the five continents and 16 nationalities represented. A total of 70 manuscripts were submitted and double-blind reviewed by both members of the permanent scientific committee of the seminar and ad hoc reviewers carefully selected by the co-chairmen and the coordinator. Thirty papers were presented at the conference. In addition, the thought-provoking keynote address by Robert A. Peterson from The University of Texas at Austin stimulated reflections and discussions concerning the process of researching fruitfully in consumer behavior domains. Bob Peterson shared his rich and outstanding research experience in a talk entitled “Lessons Learned in Later Life”. The eight papers of the special issue follow the dual theme of consumer behavior and marketing communications, albeit with great diversity. The three first papers address advertising-related issues. Elisabeth Cowley (“Processing exaggerated advertising claims”) demonstrates that consumers' brand evaluations are inflated after exposure to exaggerated claims even if they recognize that the claims are not very credible. Exaggerated claims seem to be accepted before being discredited, thus affecting memory. Jae Min Jung and James J. Kellaris 0148-2963/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2006.01.010
(“Responsiveness to authority appeals among young French and American consumers”) examine the authority principle cross-culturally in an advertising context and study how different levels of authority influence consumer attitudes and purchase intentions. They also investigate moderating effects of national culture, source credibility, and individuals' orientations to power distance. Results show a reverse authority effect across two cultures and that both perceptions of spokesperson credibility and power distance positively influence attitudes and purchase intentions. In the third article, Ingrid Poncin, Rik Pieters and Michele Ambaye (“Cross-advertisement affectivity: the influence of similarity between commercials and processing modes of consumers on advertising processing”) run an experiment to examine the issue of cross-advertisement affectivity effects. They find substantial context effects of the ad sequence in the affective reactions elicited by commercials presented in the same pod. They also demonstrate the importance of the similarity between commercials in consumer judgments. The three following articles of this special issue deal with consumer purchasing behavior. Francine V. Garlin and Katherine Owen (“Setting the tone with tune: a meta-analytic review of the effects of background music in retail settings”) study more than 150 papers dealing with the effects on consumer behavior of background music in service settings. They demonstrate through a meta-analysis of research results that appropriate background music induces positive effects on affective, attitudinal, temporal and behavioral variables. In their contribution, Joann Peck and Terry L. Childers (“If I touch it I have to have it: individual and environmental influences on impulse purchasing”) examine individual differences in touch and how these affect impulsive-buying behavior. They designed a study to investigate the link between impulse purchasing and both an environmental encouragement to touch (a sign in a supermarket) and an individual preference for autotelic touch. They find that individuals higher in autotelic need for touch (NFT) purchase more impulsively than their lower autotelic NFT counterparts. In addition, the environmental salience of touch information increases impulse-purchasing behavior. Finally, Bill Jolley, Richard Mizerski and Doina Olaru (“The effects of habit and satisfaction on player retention for online
Editorial
gambling”) experimentally test the effects of past behavior (habit) against customer satisfaction on consumer retention. They developed an online casino and a very nice online gambling experiment to test the drivers of retention. They find that playing habits, not customer satisfaction, have a strong effect on a range of responses tied to retention and add a strong contribution to the literature proving the very limited impact of satisfaction on consumer retention. The two last papers of the special issue address central consumer behavior constructs with a more psychological orientation. Todd A. Mooradian and K. Scott Swan (“Personality and Culture: the case of national extraversion and word-ofmouth”) demonstrate the general value of the personality-andculture paradigm. They hypothesize that cultures characterized by higher levels of extraversion will rely on word-of-mouth for product information more than cultures with lower extraversion. They test this hypothesis using a very impressive database including 15,000 cases from subjects in twelve nations on five continents (for the WOM data) and major cross-cultural databases for the extraversion data and find empirical support for the hypothesis. The final paper is written by Frank R. Kardes, Steven S. Posavac, David Silvera, Maria L. Cronley, David M. Sanbonmatsu, Susan Schertzer, Felicia Miller, Paul M. Herr and Murali Chandrashekaran (“Debiasing Omission Neglect”). Omission neglect (insensitivity to missing information as when products are described by a small amount of information) occurs when consumers form inappropriately extreme evaluations on the basis of weak evidence. This study investigates, through two experiments, the effectiveness of new procedures for reducing omission neglect: asking consumers to consider their criteria for judgment before receiving product information (Experiment 1), and asking consumers to rate presented and missing product attributes before providing overall product evaluations (Experiment 2). The research shows that the proposed procedures are effective in debiasing consumers' tendency for omission neglect. As co-chairs and coordinator of this conference and as coeditors of this special issue, we greatly appreciate the contributions of the international scientific committee of the La Londe seminar which year after year guarantees the quality of the contributions. This permanent scientific committee is composed of the following very distinguished scholars: Gerald Albaum (University of New Mexico), Rajeev Batra (University of Michigan at Ann Arbor), Russell W. Belk (University of
727
Utah), Christian Derbaix (FUCAM, Mons), Yves Evrard (HEC, Paris), Wayne D. Hoyer (University of Texas at Austin), Alain Jolibert (University of Grenoble), Lynn R. Kahle (University of Oregon), Michel Laroche (Concordia University), Gilles Laurent (HEC, Paris), Siew Meng Leong (National University of Singapore), Sidney J. Levy (University of Arizona), Richard J. Lutz (University of Florida), Claude R. Martin (University of Michigan at Ann Arbor), Hans Mühlbacher (University of Innsbruck), Robert A. Peterson (University of Texas at Austin), Rik Pieters (Tilburg University), Christian Pinson (INSEAD), Bernard Pras (University of Paris–Dauphine and ESSEC), Don E. Schultz (Northwestern University), Jan-Benedict Steenkamp (Tilburg University), Alain Strazzieri (Aix-Marseille University), W. Fred van Raaij (Tilburg University), Arch G. Woodside (Boston College) and Judy Zaichkowsky (Simon Fraser University). We thank all the other members of the review panel which have done a great reviewing job. Finally, we express our gratefulness to Arch Woodside, editor in chief of the Journal of Business Research, for instigating and approving this special issue. We now look forward to the 2007 seminar that will be cochaired by Joseph Sirgy (Virginia Polytechnic Institute and State University) and Søren Askegaard (University of Southern Denmark in Odense). The 2007 La Londe seminar will take place June 5–8, 2007 in La Londe les Maures, this stimulating and beautiful place on the French Mediterranean coast. Curtis P. Haugtvedt Ohio State University, United States Dwight R. Merunka IAE Aix-en-Provence, Paul Cézanne University of Aix-Marseille, Clos Guiot, 13540 Puyricard, France E-mail address: [email protected] Corresponding author. Tel.: +33 442 280 808; fax: +33 442 280 800. Dwight R. Merunka Euromed Marseille, School of Management, France Luk Warlop Katolieke Universiteit Leuven, Belgium 1 January 2006
Journal of Business Research 59 (2006) 728 – 734
Processing exaggerated advertising claims Elizabeth Cowley ⁎ University of Sydney, Economics and Business Building H69, Sydney NSW 2006, Australia Received 1 April 2005; received in revised form 1 November 2005; accepted 1 December 2005
Abstract Government policymakers allow advertisers to use wildly exaggerated, fanciful or vague claims for a product or service because they believe that nobody could possibly treat the claims seriously or be misled by them. The results demonstrate that although consumers are able to identify exaggerated claims as less credible than factual claims, their brand evaluations are inflated after exposure to exaggerated claims. The explanation is that during the process of comprehension, claims are accepted before being discredited. The temporary acceptance of the claim affects memory, even after the claim is understood as an exaggeration. © 2006 Elsevier Inc. All rights reserved. Keywords: Advertising; Brand evaluation; False claims
1. Introduction Is belief independent of comprehension? Descartes asserted that comprehension precedes, and is independent of, belief. As humans come into contact with claims or opinions, they comprehend them automatically, and then decide whether to accept or reject the information (Gilbert, 1991; Gilbert et al., 1993). Spinoza, on the other hand, believed that comprehending and accepting were part of the same process. “According to Spinoza, the act of understanding is the act of believing. As such, people are incapable of withholding their acceptance of what they understand. They may indeed change their minds after accepting the assertions they comprehend, but they cannot stop their minds from being changed by the contact with those assertions” (Gilbert et al., 1993, p. 222). Why does the disagreement matter in marketing? Selfregulatory industry groups, such as the ASA in the United Kingdom and government regulatory bodies, such as the FTC in the United States and the ACCC in Australia, have created rules using Cartesian logic. Government policymakers allow advertisers to use puffery, defined as wildly exaggerated, fanciful or vague claims for a product or service, because they believe that
⁎ Tel.: +61 2 9351 6433; fax: + 61 2 9351 6732. E-mail address: [email protected]. 0148-2963/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2005.12.004
nobody could possibly treat puffery seriously or be misled by it. Two critical assumptions underly the policy. First, consumers can identify puffed claims as not credible. Second, consumers will not incorporate a puffed claim into their evaluations or beliefs because they understand that the puffery is a ‘wild’ exaggeration. Although Cartesian logic is used by policymakers, empirical evidence supports Spinoza's view. Gilbert et al. (1993) show that interrupting the processing of false claims results in participants believing the claims to be truer. The disturbing implication of the findings by Gilbert et al. (1993) is that every encounter with misinformation or an exaggerated claim can potentially affect future behavior, even if the consumer realizes that the claim is false. The research presented here demonstrates that although consumers are able to identify an exaggerated claim as less credible, exposure to the puffed claim still shifts the evaluation of the brand to be more positive. 2. Processing exaggerated claims 2.1. A Cartesian approach Government policymakers have taken the position that consumers recognize that puffery lacks credibility, even though very little empirical evidence has been presented to support this assumption. Policymakers assume that when consumers process exaggerated claims they understand the lack of credibility and
E. Cowley / Journal of Business Research 59 (2006) 728–734
ignore the claims. Cartesian logic is implicit in this assumption, as there is no concern for partially processed claims. Previous academic research has focused on conditions under which consumers believe exaggerated claims (Kamins and Marks, 1987; Olson and Dover, 1978; Rotfeld and Rotzoll, 1980; Wyckham, 1985). Most research has shown that consumers do form beliefs based on specific puffed claims such as “the fastest headache relief possible.” Two explanations are offered for the deception. First, consumers believe the claims (Kamins and Marks, 1987; Rotfeld and Rotzoll, 1980; Shimp and Preston, 1981). “Consumers do not recognize them [puffed claims] for what they are …” (Shimp and Preston, 1981, p. 24). Second, consumers process the puffed claims as though they were fact and generate inferences on the basis of those facts (Holbrook, 1978; Shimp and Preston, 1981; Wyckham, 1987). “[Consumers] are prone to interpret evaluative claims [puffery] as constituting factual claims. It is a process of implication in which the consumer accepts an unstated factual claim as being the implied meaning of the stated evaluative claim” (Shimp and Preston, 1981, p. 24). In both cases, the concern is that consumers will make a mistake during the assessment of the accuracy of the claim. Again, Cartesian logic is implicit in the assumption that one representation of the claim will be made in memory. Are consumers unable to differentiate between fact and exaggeration? Is “the public more tolerant of advertising hyperbole and less inclined to counter argue than is the case with other message forms” (Shimp and Preston, 1981, p. 26)? Is it safe to assume that a necessary condition for forming beliefs on the basis of puffed claims is that the claim is judged to be credible? Instead of a Cartesian approach, perhaps the claim is initially represented in memory as true during the comprehension process, affecting other associated beliefs. After determining that the claim is not credible, consumers are able to store the claim with a ‘not credible’ tag, but the corresponding evaluation will not be appropriately adjusted. 2.2. A Spinozian approach Gilbert et al. (1990) provide evidence that people initially represent false information as true. In their research, participants are asked to learn new propositions, both true and false. If participants were interrupted while processing the propositions, they rated a significantly greater proportion of both true and false statements as true, compared to participants that were allowed to completely process the propositions. The result is interpreted as support for the Spinozian explanation. Gilbert and colleagues found the result to hold even when the participants were forewarned that some of the information presented would be false. Perhaps consumers temporarily believe puffed claims as part of the comprehension process. Even though consumers may expect some advertising to lack credibility, they initially represent the claim as true. In the process of determining whether the claim is true, an accepted form of the claim is represented in memory. If the claim is a generalized exaggeration such as “the very best restaurant in town” then the positive global evaluation may be later stored with a ‘not credible tag,’ but is not ignored
729
during the evaluation procedure. More specific attribute beliefs may not be affected as they are at a different level of abstraction (Hawkins et al., 2001). The prediction that a global evaluation of the brand will be affected by puffery, even though the claim is ultimately determined to be untrue, is consistent with a larger body of research that finds that people are particularly poor at rejecting, ignoring or failing to believe what they have comprehended (Bjork, 1972; Schul and Burnstein, 1985; Wyer and Budesheim, 1987), even when they are forewarned that the information will be false (Gilbert et al., 1990). The outcome of this comprehension process is also consistent with dissociations found between perceptions of truth and belief (Begg et al., 1992). 2.3. Remembering exaggerated claims Even though consumers may be able to correctly unaccept a puffed claim, it has been shown that thinking about and negating a proposition can lead to false memory (Mayo et al., 2004). Feidler et al. (1996) found that after correctly negating a fact and following a filler task, respondents were more likely to falsely recognize the fact in its non-negated form than they were to recognize unpresented facts. This is counter to the Cartesian explanation. Because under Cartesian logic only the negated claim would be stored in memory, non-negated facts could not be retrieved from memory. If the Spinozian model is correct, then factual claims should be better remembered than puffed claims because there is no alternative representation to cause interference during retrieval. However, if the processing of the exaggerated claims is interrupted, then the puffed claims should be as well recognized as the factual claims because the nonnegated fact is the only representation available in memory. 3. Hypotheses Consumers are able to identify generalized puffed claims as less credible than factual claims if processing is uninterrupted (H1). However, even after identifying the generalized puffed claims as less credible, they are still more positive in their evaluation of the brand associated with the puffed claim (H2a), but no change occurs in their specific brand beliefs (H2b). Support for the Spinozian explanation is expected. When processing is interrupted, puffed claims are rated as credible as the factual claims (H3). Consumers are better able to remember the factual claim compared to the puffed claim in an uninterrupted processing condition (H4a). However, the memory advantage for factual claims should disappear if processing is interrupted because the negation process will not occur (H4b). 4. Methodology 4.1. Design and manipulations Sixty undergraduate students at an Australian university participated in the study for course credit. The design of the study was 2 × 3. The processing at encoding was a between
730
E. Cowley / Journal of Business Research 59 (2006) 728–734
subject factor with 2 levels (uninterrupted processing, interrupted processing). The level of puffery was a within subject factor with 3 levels (a ‘the very best’ claim, a ‘the ultimate’ claim, a factual (no puffery) claim).
finished typing their student identification number and pressed the ‘return’ key, they did not return to the screen with the target advertisement. Instead, they advanced to the next task of rating of the credibility of the claims in the advertisement without any further processing of the claim.
4.2. Puffery manipulation 4.4. Dependent measures Each participant saw three target advertisements (one for the Harbor Bistro, one for the Alternative Bar, and one for the City Cruise Line), one advertisement at each level of puffery. The two puffery levels were taken from Preston (1996; 1998), who identified ‘the very best’ and ‘the ultimate’ as two of six levels of puffery. The advertisement for the Harbor Bistro claimed the bistro was ‘the very best bistro in Sydney’, ‘the ultimate dining experience’ or the factual claim of ‘dining with a harbor view.’ The advertisement included a photo of the interior of the bistro with a view of the harbor; this ensured that participants could identify the factual claim as factual. The advertisement for the Alternative Bar claimed the bar was ‘the very best club in Sydney’, ‘the ultimate club experience’ or the factual claim of ‘music in the city.’ The advertisement included a photo of the interior of the bar with a view of the stage with musical equipment to ensure that the participant knew the factual claim was factual. The advertisement for City Cruise Line claimed the cruise was ‘the very best cruise in Sydney’, ‘the ultimate harbor experience’ or the factual claim of ‘see Sydney from the harbor.’ The advertisement included a photo of the interior of the boat with a view of the harbor including the Harbor Bridge and Sydney Opera House. All of the target ads advertised hypothetical brands. 4.3. The encoding manipulation The study was run in a computer lab. Each participant viewed the stimuli on a 17″ computer terminal. All participants were warned that previous participants had experienced some technical difficulties with the software when reading the advertisements. The participants were told that “if a screen disappears, please just follow the instructions as closely as you can.” Half of the participants were allowed to view the target ads for as long as they required before making the credibility and liking judgments without interruption. The other half of the participants were allowed to view the three target ads for 500ms before being ‘interrupted’ by the screen disappearing. A pretest (n = 20) had revealed that 500ms allowed participants to read the screens but not to elaborate on the information in the copy, or to question the claims. A new screen appeared which included an apology and instructed the participant to type their student identification number in a pop-up box in order to continue. Students were forced to use the numbers above the letters on the keyboard instead of the numbers on the number pad to the right of the keyboard. Not using the number pad was important because many students can touch-type their student number on the number pad. Using the keys at the top of the keyboard meant that they had to concentrate on the location of each number key. This provided the desired interruption to the processing of the claim. When the participants
4.4.1. Credibility Directly after viewing the ads, participants were asked to provide a credibility rating for the advertisement. Participants used a 10 point scale anchored with ‘not at all credible’ (0) to ‘very credible’ (9). The instructions encouraged participants to use the entire scale. Although participants were invited to ask questions if they did not understand the instructions, no questions were directed to the study administrator. 4.5. Belief and evaluation ratings After rating the credibility of the ad, the participants were asked to predict the probability of receiving good service and good food, and whether they believed the bistro/bar/cruise would be expensive. Participants used a 10 point scale anchored with ‘not at all likely’ (0) to ‘very likely’ (9). They could find some information in the advertisements about the food and service. Price information was not available, participants had to infer the expense involved with each establishment. Finally, participants were asked for an overall evaluation of the bistro/ bar/cruise on a 10 point scale anchored with ‘not at all good’ (0) to ‘very good’ (9), and to indicate how sure they were in their response on a 10 point scale anchored with ‘not at all certain’ (0) to ‘very certain’ (9). 4.6. Recognition The participants were asked to identify the claim that they saw earlier for each of the target ads. They were presented with all three claims (one old claim seen at study, two new claims) and a “don't remember” option. 4.7. Procedure Participants signed up for a study called the Advertising Study which ostensibly was designed to investigate the relationship between leisure activities and attitudes toward advertising. They were informed that they would see copy for both existing and potential ads. In the introduction they were told that they would be viewing 10 advertisements. They would first view the ad, and then answer some questions about their impressions of the credibility of the ad, the performance of the product or service, and their overall evaluation of the brand. Participants were then warned about the ‘problem’ with the software. Before the first target advertisement, two filler ads were rated for perceived credibility, overall evaluation of the brand and two specific brand beliefs. All participants experienced the ‘disappearing screen problem’ with the software for the second filler advertisement. Experiencing the interrupted processing early in
E. Cowley / Journal of Business Research 59 (2006) 728–734
5. Results 5.1. Credibility ratings Hypotheses 1 and 3 are tested with the credibility ratings. Hypothesis 1 asserts that consumers are able to recognize the reduced credibility of an exaggerated claim. Evidence in support of the hypothesis will be claimed if participants rate the puffed claims as less credible than the factual claims when they had time to process the claims completely. An ANOVA run on credibility ratings in the uninterrupted condition reveals a significant main effect for level of puffery (F(2, 84) = 6.42, p b .01). Participants were able to identify that the puffed claims were less credible than the factual claims (fact = 7.21, ultimate = 6.07, very best = 5.07). Hypothesis 1 is supported. Note that participants did not believe, on average, that the claims were not credible. Therefore, government policymakers would be correct if they assumed that consumers can identify that puffed claims as less credible, but they are wrong to assume that consumers dismiss puffery as not credible. If comprehension and acceptance are part of the same process, then the interruption manipulation should affect credibility ratings such that no differences are found in the assessment of the
Credibility
8 7 6 5 4
Fact
Low Puffery
High Puffery
Interruption
7.26
6.84
6.94
No Interruption
7.21
6.07
5.07
Level of Exaggeration
Fig. 1. Credibility ratings.
8.00 Evaluation
the experimental session was critical because all participants were motivated to try and read as much as they could of the copy as soon as the advertisement appeared on the screen. The third advertisement was the first target ad. One third of the participants saw a factual claim, one third saw a slightly exaggerated claim, and one third saw a highly exaggerated claim. Half of the participants were allowed to process the ads without interruption, the other half had their processing interrupted after comprehension. Participants saw two more target ads amongst the next four ads and then three filler ads. The same ratings were made for each ad. If the first ad included a very best claim then the second was an ultimate claim and the third was a factual claim. Unrelated filler questions about the leisure activities of the participants followed. The questions were included to support the cover story and provide time to clear short term memory. Ten minutes later participants were asked to recognize the ad claims for the three target ads. Participants were asked to guess the hypotheses of the study and then were debriefed. None of the participants suspected that the study was testing the effect of exaggerated claims.
731
7.00 6.00 5.00 4.00 Fact
Low Puffery
High Puffery
Interruption
6.10
6.07
7.35
No Interruption
6.52
6.84
7.34
Level of Exaggeration
Fig. 2. Overall evaluation.
credibility of a fact versus puffed claim in the interrupted condition. The acceptance of all claims as credible was hypothesized in this condition because participants did not have a chance to unaccept a decidedly less credible claim. An ANOVA run on the entire sample reveals a significant interaction between puffery and interruption condition (F(2, 174) = 3.63, p b .05). A separate ANOVA was run on credibility ratings in the interrupted condition to investigate the interaction. The ANOVA reveals no significant main effect for level of puffery (F(2, 90) = 1.12, p N .10). Participants were not able to identify that the puffed claims as less credible than the factual claims (fact= 7.26, low puffery = 6.84, high puffery = 6.94). Hypothesis 3 is supported, see Fig. 1. The data support the Spinozian hypothesis that the process of comprehension is not independent from the belief decision. When processing of the claim was interrupted, the merely comprehended claims were believed to be true. 5.2. Overall evaluation Participants were asked to provide an overall evaluation of the bistro/bar/cruise and to indicate how certain they were about their response. Hypothesis 2a predicts that the inherently positive nature of a generalized puffed claim such as “the very best in town” will influence the overall evaluation of the brand. An ANOVA was run on the overall evaluation of the brand with the level of puffery and the interruption condition as the independent variables and the credibility rating as a covariate. The results reveal a significant main effect for puffery (F(2, 174) = 6.85, p b .01), but not for interruption (F(1, 174) = 2.41, p N .10). Consumers in both conditions rated brands with puffed claims as more favorable overall (fact = 6.31, low puffery = 6.71, high puffery = 7.35). Planned comparisons reveal that the “very best” claim was significantly higher than the other claims in the interrupted processing condition, and higher than the fact claim in the uninterrupted condition. A possible explanation for the mixed result for the “ultimate” claim may be that it is a more complicated claim than the “very best” which is easier to understand. The credibility covariate was weakly significant (F(1, 174) = 3.35, p b .10). Perceived credibility of the claim has a positive effect on the overall evaluation of the brand. Hypothesis 2a is supported, see Fig. 2. A separate ANOVA run on the certainty ratings for brand evaluations with puffery and interruption condition as the independent variables revealed a significant effect for the interruption condition (F(1, 174) = 3.99, p b .05), but no significant effect for the level of puffery. The significant effect for the
E. Cowley / Journal of Business Research 59 (2006) 728–734
interruption condition is interesting because evaluation confidence was higher in the interrupted condition (interrupted = 6.85, uninterrupted = 6.30). This result is consistent with recent evidence that a reduction in attitude confidence can follow the counter argumentation associated with negation of a claim (Petty et al., 2004). The null effect for the level of puffery is interesting because it indicates that an evaluation based on a factual claim is held as confidently as an evaluation based on an exaggerated claim. The result is consistent with previous research finding a low correlation between accuracy and confidence in judgments by consumers (Alba and Hutchinson, 2000; Cowley, 2004). 5.3. Attribute beliefs As expected, the attribute beliefs (good food, good service) and the inference (being expensive) did not vary with puffery level or interruption condition. ANOVAs did not reveal main effects for puffery or interruption or interaction effects. This was predicted in H2b because of the general nature of the puffed claims. Hypothesis 2b is supported, see Fig. 3. The absence of a change in specific beliefs does not support or negate the Spinozian hypothesis, instead the null effects suggest that the specificity of a puffed claim may play a role in generating inferences and changing specific attribute beliefs. Previous work has reported that specific claims affect specific beliefs. Rotfeld and Rotzoll (1980) find that almost half of the consumers in their study believed the puffed claims, but a much reduced, yet still evident percentage believed the puffery implied claims. Rotfeld and Preston (1981) and Shimp (1978) also use specific claims and find that consumers expanded on the claims, possibly because of the multiple interpretations allowing for inference generation. The effects are not found here with generalized exaggeration. The difference in results depending on the specificity of the claims themselves is consistent with related work where plausible specific claims have been found to affect belief in those claims, but that general claims do not affect specific beliefs (Hawkins et al., 2001). 5.4. Memory for the claim Hypothesis 4a states that participants in the uninterrupted condition should be better able to recognize factual claims. The data indicate that there is a difference in memory performance for puffery versus factual claims (χ2(1) = 3.87, p b .05). As ex-
% Not Remembering
732
5
Fact
Low Puffery
High Puffery
Food
5.68
5.75
6.14
Service
6.65
6.74
6.77
Expense
7.07
7.07
7.24
Level of Exaggeration
Fig. 3. Attribute beliefs.
15.00 10.00 5.00 0.00
Fact
Low Puffery
High Puffery
16.13
22.58
19.35
No Interruption
6.90
3.45
3.45
Level of Exaggeration
Fig. 4. Don't remember responses.
pected, participants were more accurate when remembering factual claims (factual = 85% correct, puffery = 64%) when processing was not interrupted. Therefore, H4a is supported. The memory advantage for factual claims was hypothesized because there was no negation process which may cause interference during retrieval. Hypothesis 4b suggests that the memory disadvantage for exaggerated claims should disappear in the interrupted processing condition because the negation process does not occur. The results show no differences in memory performance in the interrupted condition (χ2(1) = .26, p = .61). Although this may seem surprising, a number of influences are at work. First, since the interrupted group had only 500ms to look at the ad, they may have more trouble recognizing the claim and less confidence in their ability to identify the correct claim. On the other hand, the puffed claim may have been the only thing taken away from the exposure because it was prominent: increased attention to the claim may have reduced interference from other aspects of the ad. Second, part of the rationale behind H4b was that in the uninterrupted condition the dual representation of the claim may have caused interference during retrieval resulting in an increase in erroneous recognition judgments. One result that provides evidence consistent with a lack of processing account in the interrupted condition is the number of “don't remember” responses. See Fig. 4. Participants in the interrupted condition more frequently reported that they did not believe they knew the answer. These results are not significant, however. A result consistent with a negation tag account is that participants in the uninterrupted condition were more accurate when remembering factual claims versus the exaggerated claims. Although false recognition went up slightly in the interrupted condition for factual claims (possibly because of the limited processing time), it was less likely in the interrupted condition,
% Incorrect
Ratings
6
20.00
Interruption
8 7
25.00
40.00 35.00 30.00 25.00 20.00 15.00 10.00
Interruption No Interruption
Fact 22.58 13.80
Low Puffery 29.00 34.50 Level of Exaggeration
Fig. 5. Mistaken recognition.
High Puffery 22.50 34.50
E. Cowley / Journal of Business Research 59 (2006) 728–734
see Fig. 5. Taking the results of H4a and H4b together, it appears that the process of negating the initial acceptance of the claim affected recognition memory for the puffed claims. 6. Conclusions
733
this data how much of the memory performance difference is based on the negation process and how much is attributable to the encoding time itself. Future research is required to investigate this hypothesis because it is important to know under what circumstances the puffed claim is eventually forgotten.
6.1. General discussion 6.2. Implications The results provide evidence that consumers are able to identify puffed claims as less credible, however, the claims are not judged to be not credible. Only 1 in 180 responses rated the puffed claim as 0 or “not at all credible.” This is counter to previous findings with the exception of Wyckham (1987) who found upwards of 80% of consumers stating that puffed claims were not true. One explanation is that the generalized puffed claims used here are less concrete and more difficult to discredit. More specific claims such as those used in Wyckham's study (1987) refer to a single attribute or benefit, allowing the consumer a better opportunity to mentally test its credibility. Even though consumers can identify a puffed claim as less credible, they still rated the brand more favorably than brands associated with a factual claim. The general finding, that puffery is deceptive because consumer beliefs do change after exposure to puffed claims, replicates a number of past studies (Holbrook, 1978; Kamins and Marks, 1987; Olson and Dover, 1978; Rotfeld and Rotzoll, 1980; Wyckham, 1985) and supports assertions made in research review papers (i.e. Rotfeld and Preston, 1981). The explanation in previous research was that consumers believe the claims and generate inferences on the basis of the implied facts. No support is found here for these explanations. Consumers were able to identify puffery as less true and they did not generate inferences about the quality of food and service or the expense of the bistro/bar/cruise. The explanation offered here is that during the process of comprehending a puffed claim, the initial representation in memory is that the claim is true. Consumers appear to accept the claim initially, and then unaccept or discount the claim after realizing its lack of credibility. This is evidenced by the increased number of puffed claims accepted as credible in the interrupted processing condition. Processing was interrupted before the unaccepting or discounting process was complete. Associations from the initial representation to other brand beliefs may be established in memory, possibly affecting brand evaluations and other generalized beliefs about the brand. After a claim is identified as lacking credibility a negation tag is added to the representation, but other alterations to associated beliefs are not re-adjusted, or at least not completely. Consequently, knowledge that the claim is lacking credibility is stored, but remnants of the claim's persuasive impact linger. Finally, memory for the puffed claim is not as accurate as memory for the factual claims in the uninterrupted condition, but improves in the interruption condition. The explanation is that the memory representation resulting from the negation process can interfere with retrieval. This result is very limited because the processing time was different across conditions confounding the manipulation. It is impossible to establish with
The implications for policy makers are clear. Exposure to incorrect information affects beliefs even when it is identified as lacking credibility. Although puffery defined by policymakers as “wildly exaggerated, fanciful or vague claims for a product or service,” can be identified as less credible, consumers still alter their evaluations of the brand after exposure. As mentioned above, after some time has past, consumers do not recognize the claim. This compounds the deceptive danger of exposure to puffed claims for consumers because they can not correct the inflated overall evaluation at retrieval with the memory that the claims were not credible. One logical suggestion is to forewarn consumers of the effects of exposure to puffed claims as a strategy to safeguard them. The problem with this suggestion is that warnings of impending false information do not seem to alter this process (Gilbert et al., 1990). The implication for marketers is also clear. Puffery works. A legally accepted, but misleading claim, results in consumers liking the brand better and not remembering the exaggeration specifically. Given that the advertising environment is extremely cluttered reducing the amount of processing even in an uninterrupted situation (Gardener et al., 1978), consumers may be even more likely to accept puffed messages in a more natural setting because of the possible interference in the unaccepting process. This would explain the findings of Rotfeld and Rotzoll (1980) who tested brand beliefs based on exposure to puffed claims in ‘natural’ advertising settings and found very high levels of belief amongst a sample of consumers (from 42% to 69%). 6.3. Future work The brands used here were hypothetical. The participants had no knowledge of these brands. The effect of puffery on more established brands and the process by which a claim for a wellknown brand is processed may be different than the process investigated here. No current evidence in cognitive psychology research suggests that this would be true, but it remains possible. Perhaps the most promising avenue for future work is to investigate whether puffed claims heard after product trial will change a consumer's memory and evaluation for their own past experiences with the brand. Braun (1999) has shown that misleading advertising has this effect, but Cowley and Janus (2004) have shown that blatantly misleading advertising can have the opposite effect by improving memory. It would be interesting to know what effect puffed claims have on a consumer's memory for an experience, particularly since policymakers categorize puffery as blatantly misleading, but consumers do not.
734
E. Cowley / Journal of Business Research 59 (2006) 728–734
References Alba JW, Hutchinson JW. Knowledge calibration: what consumers know and what they think they know. Journal of Consumer Research 2000;27 (2):123–56. Begg IM, Anas A, Farinacci S. Dissociation of processes in belief: Source recollection, statement familiarity, and the illusion of truth. Journal of Experimental Psychology: General 1992;121(4):446–58. Braun KA. Postexperience advertising effects on consumer memory. Journal of Consumer Research 1999;25:319–34 [March]. Bjork RA. Theoretical implications of directed forgetting. In: Melton AW, Martin E, editors. Coding processes in human memory. New York NY: Wiley; 1972. p. 217–35. Cowley E. Recognition confidence, recognition accuracy and choice. Journal of Business Research 2004;57:641–6 [June]. Cowley E, Janus E. Not necessarily better, but certainly different: A limit to the advertising misinformation effect on memory. Journal of Consumer Research 2004;31:229–35 [June]. Feidler K, Wlahter E, Armbruster T, Fau D, Naumann U. Do you really know what you have seen? Intrusion errors and presuppositions effects on constructive memory. Journal of Experimental Social Psychology 1996;32:428–511. Gardener MP, Mitchell AA, Russo JE. Chronometric analysis: An introduction and application to low involvement perception of advertisement. In: Hunt HK, editor. Advances in Consumer Research, vol. 5. Ann Arbor MI: Association for Consumer Research; 1978. p. 581–9. Gilbert DT. How mental systems believe. American Psychologist 1991;46 (2):107–19. Gilbert DT, Krull DS, Malone PS. Unbelieving the unbelievable: Some problems in the rejection of false information. Journal of Personality and Social Psychology 1990;59(4):601–13. Gilbert DT, Tafarodi RW, Malone PS. You can't not believe everything you read. Journal of Personality and Social Psychology 1993;65(2):221–33. Hawkins SA, Hoch SJ, Meyers-Levy J. Low-involvement learning: repetition and coherence in familiarity and belief. Journal of Consumer Psychology 2001;11(1):1-11. Holbrook MB. Beyond attitude structure: toward the informational determinants of attitude. Journal of Marketing Research 1978;15(4):545–56.
Kamins MA, Marks LJ. Advertising puffery: the impact of using two-sided claims on product attitude and purchase intention. Journal of Advertising 1987;16(4):6-15. Mayo R, Schul Y, Burnstein E. “I am not guilty” vs “I am innocent”: successful negation may depend on the schema used for its encoding. Journal of Experimental Social Psychology 2004;40:433–49. Olson JC, Dover PA. Cognitive effects of deceptive advertising. Journal of Marketing Research 1978;15(1):29–38. Petty RE, Tormala ZL, Rucker DD. Resisting persuasion by counterarguing: an attitude strength perspective. In: Jost JT, Banaji MR, Prentice DA, editors. Perspectivism in social psychology: the yin and yang of scientific progress, APA science series. APA decade of behavior series. Washington, DC, US: American Psychological Association; 2004. p. 37–51. Preston IL. The great American blowup: puffery in advertising and selling. Madison WI: The University of Wisconsin Press; 1996. Preston IL. Puffery and other ‘loophole’ claims: how the law's ‘don't ask, don't tell’ policy condones fraudulent falsity in advertising. Journal of Law and Commerce 1998;1:49-114. Rotfeld HJ, Rotzoll KB. Is advertising puffery believed? Journal of Advertising 1980;9(3):16–20. Rotfeld HJ, Preston IL. The potential impact of research on advertising law. Journal of Advertising Research 1981;21(2):9-17. Schul Y, Burnstein E. When discounting fails: conditions under which individuals use discredited information in making a judgement. Journal of Personality and Social Psychology 1985;49:894–903. Shimp TA. Do incomplete comparisons mislead? Journal of Advertising Research 1978;18:21–8 [December]. Shimp TA, Preston IL. Deceptive and nondeceptive consequences of evaluative advertising. Journal of Marketing 1981;45(1):22–32. Wyckham RG. Implied superiority claims: parody parading as superiority. Proceedings of the 12th International Research Seminar in Marketing, La Londe Les Maures 1985:360–87. Wyckham RG. Implied superiority claims. Journal of Advertising Research 1987;27(1):54–63. Wyer RS, Budesheim TL. Person memory and judgments: the impact of information that one is told to disregard. Journal of Personality and Social Psychology 1987;53:14–29.
Journal of Business Research 59 (2006) 735 – 744
Responsiveness to authority appeals among young French and American consumers Jae Min Jung a,⁎, James J. Kellaris b a
Department of Marketing, College of Business Administration, North Dakota State University, Fargo, ND 58105-5137, United States b University of Cincinnati, United States
Abstract The “authority effect” is a powerful social influence principle frequently used in advertising to increase compliance. Young adult consumers, however, often resent authority figures. Such resentment can result in negative reactions to authority-based persuasion attempts. This study examines the differential responsiveness to authority appeals among young adults in France and the U.S., as well as the boundary conditions within which such differential responses occur. Results show that before 9 / 11 Americans had more positive attitudes when the spokesperson in an ad is low (vs. high) in authority. This reverse authority effect did not obtain among French subjects, who appear to prefer recommendations from social equals, or among Americans after 9 / 11. Perceived source credibility and power distance moderate the effect of authority on attitudes and purchase intentions. © 2006 Elsevier Inc. All rights reserved. Keywords: Social influence; Cross-cultural differences; Authority appeals; Advertising; Power distance; Source credibility
1. Introduction The “authority principle” refers to the tendency of individuals to comply with the recommendations or directives of authority figures, and is a fundamental social influence principle (Cialdini, 2001). Authority figures may vary across cultures; however, in general, they are the ones who have acquired status through education, experience, talents, or other means (Cialdini, 2001). For instance, in work places, superiors may be regarded as authority figures. Parents can be regarded as authorities by children. High ranking public officials are often regarded as authority figures. Perhaps the classic example is the controversial Milgram (1963) study, which demonstrated the shocking extremes to which individuals will go to obey authority. In the social influence literature, numerous studies have documented authority effects (e.g., Bickman, 1974; Bushman, 1988; Michener and Burt, 1975; Rasinski et al., 1985). However, authority-based appeals have received scant attention in advertising research. Instead, more efforts have been focused on celebrity endorser effects (e.g., Goldsmith et al., 2000; Heath et al., 1994; Kamins et al., 1989; MacCracken, ⁎ Corresponding author. Tel.: +1 701 231 8129; fax: +1 701 231 7508. E-mail address: [email protected] (J.M. Jung). 0148-2963/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2006.01.011
1989). This could be attributed in part to the prevalent use of celebrity endorsers in practice (Agrawal and Kamakura, 1995) and to American cultural traits that promote admiration of successful persons, yet indifferent treatment of high (vs. low) power position holders (Hofstede, 2001). The lack of research on the authority principle in advertising stands in sharp contrast to frequent use of authority appeals in advertising practice. For example, General Mills uses mothers as spokespersons for “Kix” cereal. McNeil Consumer and Specialty Pharmaceuticals features moms for their “Motrin,” children's fever reducer. In France, work place superiors and teachers are often used as spokespersons (Mooij, 1998). In all of these cases, spokespersons are featured as authorities upon whom consumers can rely. Despite the widespread use of non-celebrity authority figures in global advertising, little is known about their effectiveness and their applicability across different segments and cultures. Thus, this study seeks to address these issues using data collected from young adults in France and the U.S. 2. Background and hypotheses We identified three factors that are important to the effectiveness of authority-based advertisements. Under certain circumstances, authority figures may not be as persuasive as the
736
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
authority principle would suggest. The effectiveness of authority may reside 1) within contextual factors (e.g., national culture), 2) within the authority figures themselves (e.g., source credibility), and/or 3) within decoders of messages (e.g., their orientation to power distance). First, national culture is expected to moderate the authority effect. Cross-cultural social influence research suggests that responses to social influence strategies may also vary across national groups, depending on the fit between relevant dimension(s) of national culture and the cultural value that the influence strategy is intended to evoke (Cialdini et al., 1999; Jung and Kellaris, 2004). The authority principle, for example, may be closely related to the “power distance” (PD) dimension of national culture, which is defined as “the extent to which the less powerful members of institutions and organizations within a country expect and accept that power is distributed unequally” (Hofstede, 1991). Consequently, authority figures that influence audiences in high power distance cultures may not exert as much influence in low power distance cultures. Second, source credibility is expected to moderate the influence of authority. Consistent with previous source credibility research (e.g., Grewal et al., 1994), we define credibility as “expertise” and “trustworthiness.” For authority figures to exert power, they must have credibility (Cialdini and Rhoads, 2001). Lastly, values held by audience members are expected to moderate authority effects. Hofstede's seminal work (1991, 2001) and other studies (e.g., Singelis, Triandis, Bhawuk, and Gelfand, 1995; Triandis, 1995) in the cross-cultural psychology literature provide some insights into how individual values may affect the way individuals perceive authority. In summary, we expect authority effects to be moderated by national cultures, spokesperson characteristics, and audience characteristics. In this study, we first examine how different levels of authority influence consumer attitudes and purchase intentions. We further investigate how these authority effects are moderated by national culture, source credibility, and individuals' orientations to power distance. Fig. 1 summarizes our conceptual framework. 2.1. Authority vs. reverse authority effects In general, the authority principle works such that the higher the level of authority portrayed in an ad, the more positive attitudes (towards ads/brands) and purchase intentions should be. Under certain circumstances, however, authority may exert a negative influence. For example, individuals who resent or distrust authority should react negatively to authority-based persuasion attempts. Resentment of authority may be more common among young people who rebel against conventional societal values or among those who question authority as part of the critical world view encouraged by higher education, or among young adults who struggle to gain independence from parents. Moreover, in so much as authority represents an encroachment on the sovereignty of the individual, members of western cultures that place a high value on individual freedom may react negatively to authority. Resentment of authority may be a corollary of valuing freedom. Sources with less authority may
Individual Values (Power Distance) H4
Source Credibility H3
AUTHORITY of Ad Spokesperson
Attitude Towards Ads Attitude Towards Brands Purchase Intentions
H1 H2
Culture (National Level)
Fig. 1. Conceptual model. Dotted lines represent the relationship that are not formally hypothesized in this study.
be perceived as more trustworthy, because their motivations for making a recommendation may seem different from those of an authority figure. The low authority figure bears a greater social risk in making a recommendation and thus may elicit more positive attitudes. Hence young adults may exhibit a reverse authority effect, such that: H1. The higher the level of authority in an advertisement, the less positive young consumers' attitudes and purchase intentions. 2.2. Impact of national culture (France vs. the U.S.) Despite the universal use of authority across cultures, the degree of responsiveness to authority may vary (Manrai and Manrai, 1996). Cialdini et al. (1999) recently examined two social influence strategies (“social proof” and “commitment/ consistency”) in collectivist versus individualist countries (Poland, USA). Both strategies were found to be effective in both countries, however differentially so. A persuasive appeal based on social proof was more effective among collectivists and an appeal based on commitment/consistency more effective among individualists. Jung and Kellaris (2004) recently examined the differential responsiveness of French versus U.S. consumers to scarcity-based ad appeals. They found a positive effect of scarcity on purchase intent, with greater proneness to scarcity effects among low context American (vs. high context French) subjects. In a study of conflict resolution, Tyler et al. (2000) found cross-national differences in reactions to authority that were explained by different levels of PD at the national level. Similarly, we expect the authority principle to manifest differently among young adults across two nations as a function of cultural values (PD). France is classified as a high PD country and the U.S. as a low PD country (68 vs. 40 in PD index, Hofstede, 2001). In high PD countries, parents teach children obedience; teachers initiate communications in school; many levels exist in organizations; and decision making is centralized (Hofstede, 2001). The opposite norms are expected in low PD countries. Parents treat children as equals; students can freely initiate some communication in class; organizational structure is flat; and decision making is decentralized (Hofstede, 2001). Mooij (1998) pointed
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
how these cultural differences in PD dimension are manifested in advertising and communication strategies adopted by marketing professionals in the two countries. For example, in French advertising, the elder advises the younger (e.g., mother advises daughter). In the U.S., the younger often advises the elder (e.g., daughter advises mother). In a high PD culture, the source of communication (e.g., authority vs. common people) is important as opposed to reasoning and argument (Mooij, 1998). Further, strong boss–subordinate relationships tend to exist in high PD countries. For these reasons, one might expect a positive authority effect to be observed in a high PD country (e.g., France) and a reverse authority effect observed in a low PD nation (e.g., the U.S.). However, young adults may react more negatively to the authority-based influence attempt because their PD level is lower than that of older groups in each country (Hofstede, 2001). Thus, on the basis of Hofstede (2001) and Mooij (1998)'s work, we anticipate that:
737
tolerance of differences in hierarchy (Hofstede, 1991, 2001). High PD individuals are reluctant to refuse a request from or disagree with authority figures. They give priority to the opinions of people in authority. This suggests that those who are high in PD are more likely to accept and less likely to discount message arguments, whereas those who are low in PD are less likely to accept and more likely to discount the message arguments. Therefore, PD is expected to moderate the effectiveness of authority appeals such that individuals characterized by greater PD will be more prone to persuasive effects of authority appeals. H4. PD moderates the authority effect such that the reverse authority effect is less among high (vs. low) PD, young consumers. 3. Study 1 3.1. Methods
H2. The reverse authority effect is more prominent among young U.S. consumers than among their French counterparts. 2.3. Impact of credibility Spokesperson's source credibility (CRED) in ads is defined as the degree to which the spokesperson is perceived to have expertise on a subject and is trusted to provide an impartial opinion about the subject (Dholakia and Sternthal, 1977; MacCracken, 1989). Lack of either expertise or trustworthiness is likely to undermine the credibility of an authority figure's advocacy (Cialdini and Rhoads, 2001). Attractiveness (refer to MacCracken, 1989), which is often included to define celebrities' credibility (e.g., Goldsmith et al., 2000), is excluded in our conception of source credibility because our focus is on non-celebrity authority figures. In the advertising literature, the impact of message source credibility on ad effectiveness has long been established (e.g., Dholakia and Sternthal, 1977; Harmon and Coney, 1982). When consumers decode an advertising message, source credibility positively affects consumers' likelihood of message acceptance (Mizerski et al., 1979). Taking source credulity as a moderator, Grewal et al. (1994) showed that positive effect of price on perceived performance risk is reduced when perceived source credibility is high (vs. low). We propose source credibility as a moderator of authority's effect on attitudes (or purchase intention): the more credible an authority figure is perceived to be, the greater the impact of the authority appeals. H3. Spokespersons' credibility moderates the authority effect such that the reverse authority effect is lower among young consumers perceiving spokespersons as more (vs. less) credible. 2.4. Impact of power distance Power distance refers to the extent to which individuals expect and accept the unequal distribution of power, and
An experiment involving young French and the U.S. participants varied spokesman authority (low, mid, high) using printed descriptions of radio ads. Subjects' nationality, perceptions of source credibility, and PD were measured. Dependent variables were attitudes toward the ad, advertised brand, and purchase intention. 3.1.1. Subjects Subjects were two hundred forty-eight (N = 248) students at universities in the United States (n = 130) and the lower Loire Valley and mid-Pyrenees regions of France (n = 118). One hundred thirty-five were male (54.4%) and 113 (45.6%) female. Ages ranged from 20 to 56 with an average age of 22.9 years. The French sample was slightly older on average than the American sample (23.9 vs. 22.0 years) and had a slightly higher proportion of males; the samples were matched in other important respects. 3.1.2. Stimuli The stimuli were printed descriptions of radio ads in a dialog format. The relationship of the primary spokesperson to the recipient of a recommendation was varied to represent high, mid, and low levels of authority. Three sets of ads were developed to enhance internal validity. In the first (second/ third) ad, a yogurt (software/internet retailer) was recommended by a mother (teacher/boss) to a daughter (student/ subordinate) under the high authority condition, by one woman (person/one employee) to another woman (person/ employee) under the mid authority condition, or by a daughter (student/subordinate) to a mother (teacher/boss) under the low authority condition. The (English) text of the ads appears in Appendix A. 3.1.3. Measures Attitude toward the ad (Aad) was measured via four, sevenpoint items preceded by instructions stating “The following items concern the ad itself, not the product in the ad” and by a
738
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
prompt “The radio ad I imagined was…” The items, taken from Madden et al. (1988) were labeled pleasant/unpleasant, likeable/unlikable, interesting/boring, and good/bad. Responses were summed and averaged to form a composite scale (αCombined = .88; αUS = .89, αFrance = .86). Attitude toward the brand (Ab) was measured using the same four items, preceded by instructions stating “The following questions concern the product represented in the ad” and by a prompt “The product in the ad I imagined was…” Alpha reliability for the four item composite scale averaged .88 (αUS = .91, αFrance = .86). Purchase intent (INT) was measured via a one-item, sevenpoint Likert-type scale (very likely = 7, very unlikely = 1) preceded by the prompt “If it were available in your market, what is the likelihood that you might purchase the advertised product within the next twelve months?” and was summed across ads and averaged. Source credibility, taken from Harmon and Coney (1982), was measured via six-item, seven-point semantic differential scale preceded by the prompt “The spokesperson (party) in the ad was…” In each case the party was specified (e.g., mother for high authority, women for mid authority, or daughter for a low authority condition in ad #1). The items were labeled trustworthy/not trustworthy, open-minded/not open-minded, good/bad, expert/not expert, experienced/not experienced, trained/untrained. The second item was dropped due to a low factor loading (αCombined = .82 across the three ads; αUS = .81, αFrance = .83). Power distance scale was adapted (4 items) from Pretest six of Jung (2002). One item that had factor loading of .26 in CFA was dropped (“Opinions of superiors should be given priority over those of my peers”). Further, six congeneric scale items (Nunnally and Bernstein, 1994) were created to adequately tap the domain of the construct. Thus, the nine items were administered to the samples and two items were further dropped due to low item to total correlation. The resulting seven item PD scale (See Appendix B) produced a unidimensional factor structure with Cronbach's alpha of .86 (αUS = .89, αFrance = .83).
Two additional cultural value scales were included for discriminant validity assessment purposes. Vertical Collectivism (VC) is a cultural pattern in which one sees oneself as part of the collective and accepts inequalities among group members; Vertical Individualism (VI) is a cultural pattern in which individuals are viewed as autonomous, but they are considered different in their status (Singelis et al., 1995). Both constructs were measured originally using eight item scales (Singelis et al., 1995). After one item (Children should feel honored if their parents receive a distinguished award) was dropped, the remaining seven-item VC scale had a Cronbach's alpha of .69. For VI, one item was deleted (Some people emphasize winning; I'm not one of them) due to low factor loading. The remaining seven-item scale had a Cronbach's alpha of .85. Demographic data (sex, age, nationality) were collected to facilitate sample description. 3.1.4. Translation and procedure A French version of the stimuli and questionnaire was constructed via standard back-translation (Brislin, 1980), conducted by two independent, bilingual persons. Self-administered questionnaires containing the ad descriptions and measures were distributed to subjects in regular class sessions in return for course credit. French subjects received a French language version containing the ads. Each ad was followed by measures of attitude toward the ad, attitude toward the advertised brand, and purchase intent. Although each subject saw three ads, the manipulation of authority was completely between-subjects in order to avoid hypothesis guessing. 3.1.5. Integrity check To test the integrity of the treatments, a one-way ANOVA of authority on CRED was conducted for each of the three ads, for the American and French samples separately and combined. Results show a significant main effect of authority in ads 1 and 2 (p's b .05), but not in ad3 (p N .10) within as well as across the samples. Subsequently ad 3 was excluded from analysis. Tables 1 and 2 show correlations and descriptive statistics of variables when ad1 and ad2 are combined.
Table 1 Correlations and descriptive statistics in the combined data set
Aad Ab INT Sex Nation Age Auth Cred PD VI VC Mean SD
Aad
Ab
INT
Sex
1 .50⁎⁎⁎ .24⁎⁎⁎ − .02 − .42⁎⁎⁎ − .10 − .13⁎ .32⁎⁎⁎ .16⁎⁎ .32⁎⁎⁎ .27⁎⁎⁎ 4.12 1.04
1 .49⁎⁎⁎ .04 −.08 −.10 −.11 .37⁎⁎⁎ .17⁎⁎ .23⁎⁎⁎ .13⁎ 4.84 .82
1 .06 .09 − .06 − .07 .20⁎⁎ .06 .17⁎⁎ − .02 3.67 1.23
1 .13⁎ − .15⁎ .04 − .05 − .03 − .23⁎⁎⁎ −.07 1.46 .50
⁎b.05; ⁎⁎b.01; ⁎⁎⁎b.001 (two-tailed).
Nation
Age
Auth
Cred
PD
VI
VC
1 .19⁎⁎ .09 − .06 − .00 2.02 .82
1 .17⁎⁎ .26⁎⁎⁎ .23⁎⁎⁎ 4.41 .94
1 .07 .38⁎⁎⁎ 3.51 1.13
1 .34⁎⁎⁎ 4.39 1.11
1 3.43 .88
1
.23⁎⁎⁎ .05 − .05 − .15⁎ − .36⁎⁎⁎ − .35⁎⁎⁎ 1.48 .50
1
.16⁎⁎ −.06 −.13⁎ −.06 −.05 22.91 .82
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
3.2. Analysis and results To test our expectations concerning authority effects, we ran multiple regression analyses for each sample and for the combined sample. For dependent variables, overall average values of Aad, Ab, and INT, respectively, were calculated across ads 1 and 2 and were used in the analysis. For independent variables, we included authority (Auth), Nation, Cred, PD, VC, VI, and the interaction terms (Auth ⁎ Nation, Auth ⁎ Cred, Auth ⁎ PD, Auth ⁎ VC, Auth ⁎ VI). In addition, sex and age were included as control variables. All the independent metric variables were mean centered and Auth was zero centered to avoid multicollinearity problem (Aiken and West, 1991). Thus, each of the three dependent variables was regressed on the independent variables for the combined, American, and French data set, leading to nine regression models. Variables, “nation” and “Auth ⁎ Nation” were excluded from regression models involving single country data. Multicollinearity was not a problem since variation inflation factors were all less than 2.76 with an average of 1.47 across the regressions, which is well below the cutoff point of 10 (Neter et al., 1985). Table 3 shows results for the ad and brand attitude measures. 3.2.1. Reverse authority Results show a negative impact of authority levels on Aad (β = − .27, t = − 3.26, p b .001), Ab (β = − .26, t = − 2.99, p b .005) and INT (β = − .14, t = − 1.44, p = .075, one-tailed) for the combined sample. This means that the higher level of authority, the more negative the attitudes and purchase intentions of the young adults across cultures. Thus H1 regarding a reverse authority effect among young adults was supported. 3.2.2. National culture To test for a moderating influence of national culture on the reverse authority effect (H2), we used regression primarily and ANCOVA secondarily to shed light on the regression results. The multiple regression analysis revealed a marginally significant Auth ⁎ Nation interaction on Aad (βAuth × Nation = .13, t = 1.47, p = .07, one-tailed) and Ab (βAuth × Nation = .13, t = 1.30,
739
p b .10, one-tailed), but not on INT. Results suggest that reverse authority effects are slightly stronger among American (vs. French) participants. Standardized coefficients for Auth in each country attest to this interpretation. The reverse authority effect was present among the Americans (βAuth = − .34, t = − 3.49, p = .001 for Aad; βAuth = − .27, t = − 2.67, p b .01 for Ab; model not significant for INT), but not among the French (all three p's N .50). Next we ran a 3 (authority) × 2 (nation) between-subjects ANCOVA on Aad, with CRED as a covariate. Because CRED is an important factor determining persuasion (MacCracken, 1989) and spokesperson credibility might vary across cultures, CRED had to be controlled for to evaluate the differential effect of authority across cultures. However, we first confirmed homogeneity-of-slopes assumption that CRED be linearly related to attitudes and intentions in the same fashion across all the levels of authority and nations. Violation of this assumption would render ANCOVA invalid (Green and Salkind, 2003). Subsequently, we ran a two-way ANCOVA on Aad with CRED as a covariate. Results showed a significant covariate effect (F1, 238 = 43.04, p = .000), a reverse authority effect (MLow auth = 4.31, MMid auth = 4.12, MHigh auth = 3.88; F 2, 238 = 5.02, p b .01), nation effect (M American = 4.53, MFrench = 3.68; F1, 238 = 57.98, p = .000), and authority ⁎ nation interaction effect (F2, 238 = 4.59, p = .01, partial η2 = .04), confirming H2. Separate ANCOVA analyses were also performed for each country. Results showed that when the impact of CRED is controlled for, a reverse authority effect is clearly revealed among the U.S. participants (MLow auth = 4.98, MMid auth = 4.37, MHigh auth = 4.26; F2, 126 = 8.54, p = .000), but not among the French participants (MLow auth = 3.63, MMid auth = 3.87, MHigh auth = 3.50; F2, 111 = 1.82, ns), who seem to have responded most positively under equal authority conditions. However, an Auth ⁎ Nation ANCOVA on Ab and INT did not produce a significant interaction. Additionally, we ran a 3 (authority) × 2 (nation) between-subjects MANCOVA on Aad, Ab, and INT, with CRED as a covariate. Results show a marginally significant Auth ⁎ Nation interaction effect (Hotelling's Trace: F6, 468 = 1.93, p = .074, partial η2 = .02).
Table 2 Correlations and descriptive statistics in the French and the U.S. data
Aad Ab INT Sex Age Auth Cred PD VI VC Mean (U.S)++ SD
Aad
Ab
INT
1 .51⁎⁎⁎ .25⁎⁎ .05 .06 −.21⁎ .34⁎⁎⁎ .18⁎ .14 .17 4.54 .98
.53⁎⁎⁎ 1 .39⁎⁎⁎ .14 − .02 − .14 .33⁎⁎⁎ .20⁎ .15 .13 4.90 .84
.38⁎⁎⁎ .63⁎⁎⁎ 1 .17 − .01 − .10 .11 .06 .08 .02 3.57 1.24
Sex .04 −.04 −.08 1 −.15 .05 −.02 −.06 −.22⁎⁎ −.04 1.39 .49
Age −.04 −.14 −.13 −.23⁎⁎ 1 .09 .01 −.06 .11 −.02 21.97 2.70
Auth
Cred
− .01 − .06 − .05 .02 .21⁎ 1 .21⁎ .12 − .04 − .08 1.98 .83
.34⁎⁎⁎ .41⁎⁎⁎ .30⁎⁎⁎ −.07 −.09 .18 1 .16 .33⁎⁎⁎ .24⁎⁎ 4.45 .89
PD
VI
VC
Mean (FR)+
SD
.02 .12 .11 .04 − .13 .07 .17 1 .05 .44⁎⁎⁎ 3.67 1.14
.27⁎⁎ .29⁎⁎ .39⁎⁎⁎ − .17 − .03 − .06 .19 − .03 1 .32⁎⁎⁎ 4.77 1.10
.13 .11 .02 −.03 .06 .15 .22⁎ .22⁎ .12 1 3.71 .92
3.66 4.77 3.79 1.53 23.94 2.07 4.36 3.33 3.97 3.10
.90 .80 1.21 .50 5.37 .82 1.00 1.09 .97 .71
⁎b.05, ⁎⁎b.01, ⁎⁎⁎b.001 (two-tailed); upper triangle represents correlations among French subjects and the lower triangle represents correlations among the U.S. subjects; +is mean value for French subjects; ++is mean value for U.S. subjects.
740
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
Table 3 Multiple regression results Dependent variables
Independent variables
Combined β
1. Attitude towards ads
U.S. sample β
t
Model 1-1 Intercept Sex Nation Age Auth Cred PD VC VI Auth × Nation Auth × Cred Auth × PD Auth × VC Auth × VI
2. Attitude towards brand
– .094 − .366 .067 − .270 .296 .103 .004 .111 .132 .067 .088 .038 − .067
– .099 − .010 − .010 − .264 .348 .223 − .057 .131 .125 − .004 .114 .026 − .057
β
Model 1-2 12.099⁎⁎⁎ 1.640† − 5.830⁎⁎⁎ 1.150 − 3.255⁎⁎⁎ 4.997⁎⁎⁎ 1.703⁎ .064 1.744⁎ 1.470† 1.145 1.502† .587 − 1.059
.102 – .103 − .337 .402 .197 − .062 .035 – .129 .090 .016 − .044 Model 2-2
17.526⁎⁎⁎ 1.608† − .148 − .156 − 2.986⁎⁎ 5.486⁎⁎⁎ 3.453⁎⁎⁎ − .811 1.930⁎ 1.296† − .059 1.818⁎ .379 − .840
– .174 – .040 − .265 .359 .250 − .102 .066 – .051 .120 − .052 − .031
t
Model 1-3 5.408⁎⁎⁎ 1.222 – 1.257 − 3.488⁎⁎⁎ 4.538⁎⁎⁎ 2.094⁎ − .638 .381 – 1.439† .985 .158 − .463
–
Model 2-1 Intercept Sex Nation Age Auth Cred PD VC VI Auth × Nation Auth × Cred Auth × PD Auth × VC Auth × VI
French sample t
8.260⁎⁎⁎ 1.024 – .335 −.453 2.468⁎⁎ .396 .823 1.828⁎ – .287 1.505† .566 −1.075
– .100 – .033 − .052 .249 .039 .081 .178 – .028 .147 .060 − .114 Model 2-3
7.674⁎⁎⁎ 2.025⁎ – .482 − 2.668⁎⁎ 3.948⁎⁎⁎ 2.592⁎⁎ − 1.028 .699 – .557 1.277† − .515 − .310
15.730⁎⁎⁎ −.063 – −.880 −.293 3.551⁎⁎⁎ 2.682⁎⁎ .375 2.241⁎ – −.744 1.515† 1.310† −.930
– − .006 – − .079 − .031 .325 .238 .034 .197 – − .066 .134 .126 − .089
Notes: †p b .10 (one-tailed); ⁎p b .05 (one-tailed); ⁎⁎p b .01 (one-tailed); ⁎⁎⁎p b .001 (one-tailed). Sex: male = 0, female = 1; Nation: USA = 0, France = 1; Auth: low = −1, mid = 0, high = 1. Model 1-1, Combined sample: R2 = .35; Adjusted R2 = .31; F13, 223 = 9.17; p-value = .000. Model 1-2, U.S. sample: R2 = .28; Adjusted R2 = .21; F11, 116 = 4.03; p-value = .000. Model 1-3, French sample: R2 = .19; Adjusted R2 = .10; F11, 97 = 2.13; p-value = .025. Model 2-1, Combined sample: R2 = .26; Adjusted R2 = .22; F13, 222 = 5.94; p-value = .000. Model 2-2, U.S. sample: R2 = .24; Adjusted R2 = .17; F11, 116 = 3.28; p-value = .001. Model 2-3, French sample: R2 = .35; Adjusted R2 = .27; F11, 96 = 4.60; p-value = .000.
In sum, a reverse authority effect was manifested most strongly on attitude towards ads among the young U.S. adults when effect of CRED is controlled for as shown in Fig. 2. Thus, H2 seems to be supported concerning attitudes, but not purchase intentions. 3.2.3. Credibility As expected, perceptions of source credibility positively influenced Aad (β = .30, t = 5.00, p b .001), Ab (β = .35, t = 5.49, p b .001), and INT (β = .20, t = 2.89, p b .01) for the combined data. The positive impact of credibility on attitudes was also significant among both American and French subjects when analyzed separately. To test our expectation that the reverse authority effect is likely to be attenuated for those respondents who view the spokespersons as more (vs. less) credible, we used both regression and ANOVA analyses. The Auth ⁎ Cred interaction term positively influenced Aad among the American subjects (βAuth × Cred = .13, t = 1.44, p = .076, one-tailed), but the
interaction term was not significant in other regression models. Next, participants were divided into low vs. high credibility groups using median value and then a 3 (Auth) × 2 (Cred: low vs. high) × 2 (Nation) ANOVA on Aad was run. 5 4.5 American
4
French
3.5 3 Low Auth
Mid Auth
High Auth
Fig. 2. Impact of authority and nation on Aad: based on adjusted means controlling for CRED.
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
Results show significant main effect of authority (Mlow = 4.26, Mmid = 4.12 Mhigh = 3.82; F2, 221 = 4.22, p = .01), credibility (Mlow = 3.78, Mhigh = 4.35; F1, 221 = 21.80, p = .000), and nation (MAmerican = 4.50, MFrench = 3.63; F1, 221 = 49.47, p = .000). The Authority × Cred interaction was approaching statistical significance (F2, 221 = 1.89, p = .15). The three way interaction term was not significant, implying a similar authority × Cred interaction pattern in both countries. Similar three-way ANOVA on Ab or INT were conducted, but the results were not statistically significant. Fig. 3 depicts an Auth ⁎ Cred interaction effect on Aad, which shows a stronger reverse authority effect for the low (vs. high) CRED group. In addition, a 3 (authority) × 2 (CRED) ANOVA on Aad, Ab, or INT was conducted for each country. Results revealed a significant main effect of CRED (Mlow = 3.33, Mhigh = 4.06; F1, 104 = 9.54, p = .003) and a significant Auth ⁎ Cred interaction effect (F2, 104 = 3.58, p = .03) on INT among French subjects with clear non-crossover disordinal interaction (similar, but more distinctive than Fig. 3). Thus H3 received partial support.
741
5.5 5 Low PD
4.5
High PD
4 3.5 Low Auth
Mid Auth
High Auth
Fig. 4. The interactive influence of authority by power distance on brand attitudes (Ab) — French and U.S. combined data.
Auth ⁎ PD (F2, 235 = 2.34, p b .10). Auth ⁎ PD interaction effect was found to be more significant when the three-way ANOVA was run on Ab (F2, 234 = 3.24, p b .05), which is shown in Fig. 4. The three-way ANOVA model on INT was not significant. Thus ANOVA seem to provide provisional support for our expectations regarding a moderating effect of PD. 3.3. Discussion
3.2.4. Power distance As expected, PD positively influenced respondents' Aad (β = .10, t = 1.70, p b .05, one-tailed), Ab (β = .22, t = 3.45, p = .001), and INT (β = .15, t = 2.15, p = .03) for the combined data. The positive impact of PD was also significant among both American (p's b .05 except for INT, in which the model was not significant) and French (p's b .05 except for coefficient for Aad) subjects. To test our expectation that PD should attenuate reverse authority effect, we used both regression and ANOVA. Regression results show that PD did attenuate reverse authority effect on Aad (βAuth × PD = .09, t = 1.50, p b .10, one-tailed) and Ab (βAuth × PD = .11, t = 1.82, p b .05, one-tailed) but not on INT for the combined data set. The moderating effect of PD seems to be more consistent among French subjects (all three p's b .10, one-tailed) than American subjects (ns for Aad; p b .10, onetailed for Ab; regression model ns for INT). Thus, our expectations were met with mixed support. Next, subjects were divided into low vs. high PD groups via median split and then a 3 (Auth) × 2 (PD: low vs. high) × 2 (Nation) ANOVA on Aad was run. Results show main effects of PD (Mlow = 3.91, Mhigh = 4.21; F1, 235 = 5.97, p = .02), nation (MAmerican = 4.46, MFrench = 3.66; F1, 235 = 44.00, p = .000), and interaction effect of Auth ⁎ Nation (F2, 235 = 3.06, p b .05) and 5 4.5 Low Cred
4
High Cred
3.5
The data for the study 1 were collected about one year before the 9 / 11 calamity, an event that highlighted national security, authority of government, and compromise of personal freedom for the greater good. The calamity and aftermath may have made people more sensitive to the importance of national security and more willing to accept the infringement of personal freedom and human rights by benevolent (domestic) authority. As a result, people may have become more likely to accept intervention by the authorities: e.g., security checks in airports. Does this mean that PD may have shifted upwards across different sectors of society, resulting in a possible attenuation of reverse authority effects among young adults? A longitudinal study is needed to answer this inquiry. Study 2 was conducted almost two years after the 9 / 11 tragedy, using the same design as study 1, but with different subjects of the same age group. 4. Study 2 4.1. Subjects, design, and procedure Data were collected in 2003, three years after study 1 and two years after the 9 / 11 disaster. Participants were 169 undergraduate students enrolled in marketing classes in the U.S. Study design was similar to that used in study one: between subjects, using two ads (yogurt and software) to represent each of three (low, mid, high) authority conditions. Independent and dependent variables were identical to those in study 1. 4.2. Results
3 Low Auth
Mid Auth
High Auth
Fig. 3. Impact of authority and credibility on Aad — French and U.S. combined data.
To examine changes in responsiveness to authority appeals over time among U.S. participants, we again used regression, supplemented by ANCOVA to shed light on regression results. The same regression procedure explained in study 1 was followed.
742
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
Regression analyses revealed that reverse authority effect was not present among the post 9 / 11 American participants (p's N .40). A one-way ANCOVA of authority on Aad with Cred as a covariate also confirmed the regression results (Mlow = 4.57, Mmid = 4.51, Mhigh = 4.55; F2, 165 = .12, p N .10). In comparing American subjects across time periods, we first notice that the two groups were not statistically different in their ages (21.9 vs. 22.4 yrs), work experience (6.1 vs. 6.3 yrs), or sex composition (male/female: 61%/39% vs. 62%/38%). Further, multiple groups CFA shows evidence that PD is invariant across the two groups. Due to evidence of measurement invariance across time, standardization within each sample is omitted and the two data sets were combined prior to regression and ANCOVA analyses. To see if the reverse authority effect is reduced over time, we first ran multiple regression analyses as in study 1. We observed a significant Auth (low/mid/high) ⁎ Time (2000/2003) interaction effect on Aad (βAuth × Time = .24, t = 3.15, p b .001, one-tailed) and Ab (βAuth × Time = .21, t = 2.68, p b .01, one-tailed), supporting our speculation. Next, we ran a 3 (Auth) × 2(Time) ANCOVA on Ads with CRED as a covariate. We found a significant Auth ⁎ Time interaction (F2, 292 = 6.67, p = .001; refer to Fig. 5), which corroborates regression results. Thus, the analyses present evidence of a change in response to authority appeals across time periods in the anticipated direction. To see if PD changed over time, we first compared latent variable PD in a multiple CFA model. PD is greater in 2003 U.S. sample than in 2000 U.S. sample (p b .001). Due to the measurement invariance, we directly compared mean composite scale score; large two independent samples Z-tests indicate statistically significant changes in the American cultural values between time 1 and time 2. At least within our samples in the three year span, PD has increased from 3.67 (SD = 1.14) to 4.16 (SD = 1.15) with Z = 3.67 (p b .001, two-tailed). Thus our speculation concerning cultural change was confirmed. 4.3. Discussion We observed no authority effect among Americans at time 2. We suspect that the most significant contributor to this change would be an increase in power distance attendant to the 9 / 11 calamity. We further suspect that the 9 / 11 calamity may have elevated the perceived importance of governmental effort in
5.5 5 2000
4.5
2003
4 3.5 Low Auth Mid Auth High Auth Fig. 5. The influence of authority in two time periods on Aad: based on adjusted means controlling for CRED.
national security and authorities in charge of the job, shifting PD to higher levels among young adults. Our speculation was supported by the observed increase in PD across time periods. As people united under the causes of defending freedom and fighting terrorism, they may have become more receptive to the idea of personal sacrifice for the common good (e.g., standing in long lines for airport security checks or compromising privacy and freedom in the Patriot Act), which would have elevated their sense of respect for authority. 5. Summary This research examined authority effects experimentally in a consumer advertising context using a cross-national sample of young consumers from France and the U.S. Results of study 1 show a reverse authority effect across two cultures, with a stronger effect among American (vs. French) participants. Both perceptions of spokesperson credibility and power distance positively influenced attitudes and purchase intentions. Furthermore, credibility appears to moderate the impact of authority on attitudes and intentions. Most interestingly, power distance moderated authority effects among both nationalities. Study 2 shows that the reverse authority effect observed before 9 / 11 disappeared among post 9 / 11 Americans. In addition, compared to Americans in time 1, those in time 2 exhibited a higher power distance. 6. Theoretical and managerial implications To the best of our knowledge, this study is the first to examine the authority principle cross-culturally in an advertising context. By introducing national culture and individual level cultural values in explaining authority-based influence attempts, we shed light on how membership in a culture and a person's individual cultural values interact to form attitudes and purchase intentions. Specifically, we show that power distance at the level of an individual is important in understanding variations within a culture. In fact, individual-level power distance allowed us to predict attitudes and purchase intentions consistently within each culture: the higher (lower) power distance, the more (less) effective the authority based influence attempt. Thus, future studies that involve authority appeals should consider power distance and the interaction of PD with other possible influencers. Our research suggests that caution should be taken in predicting authority effects in specific segments of the market that might undergo drastic cultural shifts that could have positive or negative impact on the direction of authority appeals. In either case, our study suggests that prediction would be enhanced if we detect the direction of a PD value shift. For example, whereas downward shifts in power distance would predict a greater likelihood of reverse authority effects, upward shifts would predict greater likelihood of positive authority effects. Our study also presents advertisers with some provisional implications. Results imply that the classic authority principle is not likely to be effective in all the segments of the market.
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
Marketers need to know whether or not authority sources will have a positive impact on their target segments. This caution will be especially relevant for marketers who target young or adolescent consumers who view low authority figures more favorably. Additionally, our results enlighten marketers contemplating authority-based ads in countries with different perceptions of power distance. Marketers in multi-national companies should not blindly rely on country level power distance scores; instead, they should examine the effectiveness of authority country by country, segment by segment, and adopt a different level of authority as appropriate.
743
our findings using different authority figures that are less vulnerable to the nature of relationships. The present research documents cross-cultural variations in reverse authority effects among young adults in two cultures. Further investigation is required to explain the underlying source of such effects, as well as the circumstances under which one can expect positive versus negative effects of authoritybased appeals. We offer the present findings in hope of fueling wider interest in this topic and commend on-going investigation to future research. Appendix A. Test ads
7. Limitations and future research directions Despite insights gained from this study, several unanswered questions remain. Limits to generality stem from the demographic composition of our sample. We used university students, whose relative youthfulness may be a contributing factor to the reverse authority effect. Future study can include subjects from diverse age, income, education, and occupational groups. According to Hofstede (2001), education and occupation are negatively correlated with power distance. Thus, their impacts on authority appeals should be examined. Second, we argued that the 9 / 11 calamity might have caused a shift in cultural values, which might have caused a more positive orientation to authority and authority-based influence attempts. We propose that the observed differences represent a zeitgeist effect (all members' shift in value due to radical system-wide change in society). However, we acknowledge that our observations are not immune to the influence of unknown, extraneous variable(s). Third, the observed effects may stem from the specific nature of the relationships among speakers represented in the simulated ads. That is, subjects' reactions may have stemmed from specific attitudes towards parents and teachers (the nature of relationships), rather than to “authority figures” in general. Future study might attempt to corroborate
Ad #1: Imagine a radio advertisement in which (a mother/one woman/a daughter) is speaking to (her daughter/another woman/her mother). She tells her about “Soiegourt,” a new brand of yogurt. She says that it comes in a variety of interesting flavors, that it has fewer calories than Yoplait, but that it tastes richer. It sells for about the same price. At the conclusion of the ad, of course, (the mother/she/the daughter) recommends that (her daughter/the other woman/her mother) try the new brand. “I tried it last week and it was great!” Ad #2: Imagine a radio advertisement in which (a teacher/one person/a student) is speaking to (a student/another person/a teacher) about a new software product called “MicroPass-2000.” This product is designed to manage multiple passwords (mots de passes). It works in conjunction with other software products such that a user need remember only one password for all applications. It is easy to use and the cost is surprisingly low. At the conclusion of the ad, of course, (the teacher/the speaker/the student) recommends that (the student/the other person/the teacher) consider buying this software — “I installed MicroPass on my home PC and it works great!” Ad #3: Finally, please imagine a radio advertisement in which (a boss/one employee/a subordinate) is speaking to (his subordinate/another employee/his boss). He tells him about an internet site at which clothing can be purchased: www.smartdress.com. The site features name-brand clothing at discount prices, delivered to your home or office at no extra charge. All items are guaranteed. At the conclusion of the ad, of course, (the boss/he/the employee) recommends that (his subordinate/the other employee/his boss) visit the site — “I did and I think you'd love it.” Inside parenthesis indicate the power distance, either “high” “mid,” or “low” authority conditions, respectively.
Appendix B Power distance scale: scale items retained, factor loadings, CFA model fit statistics, and reliabilities (N = 248)
1. I find it difficult to disagree with someone in a higher position than mine. + 2. It is difficult for me to express my opinions to superiors. + 3. If a boss doesn't ask for my comments, I would rather keep silent. + 4. I tend to conform to the wishes of someone in a higher position than mine. 5. It is difficult for me to refuse a request if my superior asks me. 6. I tend to give priority to the opinions of people in authority. 9. I tend to find it hard to disagree with authority figures. % Variance explained Cronbach α
Combined sample (N = 248)
U.S. sample (N = 130)
F1
Lx
F1
Lx
F1
Lx
.71
.64
.75
.68
.67
.58
.77
.70
.80
.73
.73
.64
.71
.66
.72
.68
.67
.62
.85
.81
.88
.88
.83
.83
.72 .74
.67 .69
.72 .74
.68 .68
.70 .73
.65 .70
.68
.60
.79
.75
.62
.53
55.02% .86
59.80% .89
French sample (N = 118)
50.54% .83 (continued on next page)
744
J.M. Jung, J.J. Kellaris / Journal of Business Research 59 (2006) 735–744
Appendix B (continued) Data set
χ2
df
p
χ2 / df
RMSEA
GFI
AGFI
RMR
CFI
IFI
|SR| N 2
Combined sample American sample French sample
66.35
14
.000
4.74
.12
.93
.86
.13
.92
.93
None
45.46
14
.000
3.25
.13
.91
.82
.12
.93
.93
None
34.24
14
.002
2.45
.11
.92
.85
.16
.93
.93
None
Three scale items denoted as + are adopted from pretest 6 of Jung (2002). F1 = Factor loadings from principal component analysis of seven-point Likert-type scale items. Lx = Standardized factor loadings from CFA. |SR| stands for absolute value of standardized residual.
References Agrawal J, Kamakura WA. The economic worth of celebrity endorsers: an event study analysis. J Mark 1995;59(3):56–62. Aiken LS, West SG. Multiple regression: testing and interpreting interactions. Newbury Park: Sage Publications; 1991. Bickman L. The social power of a uniform. J Appl Soc Psychol 1974;4:47–61. Brislin R. Translation and content analysis of oral and written materials. In: Triandis HH, Berry J, editors. Handbook of cross-cultural psychology: methodology, vol. 2. Boston: Allyn and Bacon; 1980. p. 389–444. Bushman BJ. The effects of apparel on compliance: a field experiment with a female authority figure. Pers Soc Psychol Bull 1988;14:459–67. Cialdini RB. Influence: science and practice. Boston: Allyn and Bacon; 2001. Cialdini RB, Wosinska W, Barrett DW, Butner J, Górnik-Durose M. Compliance with a request in two cultures: the differential influence of social proof and commitment/consistency on collectivists and individualists. Pers Soc Psychol Bull 1999;25(10):1242–53. Cialdini RB, Rhoads K. Human behavior and the Marketplace. Marketing Res 2001;13(3):8–13. Dholakia RR, Sternthal B. Highly credible sources: persuasive facilitators or persuasive liabilities? J Consum Res March 1977;3:223–32. Goldsmith RE, Lafferty BA, Newell SJ. The impact of corporate credibility and celebrity credibility on consumer reaction to advertisements and brands. J Advert 2000;29(3):43–54. Green SB, Salkind NJ. Using SPSS for windows and macintosh: analyzing and understanding data. Upper Saddle River: Prentice Hall; 2003. Grewal D, Gotlieb J, Marmorstein H. The moderating effects of message framing and source credibility on the price-perceived risk relationship. J Consum Res 1994;21:145–53. Harmon RR, Coney KA. The persuasive effects of source credibility in buy and lease situations. J Mark Res 1982;19:255–60. Heath TB, MacCarthy MS, Mothersbaugh DL. Spokesperson fame and vividness effects in the context of issue-relevant thinking: the moderating role of competitive setting. J Consum Res March 1994;20:520–34. Hofstede GH. Cultures and organizations: software of the mind. New York: McGraw-Hill; 1991. Hofstede GH. Culture's consequences: comparing values, behaviors, institutions, and organizations across nations. Thousand Oaks: Sage; 2001. Jung, JM. The interactive impact of culture and individual characteristics on ethical decision-making processes, criteria, and judgmental outcomes: a
cross-national comparison between south Korea and United States. PhD dissertation. University of Cincinnati; 2002. Jung JM, Kellaris JJ. Cross-national differences in proneness to scarcity effects: the moderating roles of familiarity, uncertainty avoidance, and need for cognitive closure. Psychol Mark 2004;21(9):741–55. Kamins MA, Brand MJ, Hoeke SA, Moe JC. Two-sided versus one-sided celebrity endorsements: the impact on advertising effectiveness and credibility. J Advert 1989;18(2):4–10. MacCracken Grant. Who is the celebrity endorser? Cultural foundations of the endorsement process. J Consum Res 1989;16(3):310–21. Madden TJ, Allen CT, Twible JL. Attitude toward the ad: an assessment of diverse measurement indices under different processing ‘sets’. J Mark Res 1988;25:242–52. Manrai LA, Manrai AK. Global perspectives in cross-cultural and cross-national consumer research. New York: International Business Press; 1996. Michener HA, Burt MR. Components of “authority” as determinants of compliance. J Pers Soc Psychol 1975;31:606–14. Milgram S. Behavioral study of obedience. J Abnorm Soc Psychol 1963;67:371–8. Mizerski RW, Golden LL, Kernan JB. The attributional process in consumer decision making. J Consum Res 1979;6:123–40. Mooij MD. Global marketing and advertising: understanding cultural paradoxes. Thousand Oaks: Sage Publications; 1998. Neter J, Wasserman W, Kutner MH. Applied linear statistical models: regression, analysis of variance, and experimental designs. Homewood: Richard D. Irwin; 1985. Nunnally JC, Bernstein IH. Psychometric theory. New York, NY: McGraw-Hill; 1994. Rasinski KA, Tyler TR, Fridkin K. Exploring the function of legitimacy: mediating effects of personal and institutional legitimacy on leadership endorsement and system support. J Pers Soc Psychol 1985;49:386–94. Singelis TM, Triandis HC, Bhawuk D, Gelfand MJ. Horizontal and vertical dimensions of individualism and collectivism: a theoretical and measurement refinement. Cross-Cultural Research 1995;29(3):240–75. Triandis HC. Individualism and collectivism. Boulder, CO: Westview Press; 1995. Tyler TR, Lind EA, Huo YJ. Cultural values and authority relations: the psychology of conflict resolution across cultures. Psychol Public Policy Law 2000;6(4):1138–63.
Journal of Business Research 59 (2006) 745 – 754
Cross-advertisement affectivity: The influence of similarity between commercials and processing modes of consumers on advertising processing Ingrid Poncin a,⁎, Rik Pieters b , Michele Ambaye c a
Lille Graduate School of Management, Département Marketing-Vente, Avenue W. Brandt, 59777 EURALILLE, France b Universiteit van Tilburg, The Netherlands c Brunel University, United Kingdom Received 1 April 2005; received in revised form 1 November 2005; accepted 1 January 2006
Abstract Due to the increasing brand and product parity and the soaring volume of advertising, not only may products be increasingly similar, but commercials for products may become similar as well. Moreover, commercials frequently appear closely together in pods within and between programs, increasing the likelihood that they influence each other. The question thus becomes if and how the various commercials in pods and their affectivity “cross” influence each other, and what the influence of the similarity between commercials is on advertising processing and effectiveness. Extending assimilation–contrast theory, we find substantial context effects of the ad sequence in the affective reactions elicited by commercials presented in the same pod. The importance of the similarity between commercials as perceived by consumers, and of the interaction between this similarity and the consumers' processing modes during ad exposure are demonstrated. The findings contribute to a better understanding of context effects due to the sequence in which ads appear, and of the role of ad similarity. © 2006 Elsevier Inc. All rights reserved. Keywords: Cross-advertisement effects; Perceived similarity; Affective reaction; Sequence effects; Processing mode
1. Introduction Due to heavy competition in many mature markets, products and brands are increasingly at par, sharing many attributes and features, which may prompt price reductions and other promotional tools pressuring profit margins and the long-term marketing viability. Emotional advertising is frequently used to differentiate such similar products and brands (Rossiter and Percy, 1997), to stop and change the downward profit contribution spiral. However, the frequent use of emotional advertising may in fact have led to the situation that not only products and brands, but also their communication tools become at par, the latter at least with respect to their affectivity. This situation is even aggravated because television commercials, the dominant communication tool in most consumer ⁎ Corresponding author. Tel.: +33 320 215 996; fax: +33 320 215 959. E-mail address: [email protected] (I. Poncin). 0148-2963/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2006.01.012
markets, appear in pods or sets within and between programs, prompting immediate comparisons of commercials and the brands and products that appear in them. This makes it important to understand how ad-evoked feelings influence advertising responses and if and how the affectivity of commercials affects each others processing responses. Quite surprisingly however advertising research appears to have emphasized the influence of television commercials in isolation (Brown et al., 1998). Because the influence of the (other) surrounding commercials in a given sequence is not taken into account, it is not obvious if and how the results of such studies can be generalized to the common situation that ads appear in blocks, and thus when the affectivity of earlier ads might influence the processing and effectiveness of later ones. It is not unlikely that earlier ads might create an affective climate for the later ads in the same sequence. This climate may enhance or reduce the persuasive effects of those advertisements (Brooker, 1981). The advertising surrounding any particular ad
746
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
is, in fact, a key context for that particular ad. Additionally, according to Kent (1993) exposure to competitive clutter appears to have a more damaging impact on ad effectiveness than exposure to ads for non-competing brands. In fact, similar products and similar advertising appeals may generate brand confusion, which in turn can lead to an increase in sales of a competitor's instead of the advertiser's brand. It thus becomes important to avoid brand confusion by examining its possible determinants (De Pelsmacker et al., 2001). Therefore, some network policies provide assurance that ads for direct competitors will not appear in the same pod (sequence) of commercials. Whether such policies are relevant is not clear due to the lack of empirical research on cross-advertisement effects and the role of ad similarity. Our research focuses on the impact of the affectivity of preceding commercials on affective reactions elicited by subsequent neutral (non-emotional) commercials in pods. These latter commercials may be low or high in perceived similarity with the first ad, and we examine the influence of such similarity on ad processing and effectiveness, as measured by attitude toward the ad (Aad). In the conceptual analysis, we will argue that the influence of commercial similarity depends on the specific processing modes that consumers adopt during ad exposure. If we were indeed to find cross-advertisement effects, this would have implications for pretesting practice, among others. Frequently, copy testing is carried out either with a single commercial or in a “random” sequence of commercials. If the preceding ads create a context which influences the affective reactions induced by the subsequent ads, the validity of the pretest results and their diagnostic value are at stake. In fact, we have no guarantee that the affective reactions produced by pretest contexts are equivalent to the reactions caused by the final context. Thus copy testing in isolation might not necessarily be optimal. One solution then, on which we elaborate later, might be to test ads in the context of other ads with a known affective performance. Further, it would then also be useful to test them in different contexts to examine their malleability under different affective ad environments. But let us first examine why and how cross-advertisement effects of affectivity might arise, and how we explored their effects. 2. Cross-advertisement effects of affectivity Little is known about cross-advertisement effects of affectivity, but we do know more about the related issue of program context effects, and we draw on this literature first (Aylesworth and MacKenzie, 1998; Coulter, 1998; Goldberg and Gorn, 1987; Kamins et al., 1991; Pavelchak et al., 1988; Schumann, 1986; Shapiro et al., 2002; Yi, 1990; Tavassoli et al., 1995; De Pelsmacker et al., 2002). With respect to programelicited feelings (Goldberg and Gorn, 1987; Pavelchak et al., 1988; Coulter, 1998), two important dimensions of the affective reactions induced by a program have to be examined: the intensity of the reactions induced (sometimes termed “arousal”, Singh and Hitchon, 1989) and the valence/quality/polarity of the affective reactions (i.e., the pleasantness or pleasure
dimension). In line with previous research, and in view of the dominance of positive emotions in advertising, we focus on positively valenced emotional advertising, and examine the influence on advertising processing of variations in affective intensity, from absent, in case of an affectively neutral commercial to intense, in case of an intensely positive emotional commercial. Does positive affect, induced by the program context (or earlier commercials), improve subsequent ad processing? Both consistency and congruency theory (Pavelchak et al., 1988; Singh and Hitchon, 1989) would predict that experiencing positive feelings due to a program enhances recall, positive affective reactions triggered by the ad, ad liking and subsequent attitudes. In support of this, Aylesworth and MacKenzie (1998) observed that television ad processing was better when people were in a positive mood after seeing a program. Likewise, Goldberg and Gorn (1987) found that happy compared to sad TV programs increased the effectiveness of embedded commercials, for instance, because they were recalled better and produced higher purchase intent. Additionally, Coulter (1998) has examined the effects of emotional responses to television programming (program-induced affective reactions) on the attitude toward the ad (Aad). He found that emotional responses to the ad moderate the relationship between program liking and Aad. Moreover, the program liking–Aad linkage is strengthened if the ad and program are similar in emotional content. These findings show that program-induced positive affect has a systematic assimilative or transfer effect on various attitudes and behaviors with respect to commercials embedded in the program. In contrast, two opposite predictions emerge about the possible influence of affective intensity of program context on advertising processing, which may generalize to cross-advertisement affectivity effects. First, the negative effect hypothesis postulates that high involvement, or arousal in a program, reduces the ability of viewers to process and recall advertisements that appear in the program context. In a study that compared consumers' reactions to ads during the Super Bowl, Pavelchak et al. (1988) found that when the program is too arousing, it has a negative effect on ad and brand recall. In fact, ad responses in the winning city were inhibited in contrast to those in the losing and neutral cities. An opposite prediction can also find support: the positive effect hypothesis. According to this idea, commercials are more effective when viewed in highinvolvement programs, because the high program involvement and the associated arousal transfers to the advertisements, in line with excitation transfer theory (Zillmann, 1971). In fact, the arousal produced by a program might even be misattributed to the embedded commercial as long as the viewer is not aware of the source of the arousal. Thus, according to this positive effect perspective, “commercials embedded in exciting program segments can be expected to demonstrate increased learning (recall and recognition) relative to commercials embedded in bland program segments” (Singh and Hitchon, 1989, p.24). To reconcile these two opposing perspective on program context effects on commercial effectiveness, Tavassoli et al. (1995, p.64) propose an inverted-U relationship arguing that “it is possible that some studies manipulated involvement/arousal on the ascending part of the inverted-U-curve (i.e., comparing
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
low to moderate levels of program involvement) while others manipulated the descending part (i.e. comparing moderate to high levels)”. There is some speculation and empirical investigation about the interactive effects of polarity and intensity of affective reactions. These speculations have concentrated on the emergence of assimilation versus contrast effects, which we believe to be important for the present research. For instance, Broach et al. (1995) predicted a two-way interaction between program-induced arousal and pleasantness. According to those authors, a high-arousal program should produce an assimilation effect, whereby viewers shift their evaluations of subsequent commercials (i.e., the target) in the direction of their evaluation of the program (i.e. the stimulus). Thus, if the program were pleasant, the commercials would be viewed as pleasant. If however the program were unpleasant, the commercials would be viewed as unpleasant. On the other hand, a low-arousal program should produce a contrast effect whereby viewers shift their evaluations of subsequent commercials in the direction opposed to their evaluation of the program. Thus if the program were pleasant, the commercials would be viewed as unpleasant; if the program were unpleasant, the commercials would be viewed as pleasant. To prevent any interaction effects between program and affective reactions induced by the ad-itself, they used emotionally neutral commercials to test their hypothesis. The mean commercial pleasantness value for the pod of commercials in the high arousal pleasant program condition was significantly higher than that in the high arousal unpleasant program condition. Furthermore, the mean commercial pleasantness value in the low arousal unpleasant program condition was significantly higher than that in the low arousal pleasant program condition. Thus, the hypothesis was supported since high arousal programs produced an assimilation effect where commercial evaluations were enhanced after pleasant programs and depressed after unpleasant programs. Conversely, low arousal programs produced a contrast effect whereby commercial evaluations were enhanced after unpleasant programs and depressed after pleasant programs. While quite some research has examined program context effects, little is known about ad context effects in advertising (see Broach et al., 1997; Pieters and Bijmolt, 1997; Zhao, 1997). Aaker et al. (1986) suggested and found initial empirical evidence that affective reactions to eliciting stimuli are influenced by the sequence in which they are seen. We build on this work here. Specifically, we base our predictions about cross-advertisement affective effects on the assimilation– contrast theory (Sherif and Hovland, 1961). In the following sections, we present recent developments in this theory that may lead to a better understanding of sequence effects, and we propose specific predictions to be tested in the empirical part.
747
psychological scale for judgment. Henceforth, judgment of related stimulus is made in terms of the categories of this reference scale”. Therefore, social judgment views an attitude as consisting of a continuum of evaluations: a range of acceptable positions, unacceptable positions and positions toward which the individual has no strong commitment. In fact, assimilation and contrast are judgmental distortions, i.e., perceptual biases resulting from the tendency to perceive phenomena differently as a function of the reference scale (i.e. the context in which they are judged). Assimilation is the displacement of the judgment toward the anchor (reference scale) (i.e. the judgment that the stimulus is more similar to the anchor than it actually is). In contrast, there is the displacement of the judgment away from the anchor (the perception that the difference between the stimulus and the anchor is greater than it actually is). Developed in a cognitive judgment context, this theory has been recently applied in a more affective context (McMullen, 1997; Abeele and Gendolla, 1999). According to McMullen (1997), an affective contrast effect occurs when an individual's affect is displaced away from the valence/polarity of the reference stimulus. In contrast, an assimilation effect happens when the individual's affect is displaced toward the valence/ polarity of the reference stimulus. In line with this, Abeele and Gendolla (1999) define assimilation effects in terms of mood congruency, such that people in a positive mood make more positive judgments than people in a neutral mood, who in turn make more positive judgments than people in a negative mood. Consequently, assimilation effect led people in a more positive mood to make more positive judgments. In an important study, Stapel and Winkielman (1998, p.634) clarified “Our feelings and evaluations are experienced contextually and thus determined by their relationships to other affects and judgments. The context in which a target stimulus is embedded provides a frame of reference for interpretation and judgment. Hence, the same target can be associated with different responses depending on the context in which it is judged.” Likewise, the same commercial may be associated with different responses depending on the context in which it is judged. In our research, the context is created by the other commercials in the pod. These commercials might influence the affective reaction induced by a specific commercial. We speculate that strong positive affective reactions elicited by a previous commercial in the pod are likely to fade into a positive mood that, in turn, enhances the probability of elicitation of positive affective reactions (Davidson, 1994) and the subsequent evaluation. In the case of neutral or ambiguous commercials for the affective point of view following a strong emotional ad, the assimilation effect is likely to be highly relevant. 2.2. Perceived similarity and assimilation contrast effects
2.1. Assimilation and contrast in cross-advertisement affectivity effects According to Sherif and Hovland (1961, p.179) “an individual confronted with a series of stimuli tends to form a
In a non-advertisement context, Abeele and Gendolla (1999) showed that the use of a reference stimulus such as a standard of comparison produces a contrast effect. Yet, assimilation effects may occur when the reference stimulus induces a specific mood,
748
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
which serves as reference frame for subsequent judgment. In advertising, a specific commercial might induce a positive mood, inducing more positive affective reactions and judgments for subsequent commercials: assimilation. They found that this assimilation effect is stronger in the case of a low thematic similarity between the mood-inducing event (in our case the first commercial) and the judgmental issue (the second commercial (non-emotional). Conversely, in case of high perceived similarity (for instance when the two commercials are from the same product category), the first event (in our case the first commercial) is used as standard of reference to evaluate the second commercial to carry out a contrast effect. In fact, Abeele and Gendolla (1999) demonstrated that assimilation and contrast effect are not mutually exclusive, but rather they are concurrent processes, and thus what is observed is the net effect of the two processes (assimilation effect–contrast effect). Building on this, we predict that in case of low thematic similarity between two commercials in a sequence, a stronger assimilation effect occurs. This stronger assimilation may be between the affective reactions elicited by the first commercial and the affective reactions elicited by the second commercial. In fact, in the case of high similarity between commercials, the possibility to compare the two commercials should favour a contrast effect to occur. This leads to our first hypothesis, namely that if a neutral ad is preceded by a positive emotional ad, but has only low perceived similarity, the second ad will induce more positive affective reactive reactions and will appear more positive in terms of Aad, compared to those with high perceived similarity between the two ads. There is reason to believe that the occurrence of crossadvertisement affectivity effects depends on the processing mode that consumers are into while being exposed to the commercials, and to this topic we turn next. 2.3. Processing modes of consumers Sometimes, consumers during ad exposure focus on the experiential, hedonic qualities of the advertisements themselves, sometimes they are prompted to compare the ads on their merits, and sometimes they desire to memorize the ads, among others. These are examples of processing modes or processing goals, i.e., the specific tasks that consumers are engaged in during exposure to advertising. In 1999, Meyers-Levy and Malaviya proposed a framework describing three fundamental processing strategies that are relevant here. “When the level of mental resources a message recipient allocates to message processing is substantial, moderate or minimal, judgment formation proceeds through a systematic, heuristic or experiential strategy respectively.” In fact, in the majority of processing conditions, the amount of cognitive resources that people are willing or able to devote to processing is so meager, that only the most fleeting and scant message processing occurs. In such cases, the resources people allocate to processing are therefore minimal and they may be unable to attend to any substantive or even superficial content of the message, as they would if a heuristic strategy were employed. Rather, people are
more likely to pay attention to sensations or feelings that might be generated from the process of dealing with the advertisement or from the advertisement itself (Meyers-Levy and Malaviya, 1999). On the one hand, the experiential mode or more affective mode might be considered as the case of the “peripheral processing” of the ad. On the other hand, in the case of evaluative mode, more important resources are allocated and the processing of the ad will be systematic. In this latter case, the consumer focuses their attention on important and diagnostic aspects of the message and, therefore, the ad's evaluation is more cognitive. McMullen (1997) found support that assimilation is more likely to occur when consumers are in an experiential mode because they are focusing on their immediate affective reactions to the mental simulation. 2.4. Ad processing mode and assimilation contrast theory In the case of the evaluative processing mode, more cognitive processing of the ad is expected, which is likely to reduce the affective reactions elicited and might result in a more systematic comparison between the commercials. Consequently, the context – the preceding ads – is more likely to be used as a comparison standard and therefore a contrast effect between the ads. Meyers-Levy and Malaviya (1999) mention the phenomenon of judgment process correction, which might be related to the contrast effect. In some circumstances people will attempt to correct their initial assessment for biases that they perceive. Three conditions need to be met: First the message recipient must become aware that the contextual data may have influenced his/her views about the persuasive message inappropriately. This occurs because salient features of a judgment task or communication context bring the influence to mind. Secondly, the person must be able to identify a naïve theory that might account for why, how and to what extent the biasing data could have this effect (Petty and Wegener, 1993). Thirdly, the message recipient must possess and be willing to expand the cognitive resources required to engage in a correction process (Martin et al., 1990)). In most cases however, the influence of the biasing source on judgments is believed to be assimilation (Martin and Achee, 1992). If any of the required conditions are not met, the presumably spontaneously occurring assimilation effect of context on people's initial judgments remains intact. Moreover if such conditions are met, people will attempt to screen out the inappropriate influence on their initial assessment. Yet, because the process of correcting and screening out bias is imperfect and presumably prone to over-adjustment, a contrast effect may occur. If this is the case, people's updated judgments become fairly extreme and appear to be the opposite direction of the interpretation implied by the contextual data (Martin and Achee, 1992). Therefore, if the consumer thinks that the context biases her evaluation or alters the evaluation to be more positive, she will try to correct this bias by updating her evaluation as more negative. When the processing mode is experiential, affect will dominate. Consequently, the affective context created by the
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
preceding commercials is more likely to be used as frame of reference and is likely to influence the quality of the affective reactions perceived as congruent. Therefore, some assimilation effects are expected between the affective reactions elicited by commercials aired in the same pod. These assumptions are in line with the predictions made by McMullen (1997). As was previously mentioned, the essential determinant of the processing mode is the amount of cognitive resources that the consumer is able and motivated to allocate. Therefore, in the everyday advertising context, the consumer is expected to be in an experiential mode most of the time. However, in some specific situations (i.e. situational involvement in a product category, pre-test situation…), the consumer might be in an evaluative mode. For instance, in television copy testing consumers may frequently be in an evaluative mode or systematic processing mode stimulated by the task, the environment, and the fact that they are paid to “judge” commercials. De Pelsmacker et al. (2001, p.229) point out that “most pre-tests take place in an experimental setting. Consumers may behave fundamentally differently when exposed to an ad in a real-life situation. Hence, some pre-test methods, especially the explicit ones such as direct rating method, are susceptible to consumer jury effects.” Therefore, the difference between these two contexts (everyday life versus copy testing) may lead to systematic differences in the affective reactions of consumers to commercials. If this occurs, the validity of pre-testing practices may be in serious doubt. Moreover, it might have implications when ads for directly competing products are included in the same pods. In short, in some contexts, the affective reactions to the commercial might be assimilated or contrasted with the reactions to other commercials aired in the same pod. We anticipate that the use of the experiential mode for processing will lead to more assimilation between the affective reactions elicited by the two commercials in comparison with the use of more systematic processing (evaluative mode). This leads to the second hypothesis to be tested: When the processing mode is experiential, a neutral ad preceded by a positive emotional ad will induce more positive affective reactions and will appear more positive in terms of Aad, compared to an ad which is considered by an evaluative mode. The strongest assimilation effect and the weakest contrast effect are expected, in the case of an experiential mode of ad-processing and low affective/content perceived similarity. The strongest contrast effect is expected in the case of an evaluative mode and high-perceived similarity. At this stage, two main effects are predicted: one for perceived similarity and one for processing mode. Although it is more speculative, an interaction effect between mode and perceived similarity is likely, such that the effect of perceived similarity might be influenced by the processing mode. When perceived similarity is high, the evaluative mode might lead to a stronger contrast effect. When perceived similarity is low, the mode of processing might be less effective. Therefore, we propose that the positive affective reactions induced by a neutral commercial will be more intense (higher) in case of assimilation effect with the preceding emotional commercial. We predict that the attitude toward the neutral ad (Aad) will be more positive in
749
comparison with the Aad score for the same commercial when aired alone or in a condition where a contrast effect between the two ads is expected. This leads to the third hypothesis to be tested, namely that in the case of the experiential treatment mode in an ad, if a neutral ad is preceded by a positive emotional ad which shares low perceived similarity, the second ad will induce more positive affective reactions and appear more positive in terms of Aad. This is opposed to the case where commercials might be highly similar and are processed in an evaluative mode. Support for the three hypotheses would demonstrate the importance of cross-advertisement affectivity effects, and would show how the affect evoked by earlier ads may increase or decrease the effectiveness of subsequent ads, under specific but very common situations, namely depending on the similarity of the ads and the processing modes of the consumers. In the following sections, the design and the results of a controlled experiment to test these predictions are described. 3. Methodology Laboratory conditions seemed especially suited, due to the need to examine the causal relationship between the ad processing mode, perceived similarity and the changes observed in the affective reactions induced by the commercials. Moreover, a high level of control was required to correctly assess the respective importance of each variable and limit the importance of extraneous variables. 3.1. Pilot study results A pilot study was carried out with 20 students who rated the degree to which they perceived 6 different commercials as being similar or dissimilar to each other We asked participants to globally judge the perceived similarity between the two commercials (very dissimilar versus very similar). Then we asked them to identify which aspects they concentrated on to make this judgment. Finally we asked them to judge the perceived similarity of several specific characteristics such as product category, affective reactions elicited, ad-type…. During this pilot study, it was noted that consumers used different dimensions to judge the similarity between two ads. Belonging to the same product category is a necessary, but insufficient, condition. The duration, the ad-thematic characteristic, as well as other characteristics, might help to make this judgment. Based on this first result, two pairs being perceived as the most similar, and two pairs perceived as the most dissimilar, were selected. Each of the chosen commercials was also pilot-tested with 20 other participants to evaluate the commercial's ability to elicit affective reactions when aired alone. This method of ad-testing ensures a low variance between the affective reactions elicited by the same ad for the different consumers (homogeneity response of the consumer responses). It also allows the possibility of checking the high variance in the intensity of the affective reactions induced by commercials — described as being emotional versus neutral. Finally, several pilot studies were conducted with other consumers in order to confirm the
750
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
understanding of the instructions used to manipulate the way of processing commercials and to check the understanding of the final questionnaire (see Appendix for the instruction used to manipulate either the experiential mode or evaluative mode). 3.2. Design of the experiment The design used in this study is a “between-consumers” design: a 2 × 2 (experiential mode versus evaluative mode and low versus high perceived similarity), replicated twice. Each block of commercials was designed to present a neutral commercial followed by a strong positive emotional commercial. This, in turn, was followed by either a neutral testcommercial, pilot-tested as being perceived similar to the emotional commercial, or a neutral commercial pilot-tested as being perceived as dissimilar to the preceding emotional commercial. The use of repeated measures in the design, with two sequences of different commercials, allows the possibility of ruling out idiosyncratic effects resulting from the specific characteristics of the ads. Each participant was randomly assigned to one experimental condition and individually exposed to two strings of three commercials. 3.3. Stimuli The stimuli were six 30 s TV commercials. Originally aired in Switzerland, they were unknown in Belgium and advertised unknown brands, therefore removing the concern of a differential familiarity with the stimuli. As in the pilot study, the ads that were judged as being similar were ads that belonged to the same product category. 3.4. Measures and sample In line with previous research in emotion and advertising (Brown et al., 1998), emotional responses were collected using verbal measurements composed of an adjective list of affective reactions. Attitude toward the ad was measured with a simple two-item scale as used by Derbaix (1995). Measures of manipulation effectiveness were added as well (processing mode and perceived similarity). A sample of 60 undergraduate students across academic disciplines at a Belgian university participated in the study. Students are huge TV-media consumers and therefore an ideal target. In addition, the product categories chosen for the different commercials (a travel agency and a soft drinks company) were well-known to them. Participants were randomly assigned to the four cells of the two (high-low similarity) by two (processing mode: evaluative vs. experiential) between-subjects design. 4. Results 4.1. Manipulation checks First of all, checks were performed on the manipulated variables. We examined the perceived similarity between the
commercials and checked the processing mode (used by the participants in the different conditions). On the one hand, two groups (Group 1 and Group 2) needed to be in a condition where perceived similarity would be expected between the emotional commercial and the neutral test commercial. Conversely, perceived dissimilarity was expected in the two other groups. To check if the ads used were perceived as expected, we used a similarity measure. The test (F(1, 55) = 4.62, p b 0.05 (one tailed)) showed that commercial X1 was perceived as more similar to commercial Y1 than to commercial Y2. Similarly, the test (F(1, 55) = 5,65, p b 0.05 (one tailed)) confirmed that commercial X2 was perceived as more similar to commercial Y2 than to commercial Y1, as expected. Therefore we concluded that the check of the perceived similarity supports the initial expectations. On the other hand, the question that tapped the processing mode of the participants during exposure to the commercials was related to the attention to information versus emotion during the ads. One can expect that participants in an experiential mode pay more attention to the emotions compared to participants in an evaluative mode, who are expected to pay more attention to information. The observed means in the different conditions were in the expected direction and the main effect of the processing mode was statistically significant F (1, 55) = 5.52; p b 0.05 (one tailed). Thus, the similarity and processing mode manipulations were effective. 4.2. Effects on affective reactions In order to test our prediction, the main effects of similarity and mode and their interaction on both positive and negative emotions need to be examined. If the theory is supported, we would find significant main effects and an interaction between processing mode and perceived similarity, but only for positive emotions. As expected, for the negative affective reactions, no main effects were observed for either similarity or processing mode. Positive affective reactions elicited in the case of dissimilarity were more intense than in case of similarity. Perceived similarity therefore has an impact in the expected direction. The positive emotions elicited in the evaluative condition appear to be more important in comparison with the experiential condition, which is the reverse of what was expected. However, the dependent measures show only a significant main effect for perceived similarity F(1, 53) = 14.06; p b 0.001 (one tailed). But neither the main effect of a mode, nor the interaction effect of similarity by processing mode is statistically significant. Only the first hypothesis is validated. 4.3. Effects on the attitude toward the ad (Aad) Attitude toward the ad (Aad) was tapped with a two-item scale following a check of internal consistency (α = 0.74 and 0.81), and a single score for Aad was computed. First, we detected heterogeneity between the two replicates (F(1, 53) = 4.28; p b 0.05 (one tailed)). This result might be due to specific characteristics of one or both neutral test
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
751
Table 1 Observed means and MANOVA results
Sequence 1 Similarity Dissimilarity
Sequence 2 Similarity Dissimilarity
Positive affective reactions
Attitude toward the ad (Aad)
Min = 7, Max = 49, a higher value meant that the affective reactions were more positive. One tailed tests were used.
Min = 2, Max = 10, a higher value indicates more positive affective reactions. One tailed tests were used.
Experiential Evaluative mode mode 14.57 14.93 14.75 Similarity b dissimilarity 19.21 22.79 21.00 16.89 18.72 Evaluative N experiential
Experiential Evaluative mode mode 5.43 5.20 5.14 6.29 5.29 5.74 Evaluative N experiential
5.31 Similarity b dissimilarity 5.71
Experimental Evaluative mode mode 18.00 15.80 16.86 Similarity b dissimilarity 21.57 24.00 22.78 19.78 19.75 Evaluative ≈ experiential
Experimental Evaluative mode mode 6.14 5.33 6.36 7.00 6.25 6.13 Experiential ≈ evaluative
5.72 Similarity b dissimilarity 6.68
Main mode effect Main similarity effect F(1, 53) = 0.41; F(1, 53) = 14.06; p = 0.261 p = 0.00 Interaction effect mode ⁎ similarity F(1, 53) = 1.47; p = 0.115
Main mode effect Main similarity effect F(1, 53) = 0.20; F(1, 53) = 2.54; p = 0.329 p = 0.058 Interaction effect mode ⁎ similarity F(1, 53) = 2.82; p = 0.05
commercials. However, while rather small, the main effects of perceived similarity are in the expected direction for both sequences. Conversely, the observed means for the impact of the processing mode are nearly equal in the two conditions (no effect of the processing mode) during the replication. Consequently, the dependent measures indicate a main effect for perceived similarity: F(1, 53) = 2.54; p b 0.1 (one tailed) and an interesting interaction effect between perceived similarity and processing mode F(1, 53) = 2.82; p b 0.05 (one tailed). The main effect of processing mode is nonsignificant F(1, 53) = 0.20; n.s. Table 1 provides more detail To gain better understanding of the interaction between perceived similarity and processing mode, it is graphically presented in Fig. 1. The similarity effect is mainly due to the evaluative mode of processing and nearly non-existent in the experiential mode: (Experiential mode: Mean similarity = 5.78, Mean dissimilarity = 5.75, t = 0.05 n.s.; Evaluative mode: Mean similarity = 5.27, Mean dissimilarity = 6.64, t = 3.00; p b 0.01).
In the evaluative mode, and when the two commercials are perceived as dissimilar, an assimilation effect occurs between the attitude toward the positive emotional ad and the attitude toward the neutral ad. Moreover, in case of perceived similarity, it seems that a contrast effect occurs between the two ads leading to a lower Aad score for the neutral ad. In the experiential mode, the similarity effect is absent, but the observed difference between the two conditions (similarity versus dissimilarity) is not statistically significant. In short, a main effect of processing mode was not present in the current data, which is counter to our second hypothesis. However, and as predicted by our first hypothesis, more assimilation occurs between the affective reactions induced by two commercials perceived as dissimilar, in comparison with two ads described as alike. The main effect of the perceived similarity is also significant for the attitude toward the ad. Moreover and most importantly, our third hypothesis was supported, though less strongly than we had hoped for. Thus, we
6,8 6,6
Similarity Dissimilarity
6,4
Aad
6,2 6 5,8 5,6 5,4 5,2 5 EXPERIENTIAL
EVALUATIVE Mode
Fig. 1. Interaction effect between mode and similarity.
752
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
detected an interaction effect of perceived similarity and processing mode on the positive affective reactions (for both pleasure and arousal dimensions), and on Aad. This indicates that processing modes of consumers during ad exposure may influence the impact of the perceived similarity. Conversely, perceived similarity between commercials may influence ad processing as well. Moreover, it is worth noting that controlling the attention devoted to emotions (by addition of the verbal measure of attention devoted to emotions as covariate) does not drastically modify the results. The covariate strengthens the presence of the main and interaction effects for similarity and mode. 5. Discussion We have found empirical support for the presence of systematic cross-advertisement affectivity effects. Affective responses to earlier commercials in a pod act as the context in which consumers process subsequent commercials. Specifically, we obtained assimilation and contrast effects. When commercials are perceived to be dissimilar (e.g., because they are from different product categories), the positive affect evoked by the first commercial in a pod may assimilate to the next commercial, thus improving its effectiveness. Moreover, the processing mode that consumers are in during ad exposure also impacted on the observed assimilation effect. However, this effect was the reverse of what was expected. In case of an evaluative mode, more assimilation occurred between the positive affective reactions induced by two commercials, than in case of an experiential mode. The analysis of Schwarz and Clore (2003) suggests several potential explanations for this latter finding. Perhaps the result is due to the specific manipulation of the two conditions and to the subsequent measures. In fact, in case of the experiential mode, only the affective reactions elicited by the neutral test commercial were questioned. On the other hand, in the evaluative mode, participants were also asked about the affective reactions elicited by the emotional commercial. Therefore, in case of evaluative mode, the emotions elicited by the emotional commercial were perhaps more salient for the participants. To examine this possibility, a small follow-up study was carried out. In this second study, 75 participants were placed in an experiential mode and only the perceived similarity was manipulated. Based on the result of this second study, it appears that this salience effect may not be due to testing ad-induced affective reactions for both ads, but only due to the way of processing the ad and the instructions given in the evaluative mode. In fact, although in this second study, emotions induced by both ads were measured, no differences were found between the perceived similarity and dissimilarity conditions. Finally, the presence of an interaction effect between the similarity and the mode of processing was very interesting and opened the perspective for possible future research. The impact of the perceived similarity depends on how the ad is processed. This makes sense, for instance, if a particular product category is of interest and if the ad is processed more intensively. The
difference between the two commercials might, depending on the characteristic of the two commercials, appear to be larger or smaller. This effect was also shown for Aad. This effect is particularly interesting in a practical sense since most advertisers accept Aad as a measure of ad effectiveness. In fact, similarity influenced the affective reactions on Aad both directly and indirectly. The existence of assimilation effects between ads is a remarkable result taken into account the limited size of the sample. 6. Conclusion It is important to conduct follow-up research under more natural conditions to improve the ecological validity of the current results. Future research using less direct and reactive measures, such as non-verbal measurement, can be useful to corroborate our results. Finally, we used a small and student based sample. A replication of this study with a more representative sample might be a suggestion for the future. For a fuller understanding of the sequence effects phenomenon, new experiments focusing on other characteristics of ads are necessary. For instance, is the impact of a commercial eliciting positive affect different from the impact of a commercial eliciting negative affect on the affective reactions triggered by subsequent commercials? Are assimilation or contrast effects expected in this case? What happens when the intensity of the affective reactions elicited by the commercial or the product category is varied? It might also be interesting to take into account personal/individual characteristics of the participant, such as mood, involvement in the product category as a moderator of the relationship between the affective reactions elicited by the ad. Continuing this research by looking at the correction process used by consumers when facing several commercials would be interesting. Friestad and Wright (1994) stressed the need to take into account consumers' personal persuasion knowledge in order to understand the advertising process better. It is especially relevant in this context of research since we have observed an interaction effect between the processing mode and the perceived similarity. The current findings have implications for media planning and copy testing as well. Often, copy tests are conducted with single commercials in a random or not strongly controlled sequence of ads. The present results illustrate that the affective reactions elicited vary as a function of the contexts created by other commercials. Thus, it might be useful to test advertisements in sequences, and to control the affective intensity and the similarity to the other commercials in the pod. Moreover, commercials may have to be tested in different ad contexts to examine their malleability to different affective contexts, and with different processing modes. In real life, consumers process advertisements under various conditions, sometimes being focused on experiencing ads, and sometimes on comparing them, for instance. However, advertising research appears to have developed mostly around the implications of an evaluative processing mode (under high and low involvement), partly ignoring the influence that experiential
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
processing modes may have on advertising processing and effectiveness (see Meyers-Levy and Malaviya, 1999). As the present findings suggest, cross-advertisement effects differ by processing mode, and we speculate that more generally experiential processing modes may have quite different effects than the well-known evaluative modes. Moreover, since processing modes have indeed different effects on advertising responses, it may be worthwhile in media placement decisions to take this into account, by exploring if and how reception contexts induce different kinds of processing modes and how various advertisement versions can be optimally matched to the these processing modes.
753
Following the viewing of the advertisements, the respondents received the following instructions: Could you please now describe how you felt when watching advert X, the previous advert, and what your affective reactions were. Before answering, please try to relive what you felt during the advert. Of course, there is no right or wrong answer. In the experiential mode, the respondents were only questioned about their affective reactions triggered by the neutral test advert. However, in the evaluative mode, they were questioned about the two test adverts (the positive emotional advert and the neutral advert).
Acknowledgement The authors thank Claude Pecheux for useful comments on a previous version of this paper. Appendix A. Instructions To manipulate the ad treatment mode, the following instructions were used. Firstly, to stimulate a systematic and evaluative mode of treatment, the respondents were asked to read the following instructions before watching the ads: Adverts usually try to give a message or information to consumers. Only spectators can judge whether the message really comes across. We therefore need your opinion as a spectator to evaluate the adverts that you are going to see. Please follow the instructions carefully and remember that what we are interested in is your opinion as a spectator. There is of course no right or wrong answer. You will see a series of TV adverts. Please listen to the message and the content of the adverts and observe the techniques used by the advertiser to get the message over to the consumer. In short, please listen to the information being transmitted. Following the viewing of the adverts, the respondents received the following instructions: Some adverts can provoke ‘affective reactions’ in spectators. Only the spectators can judge if they really work. This is why we need your help to analyse these ‘affective reactions’, i.e. the emotions triggered by the adverts. For each reaction, we would like you to try to analyse the extent of this reaction. In order to stimulate an experiential treatment mode, the respondents were asked to read some other instructions before watching the advertisements: We are interested in the affective reactions triggered by adverts, such as how you feel and your emotions. You will now see a series of TV adverts and we would like you to watch these in a relaxed way, letting the emotions you feel when you watch them take over. Please be aware of what you are feeling when you watch these adverts and the affective reactions they have on you. In short, be aware of your emotions.
References Aaker D, Stayman D, Hagerty M. Warmth in advertising: measurement, impact and sequence effects. Journal of Consumer Research 1986;12:365–81 [March]. Abeele A, Gendolla G. Satisfaction judgments in positive and negative moods: effects of concurrent assimilation and contrast producing processes. Personality and Social Psychology Bulletin 1999;25(7):883–95 [June]. Aylesworth AB, MacKenzie SB. Context is key: the effect of program induced mood on thoughts about the ad. Journal of Advertising 1998;27:17–33 [Summer]. Broach C, Page Jr T, Wilson R. Television programming and its influence on viewers' perceptions of commercials: the role of program arousal and pleasantness. Journal of Advertising 1995;24(4):45–54. Broach C, Page Jr T, Wilson R. In: Wells W, editor. The effects of program context on advertising effectiveness in measuring advertising effectiveness. Mahwah NJ: Lawrence Erlbaum Associates; 1997. p. 203–14. Brooker G. A comparison of the persuasive effects of mild humor and mild fear appeal. Journal of Advertising 1981;10(4):9-40. Brown SP, Homer PM, Inman JJ. A meta-analysis of relationships between adevoked feelings and advertising responses. Journal of Marketing Research 1998;35:114–26 [February]. Coulter K. The effects of affective responses to media context on advertising evaluations. Journal of Advertising 1998;27(4):41–51. Davidson R. On emotion, mood and related affective constructs. In: Ekman Paul, Davidson Richard, editors. The nature of emotion. Oxford University Press; 1994. De Pelsmacker P, Geuens M, Van den Bergh J. Marketing communications. Essex: Pearson Education Limited; 2001. De Pelsmacker P, Geuens M, Anckaert P. Media context and advertising effectiveness: the role of context appreciation and context ad similarity. Journal of Advertising 2002;31(2):49–61. Derbaix C. The impact of affective reactions on attitudes toward the advertisement and the brand: a step toward ecological validity. Journal of Marketing Research 1995;32:470–9 [November]. Friestad M, Wright P. The persuasion knowledge model: how people cope with persuasion attempts. Journal of Consumer Research 1994(21):1-31. Goldberg M, Gorn G. Happy and sad TV programs: how they affect reactions to commercials. Journal of Consumer Research 1987;14:387–403. Kamins M, Marks L, Skinner D. Television commercial evaluation in the context of program induced mood: congruency versus consistency effects. Journal of Advertising 1991;20(2):1-14. Kent R. Competitive versus noncompetitive clutter in television advertising. Journal of Advertising Research 1993:40–6 [April]. Martin L, Achee J. Beyond accessibility: the role of processing objectives in judgment. In: Martin L, Tesser Abraham, editors. The construction of social judgments. Hillsdale, NJ: Lawrence Erlbaum; 1992. p. 195–216. Martin L, Seta JJ, Crelia RA. Assimilation and contrast as function of people's willingness and ability to expend effort in forming an impression. Journal of Personality and Social Psychology 1990;59:27–37.
754
I. Poncin et al. / Journal of Business Research 59 (2006) 745–754
McMullen M. Affective contrast and assimilation in counterfactual thinking. Journal of Experimental Social Psychology 1997;33:77-100. Meyers-Levy J, Malaviya P. Consumers' processing of persuasive advertisements: an integrative framework of persuasion theories. Journal of Marketing 1999;63:45–60. Pavelchak MJ Antil, Munch J. The super bowl: an investigation into the relationship among program context, emotional experience and ad recall. Journal of Consumer Research 1988;15:360–7. Petty R, D. Wegener. Flexible correction processes in social judgment: correcting for context-induced contrast. Journal of Experimental Social Psychology 1993;29:137–65. Pieters RG, Bijmolt TH. Consumer memory for television advertising: a field study of duration, serial position, and competition effects. Journal of Consumer Research 1997;23:362–72 [March]. Rossiter J, Percy L. Advertising communications and promotion management. New York: McGraw-Hill; 1997. Schumann D. Program impact on attitude toward TV-commercials. In: Saegert J, editor. Proceedings of the division of consumer psychology; 1986. p. 67–73. Washington DC. Schwarz N, Clore GL. Mood as information: 20 years later. Psychological Inquiry 2003;14(3–4):296–303.
Shapiro S, MacInnis D, Park CW. Understanding program-induced mood effects: decoupling arousal from valence. Journal of Advertising 2002;31 (4):15–26 [Winter]. Sherif M, Hovland CI. Social judgment: assimilation and contrast effects in communication and attitude change. New Haven, CT: Yale University Press; 1961. Singh S, Hitchon J. The intensifying effects of exciting television programs on the reception of subsequent commercials. Psychology and Marketing 1989;6 (1):1-31. Stapel D, Winkielman P. Assimilation and contrast as function of context–target similarity, distinctness and dimensional relevance. Personality and Social Psychology Bulletin 1998;24(6):634–46 [June]. Tavassoli N, Shultz C, G.Fitzsimons G. Program involvement: are moderate levels best for ad memory and attitude toward the ad? Journal of Advertising Research 1995:61–72 [September/October]. Yi Y. Contextual priming effects in print advertisements. Journal of Consumer Research 1990;17:215–22. Zhao X. Clutter and serial order redefined and retested. Journal of Advertising Research 1997:57–73 [September–October]. Zillmann D. Excitation transfer in communication mediated aggressive behavior. Journal of Experimental Social Psychology 1971:419–34.
Journal of Business Research 59 (2006) 755 – 764
Setting the tone with the tune: A meta-analytic review of the effects of background music in retail settings Francine V. Garlin ⁎, Katherine Owen School of Marketing, University of Technology, Sydney, PO Box 123, Broadway NSW 2007, Australia Received 1 April 2005; received in revised form 1 November 2005; accepted 1 January 2006
Abstract Among the many in-store elements purported to impact patrons, background music is a leading feature of academic enquiry [Turley LW, Milliman RE. Atmospheric Effects on Shopping Behavior: A Review of the Experimental Evidence. J Bus Res 2000;49(2):193–211.]. Collectively, research examines a range of retail contexts, focuses on many different dimensions, and, uses different methods to explore the phenomena of background music in commercial settings. Therefore, conclusions are difficult on the extent to which the influences of background music on customer behavior are generalizable. The purpose of this study is to synthesize the results of extant research to identify common effects and the circumstances under which they differ. Our meta-analysis uses a sample size of 148, taken from 32 studies. A conservative approach to the analysis reveal small-to-moderate, yet quite robust effects in terms of background music and the dependents: value returns, behavior duration and affective response. © 2006 Elsevier Inc. All rights reserved. Keywords: Background music; Consumer behavior; Retail; Meta-analysis
Retailers have long been aware of the potential for the perceptual elements in their store environments to influence customers' behavior, and apply a tacit approach based on experience and observation (e.g. Areni, 2003a,b,c; DeNora and Belcher, 2000), or a conscientious design of their trading space to optimise the customer experience and/or maximise business returns (Babin and Attaway, 2000). In their review of sixty published studies on the impact of the purchase and consumption environment, Turley and Milliman (2000, 195) note that, “Music is the most commonly studied general interior cue”. Similarly, our interest is focussed on the effects of background music, and the extent to which this contributes to customer value. We report on the findings of a work-in-progress involving a quantitative meta-analysis that aims to, inter alia, identify common value effects from the use of background music in retail settings. The research has been kindly sponsored by the Australasian Performing Right Association.
⁎ Corresponding author. Tel.: +61 2 9514 3522; fax: +61 2 9514 3535. E-mail address: [email protected] (F.V. Garlin). 0148-2963/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2006.01.013
1. Literature review By providing the right “clues” in the physical setting, customer value is not reducible to “functionality versus price. Instead it is composed of both the functional and emotional benefits customers receive, minus the financial and nonfinancial burdens they bear.” (Berry et al., 2002, 86). These researchers suggest that the emotions derived from the customer's sensory experience of the physical setting can be the primary influence on their response. An emotionally taxing environment can negatively impact patronage even more than price considerations. “[O]rganizations must manage the emotional component of experiences with the same rigor they bring to the management of product and service functionality” else they may be at risk of reducing the overall value of their offer (Berry, Carbone and Haeckel, 2002, 86). The “values” available from such efforts are fully supported by decades of academic research. The dominant underlying theoretical framework for the study of physical consumption settings derives from environmental psychology (DeNora and Belcher, 2000), which
756
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
examines the relationship between the stimulus-organismresponse (SOR). Using this paradigm Mehrabian and Russell (1974) developed a model (M–R Model) representing the influence of environmental cues on mediating emotional states (e.g. degrees of pleasure and arousal), which in turn result in either approach or avoidance behaviors. Simply put, “Consumers avoid unpleasant and approach pleasant environments” (Dube and Morin, 2001, 107). Since this was first adapted to a retail setting (Donovan and Rossiter, 1982), a stream of research has developed in the area. Among others, studies have found atmospheric effects to impact sales, amount spent, gross margin, actual and perceived time in the environment, patronage, unplanned purchases, brand/store image and evaluation, rate of purchase, pace of shopping, brand choice, brand switching and satisfaction (Turley and Milliman, 2000). In terms of background music, Turley and Milliman (2000) suggest that most researchers are concerned with time spent instore as the dependent variable. However the extensive literature search and review for the current study leads to a different conclusion. This is not unexpected since Turley and Milliman's study looked broadly at atmospheric effects and utilised only 16 papers related to music effects. We reviewed 157 papers, from which 150 papers explicitly discuss background music effects. From these 150 papers, we identified five categories of dependent variables, shown in Table 1 together with the proportion of papers examining each variable. Table 1 is followed by a brief overview of exemplar work from each dependent category.
not directly cause people to behave in certain ways … perceptions of the servicescape lead to certain emotions, beliefs and physiological sensations which in turn influence behaviors” (Bitner, 1992, 62). Many studies explore these intermediate relationships, particular music-mood effects (McGoldrick and Pieros, 1998). Enough research has been conducted in both psychology and marketing to confirm the predictive effect of music variables on mood (Tansik and Routhieaux, 1999). This is useful because music is one of the easiest ways managers can influence how their customers feel (Hosea, 2004). The other significant body of work exploring background music and affective variables is that which utilises the M–R Model. This model is most commonly used to represent the effect of music on the degree of pleasure and arousal experienced by customers. 1.2. Financial returns The evidence from both research and practice clearly suggests that customers' affective and cognitive responses to experiences in-store influence the likelihood of behaviors which directly impact an organization's financial returns. The literature reports on studies which highlight managers' implicit beliefs in the ability of background music to facilitate top- and bottom-line returns to business (Areni, 2003a,b; DeNora and Belcher, 2000). Other researchers have secured significant results using more objective measures, such as those included in Table 1 (Areni and Kim, 1993; Lammers, 2003; Milliman, 1982, 1986; North et al., 2000, 2003; Yalch and Spangenberg, 1993).
1.1. Affective variables 1.3. Attitude and perception Music in service settings can reduce even relatively extreme emotions such as intense anxiety (Lee et al., 2004; Tansik and Routhieaux, 1999). However, “the perceived servicescape does Table 1 Dependent variables studied in background music research⁎ Variable category
% Studies Variations or descriptors (n = 150)⁎⁎
Affective Financial returns Attitudinal/ perceptual
41% 25% 24%
Temporal effects
20%
Behavioral
10%
Mood, arousal, pleasure, emotion, nostalgia Value of sales, repeat purchase, items purchased, rate of spend, quantity purchased, gross margin Liking, brand loyalty, product evaluation, quality perceptions, experience satisfaction, perception of visual stimuli, service quality perceptions, price sensitivity, expectations, intentions, social identification, status perceptions Duration perceived/actual, service time, unplanned time, time to serve customers, time to decision-make, time to consume, duration of music listening Patronage frequency, store choice, behavior speed, affiliation, items examined/handled, in-store traffic flow, impulse behavior., recommend service, number of customers leaving before served.
⁎Where studies examine multiple dependent variables they are included in more than one category. ⁎⁎Reference list available on request from the authors.
Background music in-store is attributed to influence customer perceptions, specifically degree of attention and information processing of critical store elements such as visual stimuli and salesperson's arguments (Chebat and GelinasChebat, 1993; Chebat et al., 2001). In addition, North et al. (1999) suggest that music can ‘prime’ the selection of certain products by stimulating customers to recall related knowledge. Feelings of pleasure intensity and customer perceptions generated from background music effects can also enhance or create customers' attitude towards the store and/or its elements (See Dube and Morin, 2001; Gorn et al., 1993; Grewal et al., 2003; Oakes, 2003). 1.4. Temporal effects Several temporal dimensions are examined in the literature, including customer perceptions of time duration, actual time duration, time of day/week/year, wait time, and task completion time (e.g. decision-making, serving, consuming, shopping). Collectively, these studies reveal clear relationships between background music and the various temporal dimensions. Several musical variables have been manipulated in this research stream including tempo (e.g. Akhter et al., 1987; Holbrook and Gardner, 1993; Chebat, Gelanis-Chebat, and Filliatrault 1993), style/genre (North et al., 1999), volume
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
(Kellaris et al., 1996), complexity (North and Hargreaves, 1999) and modality (Kellaris and Kent, 1992). In general, the literature provides evidence of the effects of music compositional variables on a range of temporal dimensions. However this is a cautionary tale. The research also highlights the complexity of relationships between musical elements and behavioral outcomes. As with all research domains, this signifies the need for careful methodological design, since confounds and inconsistent findings are often the result of inconsistent operational definitions (see Turley and Milliman, 2000). 1.5. Behavioral variables A substantially smaller collection of papers has been included in the behavioral category. This is because many primary behaviors of interest are captured in the other dependent sets. In brief, the literature reports on the capacity for background music to encourage customers to affiliate with staff (Areni, 2003a; Dube et al., 1995) and with other customers (Mattila and Wirtz, 2001; Sweeney and Wyber, 2002), can increase the likelihood of exploring a store and browsing (Tai and Fung, 1997; Akhter et al., 1987), influence patronage and store choice (Areni, 2003b; Caldwell and Hibbert, 2002; Donovan and Rossiter, 1982; Grewal et al., 2003), and provide potential for increased impulse buying (Mattila and Wirtz, 2001). Lastly, background music has been shown to impact the speed with which customers shop or consume (Milliman, 1982; Oakes, 2000). 2. Purpose of the study The empirical studies that have examined the impact of background music on consumers have been applied in a variety of settings, utilizing a range of research methodologies. Therefore conclusions are not easily made on the extent to which effects are significant and generalizable to other contexts. While inherently difficult to determine an “exact” value of music in business backgrounds, this does not preclude the need for furthering an understanding of these effects and their magnitude. This study aims to synthesize the results of existing empirical studies into the effects of background music in retail settings. From the analysis we anticipate the results will identify common value effects and the circumstances under which they differ.
757
which is best suited to the objectives of the research and the outcomes sought. Meta-analysis is a quantitative methodology that seeks to objectively integrate the results of a large volume or complex research area. In the context of the proposed research we aim to identify the key impact and values of background music and, where possible, indicate the magnitude of the effect of these variables on purchase behavior and/or business investment returns. The review includes published studies, but where possible we have sought to identify and obtain unpublished work to minimize bias. Following an initial categorisation of the collection of studies into methodological subsets, we operationalised the research objectives (the impact and value of background music) with working hypotheses and a clear set of criteria on which to assess the studies in each subset. The SAS statistical package was used to undertake the first meta-analysis enquiry. 3. Methodology 3.1. Literature search The literature search focused on identifying all published studies on the effects on customers or staff of background music in retail or other business settings. The collection commenced with Smith and Curnow (1966) who were the first to address background music effect. A total of 11 databases were searched including: ABIInform, Academic Search Elite, Business Source Premier, Communication and Mass Media Complete, Expanded Academic, Google, Hospitality and Tourism Index, Medline, Professional Development Collection, Psychinfo and Science Direct. Keyword combinations applied included: background music or music and customer/consumer, musical effect, marketing, patron, purchase, buying, environment(al), atmospheric(ere), servicescape, retail, restaurant, environmental psychology, duration, behavior. References from bibliographies were also examined to identify further studies. Finally, we pursued non-published studies over the last 5 years through Google, through sites of active authors, and through direct enquiry to authors. 3.2. Inclusion criteria
The overall objectives of the research are to: 1) identify the impact of using music in the retail context; 2) identify the qualitative and quantitative values of using music in the retail context; and, 3) explore the implications for return on investment from the use of music in a retail context.
Studies that were included in the meta-analysis studies had to satisfy the following criteria. They had to: 1) have studied background music affects on patrons/clients/staff; 2) show a direct or clear indirect affect on purchase behavior or customer loyalty; and 3) have reported sample sizes, outcome statistics (e.g. t, F, χ2) or information such as group means and standard deviations that enable computation of the d statistic recommended by Arthur et al. (2001).
2.2. Overview of research design
3.3. Coding scheme
The study uses meta-analysis to provide a comprehensive review of the empirical studies in the research area—a design
The coding scheme developed for the meta-analysis incorporated three sections: effects of background music,
2.1. Research objectives
758
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
known or used moderators, the type(s) of dependent variable(s) used in studies (Table 1), and covariates. Also coded was the type of study and method of analysis. Prior to commencing the coding we collected abstracts from the final collection of studies (150) as an initial check on their eligibility for coding. This resulted in further articles being excluded because of inadequate data, problems in experimental design that negated results, and methods or data not suited to our needs. 3.4. Meta-analysis procedure To compare the effects between studies the outcomes were converted into a common metric. In this analysis we use the sample-weighted mean d statistic developed by Glass (1976 in Glass et al., 1981) and outlined in Arthur et al. (2001): d¼
ME −MC SW
ð1Þ
where ME is the mean of the experimental or manipulated group and MC is the mean of the control or reference group, and SW is the pooled, within-group standard deviation. Although the correlation coefficient (r) is often considered the preferred statistic for meta-analysis, (Hunter and Schmidt, 1990) this method is appropriate for post-test-only studies which is the majority of studies into background music affects. Further, the advantages from the use of r are similarly available from the d statistic, viz, “d is not a function of sample size, and, because it is calculated as a difference of two proportions, also can be interpreted readily” (D'Alessio and Allen, 2000, 139). Where only test statistics were available (e.g. t, F, χ2 ) these were converted to d using the appropriate conversion formula and adjustments were made for unequal group sizes (Arthur et al., 2001; Rosnow and Rosenthal, 2003). In the absence of group sizes we made the assumption they were equal. In instances where only significance levels (e.g. p = .03) and means were reported an estimate was made for the appropriate test, and where only a “not significant” was given, a coarse estimate of the statistic was made (e.g. 1.0 for F). In many studies no variances were reported. These were estimated where means were reported by calculating the differences between means and dividing this by the F test to arrive at an approximation of the means squared error (i.e. MSE = BSE / F test). Prior to analysis we examined the magnitude and variance of effect sizes across the different test statistics. Effect sizes calculated from studies that provided means and standard deviations had the highest mean effect (m = 0.53, standard deviation of 0.25 and number of samples 34 with a mean sample size of 123). The effects for t tests were similar (0.51) while the F tests (F = 1 or F N 1) were lower at 0.35 and 0.42 with average samples of 123 and 138, respectively, and standard deviations of 0.25 and 0.28. The effect most requiring estimation was from the F test. Our concern was to err on the side of conservative estimates of the effects of background music. While a coarse comparison, this indicates that we are more likely to under- than overestimate the effects.
Once all effects were calculated a further adjustment was made to those that had used measurement scales for dependent or independent variables (Arthur et al., 2001). These scales are invariably less than perfect measures of an outcome and so tend to underestimate the effect. Following these adjustments the sample effect is calculated using individual and total sample sizes to provide a sampleweighted mean effect. X d¯ ¼ di ni =NT ð2Þ where d¯ is the mean effect size, di is the effect size and ni the sample size for sample i, and NT is the combined sample size from all studies. This mean is further adjusted to arrive at a mean corrected for unequal sample size, which avoids attenuation of effects due to substantial differences in sample sizes (Hunter and Schmidt, 1990). Hunter and Schmidt (1990) propose the following adjustment in the form of a multiplier (A): A = (1 + (0.75 / N¯ − 3)), where N is the average sample size across studies. 4. Analysis and results The final sample for analysis comprised 148 data points (samples) from 32 studies which are listed in the Appendix. These were then separated into three groups: Value, Affect, and Duration. Of the three, only studies in the value group were directly linked to some form of purchase behavior or intentions. Affect and Duration indirectly effect purchase behaviors as previously outlined, and so are discussed in the context of the links hypothesised in each area. The magnitude of effects is assessed using three criteria. First is the size of the effect. Following Cohen (1988) effects are evaluated as small (0.2), medium (0.5), and large (0.8). In the context of background music we would not expect to see large effects, this being one of many environmental factors that affect behaviour. The expectation is that effects will range from small to medium. To determine whether underlying moderators are influencing the results we assess the percentage of variance that is due to sampling error. For example, if this is around 100% we know that all variance in sample means is simply random and that underlying moderators are unlikely. A general rule of thumb is that if 75% of the variance can be attributed to error, then the likelihood is that the effect is the true effect size for that population (Arthur et al., 2001). Supporting this test is a chisquare test for homogeneity where rejection of the test is a further indication of possible moderators. Finally, we calculate confidence intervals to determine the range the effect sizes can take. The wider the range the more variable an effect can be across studies (also indicating the possibility of underlying moderators). The confidence intervals are graphed for easy reference. 4.1. Value The value sample included studies that examined the effects of background music on sales/purchases, intention to purchase
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
or patronise, intention to return or recommend, and evaluation of service/products. Where a study reported several different purchase outcomes we chose the one that would be most consistent with other studies or that was likely to be more reliable. For example, Milliman (1986) reports individual purchases and daily gross sales from a supermarket where each day receives a different music condition. We included the latter to reduce the likelihood of confounding individual specific factors of those they surveyed. Areni and Kim (1993) reported number of purchases and value of purchases. The latter was taken because this is comparable to other studies. The effects examined in these studies included tempo, volume, complexity, genre, liking/familiarity and absence/ presence of music. Of the 20 studies included 13 were from natural settings and five of these in hospitality. In all laboratory settings a service or retail environment was simulated. A total of 41 data points were used in the analysis. Effects were reorganised to reflect the expectation that music as opposed to no music, lower tempo, lower volume and less complexity would be associated with greater purchases or a more favourable view of the venue (e.g. would recommend, return or evaluate favourably). Of these, ‘evaluation’ is likely to be the most ambiguous because of the influence of fit. Similarly, genre per se has no hypothesised direction. In all studies we included, the genre effect “classical” is compared to another form, mostly “popular”. This is because classical with some other genre has been the most common comparison. Differences in genre effects will also be evident in the absence–presence analysis if they are significant because these include a range of genres compared against no music. Highlighted earlier was that in many studies researchers examined several manipulations of an effect (e.g. 3 genre). This results in some overlap in samples because each comparison is treated as a separate data point and coded accordingly (this is particularly the case where comparisons were against a control condition and where we changed genre comparisons to presence–absence of music). In analysis we included each effect regardless of overlap but checked whether results were sensitive to sample size or the omission of one of the effects. In all but one instance including all effects did not impact the overall result. The exception was Wilson (2003) where the study sample was 300 and she reported five genres and no music. For analysis each genre was coded against the no-music group (50 × 2 × 5) resulting in an implicit sample of 500. To counter this inflation the sample of each comparison was reduced so that the sample of the combined groups was 300. Table 2 presents the results for the full sample and each primary effect area. The confidence intervals are illustrated in Fig. 1. We have combined the studies of tempo, volume and complexity because their movement from low-to-high/fast volume/tempo and less-to-more complex music is arousing, although these could have different consequences. Only one of these studies examined complexity, while three examined volume. Turning first to the overall sample, significant variation is evident in the effects of underlying studies given the high rejection for chi square. Familiarity/liking comprised five data
759
points from three studies with a sample-weighted mean of .35 indicating a small but clear positive effect on patronage. All the variance is due to error and so these studies can be considered to be representative of a population of studies. That is, the result is quite robust. This is also the case for the absence–presence of music. Almost regardless of the type of music, its presence has a small positive affect on patronage that is also robust. The result for genre indicates a medium positive outcome from playing classical as opposed to other forms of genre. However, this is clearly not always the case as indicated by the confidence interval. In a study by Lammers (2003) the alternative music was soft rock, with no significant difference between the two genres in terms of patronage outcome. Excluding this study from the sample increases the effect to 0.66 and reduces the variance across studies to only just significant. The final effect given in Table 2 is the combined tempo, volume and complexity comparison. The sample-weighted mean is almost zero with significant variation among the studies. The low mean is persistent whether only tempo or volume is examined and if the studies are separated by sales/ intentions and evaluations. Clearly context plays a significant role and needs to be explored. 4.2. Duration Duration has been examined based on actual time spent (or pace) in a location or at an activity, and/or respondents' perception of time spent. Of 19 studies that had examined duration eight reported affects on actual duration and five of these were in a natural setting. A total of 11 studies examined perceived duration. For the purpose of analysing the results actual duration and perceived duration were separated because the effect of music is different to the outcome. In looking at the effect of music on duration/pace in a retail environment most research has reported that consumers stay longer when some form of background music is present as opposed to no music, when the music is low rather than high in volume (Sullivan, 2002), is slow rather than fast tempo (Milliman, 1982), is liked (Caldwell and Hibbert, Table 2 Value: results for the full sample and corresponding primary effect Statistics
All
Familiar/ like
Tempo, vol, complexity
Absence– presence
Genre
Data points Total sample size Sampleweighted mean Corrected SD % Variance sampling error
39 5338 0.14
5 387 0.35
13 3542 − 0.05
16 1222 0.28
6 454 0.50
0.23 34.99
0.00 100.0
0.20 27.46
0.07 90.83
0.44 22.08
0.35 0.35 3.14
− 0.44 0.33 47.35
0.14 0.43 17.62
− 0.36 1.37 27.17
ns
p = .05
ns
p = .05
Confidence interval 95% Lower 0.60 Upper 111.45 Chi square p = .05 significance
760
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764 1.50 1.00 0.50 0.00 -0.50
All
Familiar/like
Tempo
Absence-Presence
Genre
-1.00
Fig. 1. Sample-weighted mean and confidence intervals.
2002), and possibly lower in complexity. Seven of the eight studies examined one of these effects and a further study, genre (Areni and Kim, 1993). The latter was excluded from analysis because no hypothesis was established as to how genre per se should affect duration. Table 3 show three sets of results covering the full sample, an adjusted sample and a tempo-only sample. The sampleweighted means are all negative indicating that that reduced tempo/volume and more familiar music resulted in subjects staying marginally longer at a venue than if the tempo or volume was high or the music less familiar. The mean effect of the full sample is higher than the other results but this also has a wider confidence interval, a low error variance, and fails the test for homogeneity of variance. This is indicative of moderating factors that are influencing the results. A primary source of this variance is from a study by McElrea and Standing (1992) in which they examined the pace at which a sample of women (40) consumed a soft drink under slow and fast tempo. The range of tempo was 54–132 which is substantially wider than most studies and resulted in a very high consumption pace under the high tempo condition. In the adjusted sample this study was omitted. Both the adjusted and tempo solutions indicate the sample of studies can be considered to come from the same population (i.e. homogeneous variance with the former just rejected) and so no apparent moderating effects are present (although the error variance is substantially below 75% which conflicts with this). The sample-weighted means are at the lower range of possible effects indicating that the effect of background music on duration of stay is small. However, the range of effect can vary from almost non-existent to moderate as given by the confidence intervals (Fig. 2).
The second area examined under duration is perceived duration. Most of these studies focus on the effect of background music on perceived waiting. For retailers the concern is to minimise waiting time and/or for the time to pass as pleasantly as possible. Perceived duration was examined in 11 studies and most were concerned with perceived waiting time. All but one of these studies was carried out in laboratory or simulated environments. Three studies used simulated retail environments and another used a service environment (student registration). As with actual duration Yalch and Spangenberg, 1993 was excluded because the study tested for genre. All samples indicate the presence of moderators. A major source is the Bailey and Areni (2006) study which manipulated the number and duration of musical pieces under active and passive conditions. Their results are forthcoming and not discussed here. The only study in a natural setting (Smith and Curnow, 1966) examined the effect of volume on perceived duration and had a particularly low effect of .07 and the most substantial sample size (1100). However, no volume levels were given, which raises the question whether this study should remain in the analysis. Omitting both of these studies substantially reduced the underlying heterogeneity and increased the sample-weighted mean to 0.25. The positive mean indicates that the higher the volume, tempo and less liked the music the longer the perceived duration. The effect is more marked for studies examining tempo, with the effect at 0.39. The very low effect of 0.05 for liking/familiarity is a result of the Bailey and Areni (2006) manipulation of length and number of musical pieces. If these are excluded the effect increases to .34 (χ2 = 3.6) with 3 studies and 477 in the sample (Table 4) (Fig. 3). 4.3. Affect
Table 3 Duration: results for the full sample and tempo effect Statistics
Full sample
Adj sample
Tempo
Data points Total sample size Sample-weighted mean Corrected SD % Variance sampling error
9 1518 −0.32 0.28 23.64
8 1478 − 0.28 0.16 44.91
4 1185 − 0.22 0.12 50.19
−0.86 0.23 38.08 p b 0.05
− 0.60 0.04 17.82 p b 0.05
− 0.45 0.01 7.97 ns
Confidence interval 95% Lower Upper Chi square significance
The final area examined is affect. The MR model is the most cited for affect studies, which comprises three dimensions: 0.40 0.20 0.00 -0.20 -0.40 -0.60 -0.80 -1.00 Full sample
Adj sample
Tempo
Fig. 2. Sample-weighted mean and confidence intervals.
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
761
Table 4 Perceived duration: results for the full sample and corresponding effects
Table 5 Affect: results for the full sample and corresponding effects
Statistics
Full sample
Adj sample
Tempo
Fam/liking
Statistics
Data points Total sample size Sample-weighted mean Corrected SD % Variance sampling error
19 3178 0.15 0.36 15.61
16 1698 0.25 0.30 29.17
4 497 0.39 0.31 24.95
11 1359 0.05 0.46 13.36
Pleasure absence– presence
Pleasure tempo
Arousal tempo
Data points Total sample size Sample-weighted mean Corrected SD % Variance sampling error
7 698 0.38
7 703 0.27
6 675 0.48
0.22 45.28
0.25 39.45
0.38 20.14
− 0.06 0.81 15.46 p = .05
− 0.22 0.75 17.75 p = .05
−0.27 1.23 29.79 p = .05
Confidence interval 95% Lower Upper Chi square significance
− 0.56 0.85 121.74 p = .05
− 0.35 0.84 54.86 p = .05
− 0.23 1.01 16.03 p = .05
− 0.85 0.95 82.32 p = .05
arousal, pleasure, and dominance. The common hypotheses are that increased pleasure is associated with higher evaluations of a service or venue, and increased arousal with a greater tendency to affiliate with other patrons or with staff, although several other outcomes are possible. Effects that have been examined include, genre, tempo, volume, mode, complexity, familiarity and liking and absence–presence of music. The consensus among many researchers in this stream is to exclude ‘dominance’ from studies, as prior research indicates that dominance is of little predictive value (Sherman et al., 1997). Although studies including dominance were initially coded for analysis they were subsequently excluded. We collected 61 data points from 15 studies that covered pleasure (25), arousal (14), and dominance (9). A further nine studies employed various measures relating to “satisfaction”, but these are not examined here. Despite the high number of data points many of the effects have only been examined in one study or study group (e.g. complexity (North and Hargreaves, 1999, 1996), mode (Kellaris and Kent, 1993). Only three effects had sufficient studies for analysis: absence–presence and tempo for pleasure, and arousal in the pleasure dimension. These represent 6, 5, and 5 studies respectively and 7, 7, 6 data points. 1.50 1.00 0.50 0.00 -0.50 -1.00
Full sample
Adj sample
Tempo
Famil/like
All effects exhibit wide variation as illustrated in Fig. 4. Absence–presence has a small-to-moderate positive effect on pleasure but clearly underlying factors that influence pleasure exist. Most of the variance comes from a study by Yalch and Spangenberg (1988, 1990) where they monitored and then interviewed shoppers in the clothing section of a department store and found age to be a moderating factor in pleasure. The effect for tempo was highest for arousal as would be expected since higher tempo is not always pleasurable. The high variance across studies is primarily from Mattila and Wirtz (2001) where the arousal effect was significantly lower than for other studies. Unfortunately they provided only the source of their music and not its musical characteristics (beyond low and high arousal) and so precluding our ability to assess whether the nature of the tempo differed from the other studies (Table 5). 5. Conclusion The literature indicates numerous influences on customer behavior from the use of music in a retail context. This was supported in the meta-analysis that used standardised measures of the effect of musical characteristics on key financial, behavioral, attitudinal and affective factors to assess how broadly applicable these many findings are. These affects were small-to-moderate but are clearly evident in the following relationships:
Arousal Tempo
• Familiarity/liking has a positive effect on patronage; • The mere presence of music has a positive effect on patronage as well as felt pleasure; • Slower tempo, lower volume and familiar music results in subjects staying marginally longer at a venue than when the tempo or volume are high, or the music less familiar; • A higher volume and tempo, and the less liked the music, the longer customers perceive time duration. This has most implications for waiting customers. • Tempo has the greatest effect on arousal.
Fig. 4. Sample-weighted mean and confidence intervals.
The consistency of the findings in this study with the literature connotes implications for return on investment from
Fig. 3. Sample-weighted mean and confidence intervals.
1.5 1 0.5 0 -0.5
Confidence interval 95% Lower Upper Chi square significance
Pleasure Absencepresence
Pleasure Tempo
-1
762
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
the use of music in a business context. Taken collectively, past research demonstrates the capacity for appropriate background music to illicit positive effects on affective, attitudinal/ perceptual, temporal and behavioral variables. A considerable body of work presents evidence for these effects to provide returns to business in the form of sales value and volume, repeat purchase, rate of spend, quantity purchased and gross margin. Many indirect returns to business are apparent, such as positive perceptions of quality and venue/store brand image. 5.1. Limitations and future research The most significant limitation faced in this study relates to the limitations evident in many of the studies sourced for the meta-analysis. Studies were excluded because they did not report useable statistics, or they had serious design issues that comprised the results. In the main these were a result of confounding effects or poor experimental design, such as the lack of a control group or poor balance across conditions in field studies. This also limited the number of variables we could adequately analyse, and precluded a more comprehensive analysis of moderator effects and complex relationships. Unfortunately, most of these problems arose in field settings which are inherently more difficult to design and control, therefore we could not benefit from a more extensive inclusion of studies undertaken in naturalistic versus laboratory environments. Several studies that were included in the meta-analysis utilized incomparable statistics requiring adjustment. The possibility of inflating results is a consequence; however we countered this possibility by taking a conservative approach with our estimations. Finally, the meta-analysis and review is unlikely to include the population of existing work in this area of research. In particular studies may not have been published, perhaps due to the absence of significant effects. In terms of future research, well-designed replication studies that aim to overcome some of the limitations evident in prior research are clearly required. Despite the inherent difficulties, field studies outside the laboratory could increase our understanding of the background music phenomena. Future research examining the relationship between background music, intermediating cognitive and affective processes, and consequent behaviors are clearly warranted. Currently, the majority of studies have focused on intermediate effects of music alone or have ignored the intermediate effects altogether. The meta-analysis also indicates the need to examine the effects of context, and to explore the types and nature of underlying factors that are influencing or confounding variables of interest. More specifically, what is the relationship of background music with other elements of the ambient environment, and how might this influence behaviour? What are the similarities or differences between the effects of background music on patrons versus staff? Do different facets of the musicscape elicit different responses from patrons compared to staff? For example, ‘familiarity’ of music played in-store could have a very different effect on staff who might be more accustomed to the repertoire than customers might.
Finally, a reviewer of this paper asked “a question that cannot be answered with data: Are researchers studying the right variables … what else should we be studying?” Inspired by this, we suggest that researchers interested in the influence of the music stimulus itself explore the individual qualities of music using a set of non-evaluative attributes rather than genres. In a consumer study conducted by Garlin and McGuiggan (2002) various attributes were used in lieu of genre to describe movie content. Genre use in leisure-media research is criticised as being a highly limited unit of analysis, which is subject to wide interpretations between individuals and between producers and audience (Garlin and McGuiggan, 2002). The same situation clearly exists with music genre. Music researchers have identified the use of genre classifications as highly problematic. “No unique position can be taken regarding genre …. music genre is an ill-defined notion, that is not founded on any intrinsic property of the music, but rather depends on cultural extrinsic habits” (Aucouturier and Pachet, 2003, 83, 84). Like film genre, these authors note that music genre is dynamic and evolves with the appearance of new genres, merging genres and subgenres. Many nuances between musical pieces may also be differentially perceived by consumer researchers into background music effects and their subjects alike. “What is the difference between ‘Classical’ and ‘Opera’? …. What exactly is ‘Modern’ compared to ‘Jazz’”, or could the genre of “an orchestral rendering of Gershwin's Porgy and Bess” be perceived as the same as “a Prokoviev Symphony piece” by the uninitiated customer? (Aucouturier and Pachet, 2003, 87, 88, 90). The reason that genre use is raised as a significant issue for consideration is due to the high incidence of genre as a variable of interest in background music studies when this may be the inappropriate unit of analysis. Using a taxonomic approach, compiling musical attributes derived from consumers' and retailers' own descriptions as well as extant literature, would be a starting point. Appendix A. List of studies included in the meta-analysis Alpert JI. Apert MI. Background music as an influence in consumer mood and advertising responses. Advances in Consumer Research 1989; 16 (1): 485–91. Areni CS. Kim D. The influence of background music on shopping behavior: Classical versus top-forty music in a wine store. Advances in Consumer Research 2003; 20(1): 336. Bailey N. Areni CS. When a few minutes sounds like a lifetime: Does atmospheric music contract perceived time? Journal of Retailing 2006; 82(3). Caldwell C. Hibbert SA. The influence of music tempo and musical preference on restaurant patron's behaviour. Psychology and Marketing 2002; 19 (11): 895–917. Cameron MA. Baker J. Peterson M. Braunsberger K. The effects of music, wait-length evaluation, and mood on a lowcost wait experience. Journal of Business Research 2003; 56(6): 421–530. Chebat J-C. Gelinas-Chebat. C. Fillatrault P. Interactive effects of musical and visual cues on time perception: An application to waiting lines in Banks. Perceptual and Motor Skills 1993; 77(3): 995–1021.
F.V. Garlin, K. Owen / Journal of Business Research 59 (2006) 755–764
Dube L. Chebat J-C. Morin S. The effects of background music on consumer's desire to affiliate in buyer selling. Psychology and Marketing 1995; 12 (4): 305–19. Dube L. Morin S. Background music pleasure and store evaluation Intensity effects and psychological mechanisms. Journal of Business Research 2001; 54. Grewal D. Baker J. Levy M. Voss GB. The effects of wait expectations and store atmosphere evaluations on patronage intentions in service-intensive retail stores. Journal of Retailing 2003; 79:259. Groenland EAG. Schoormans JPL. Comparing MoodInduction and Affective Conditioning as Mechanisms Influencing Product Evaluation and Product Choice. Psychology and Marketing 1994; 11:183–197. Holbrook MB. Gardner MP. An approach to investigating the emotional determinants of consumption durations: why do people consume what they consume for as long as they consume it? Journal of Consumer Psychology 1993; 2 (2): 123–42. Hui MK. Dube L. Chebat J-C. The impact of music on consumer's reactions to wait