Review of Marketing Research (5) [5] 0765621258, 9780765621252


307 86 1MB

English Pages 217 Year 2008

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Review of Marketing Research......Page 2
REVIEW OF MARKETING RESEARCH......Page 6
AD HOC REVIEWERS......Page 7
CONTENTS......Page 8
PUBLICATION MISSION......Page 10
ARTICLES IN THE FIRST VOLUME......Page 11
ARTICLES IN THE THIRD VOLUME......Page 12
ARTICLES IN THE FOURTH VOLUME......Page 13
ARTICLES IN THIS VOLUME......Page 14
REFERENCES......Page 17
CONTENTS, VOLUME 1......Page 18
CONTENTS, VOLUME 2......Page 19
CONTENTS, VOLUME 3......Page 20
CONTENTS, VOLUME 4......Page 21
Introduction......Page 24
Intuition and Analysis as Distinct Systems of Thought......Page 25
Determinants of System Engagement and Utilization......Page 31
Application to Topics in Consumer Behavior......Page 35
Persuasion Knowledge......Page 36
Emerging Issues for Marketing Research......Page 44
Conclusions......Page 51
References......Page 52
Introduction......Page 59
Innovators as Influential......Page 60
Life According to Rogers, Bass, and Moore......Page 63
Data Analysis......Page 68
Empirical Results and Implications......Page 70
Discussion......Page 73
References......Page 76
Introduction......Page 79
Conceptual Framework......Page 82
The Empirical Study......Page 83
Conclusions and Directions for Future Research......Page 90
References......Page 93
Introduction......Page 96
Optimal Positioning in STP......Page 100
The Proposed Clusterwise Unfolding Model......Page 103
Application: Portable Telephones......Page 110
Discussion......Page 118
References......Page 122
Introduction......Page 125
Alternative Methods for Massive Number of Attributes......Page 126
Details of the Methods and Applications......Page 127
A Comparison of Methods......Page 144
References......Page 148
Introduction......Page 151
Laddering......Page 154
Proposed Resolution......Page 166
Study......Page 169
Notes......Page 178
References......Page 179
Appendix A. Glossary of Laddering Terms......Page 185
Appendix B. Computation of Laddering Quality Metrics......Page 187
Appendix C. Code Distributions......Page 193
Appendix D. Summary of Quasi-Reliability and Quasi-Validity Measures by Laddering Method......Page 195
Introduction......Page 196
Internet Advertising......Page 197
The Internet as a Social Medium and Online Word-of-Mouth......Page 205
Conclusion......Page 210
References......Page 211
ABOUT THE EDITOR AND CONTRIBUTORS......Page 214
Recommend Papers

Review of Marketing Research (5) [5]
 0765621258, 9780765621252

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Review of

Marketing Research

Review of Marketing Research VOLUME 5

Naresh K. Malhotra Editor

M.E.Sharpe Armonk, New York London, England

Copyright © 2009 by M.E.Sharpe, Inc. All rights reserved. No part of this book may be reproduced in any form without written permission from the publisher, M.E. Sharpe, Inc., 80 Business Park Drive, Armonk, New York 10504.

Library of Congress ISSN: 1548-6435 ISBN: 978-0-7656-2125-2 Printed in the United States of America The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences Permanence of Paper for Printed Library Materials, ANSI Z 39.48-1984.

N

IBT (c) 10

9

8

7

6

5

4

3

2

1

CHAPTER TITLE

REVIEW OF MARKETING RESEARCH EDITOR: NARESH K. MALHOTRA, GEORGIA INSTITUTE OF TECHNOLOGY

Editorial Board Rick P. Bagozzi, University of Michigan Russ Belk, University of Utah Ruth Bolton, Arizona State University George Day, University of Pennsylvania Morris B. Holbrook, Columbia University Michael Houston, University of Minnesota Shelby Hunt, Texas Tech University Dawn Iacobucci, Vanderbilt University Arun K. Jain, University at Buffalo, State University of New York Barbara Kahn, University of Miami Wagner Kamakura, Duke University Donald Lehmann, Columbia University Robert F. Lusch, University of Arizona Nelson Oly Ndubisi, Monash University, Malaysia Debbie MacInnis, University of Southern California A. Parasuraman, University of Miami William Perreault, University of North Carolina Robert A. Peterson, University of Texas Nigel Piercy, University of Warwick Jagmohan S. Raju, University of Pennsylvania Brian Ratchford, University of Texas, Dallas Jagdish N. Sheth, Emory University Itamar Simonson, Stanford University David Stewart, University of California, Riverside Rajan Varadarajan, Texas A&M University Michel Wedel, University of Maryland Barton Weitz, University of Florida

v

v

vi

APPLIED THEORY IN WORKPLACE SPIRITUALITY

AD HOC REVIEWERS Kevin Crowston, Syracuse University Renana Peres, The Hebrew University of Jerusalem, Israel Ashutosh Prasad, University of Texas at Dallas S. Siddarth, University of Southern California Catarina Sismeiro, Tanaka School, Imperial College, London Katherine Stewart, University of Maryland

vi

CHAPTER TITLE

vii

CONTENTS Review of Marketing Research: The First Five Volumes Naresh K. Malhotra Contents, Volume 1 Contents, Volume 2 Contents, Volume 3 Contents, Volume 4

ix xvii xviii xix xx

1. Consumer Judgment from a Dual-Systems Perspective: Recent Evidence and Emerging Issues Samuel D. Bond, James R. Bettman, and Mary Frances Luce

3

2. Can You See the Chasm? Innovation Diffusion According to Rogers, Bass, and Moore Barak Libai, Vijay Mahajan, and Eitan Muller

38

3. Exploring the Open Source Product Development Bazaar Balaji Rajagopalan and Barry L. Bayus

58

4. A New Spatial Classification Methodology for Simultaneous Segmentation, Targeting, and Positioning (STP Analysis) for Marketing Research Wayne S. DeSarbo, Simon J. Blanchard, and A. Selin Atalay

75

5. Methods for Handling Massive Numbers of Attributes in Conjoint Analysis Vithala R. Rao, Benjamin Kartono, and Meng Su

104

6. A Review and Comparative Analysis of Laddering Research Methods: Recommendations for Quality Metrics Thomas J. Reynolds and Joan M. Phillips

130

7. Metrics for the New Internet Marketing Communications Mix Randolph E. Bucklin, Oliver J. Rutz, and Michael Trusov

175

About the Editor and Contributors

193

vii

REVIEW OF MARKETING RESEARCH

ix

REVIEW OF MARKETING RESEARCH The First Five Volumes NARESH K. MALHOTRA

OVERVIEW Review of Marketing Research, now in its fifth volume, is a fairly recent publication covering the important areas of marketing research with a more comprehensive state-of-the-art orientation. The chapters in this publication review the literature in a particular area, offer a critical commentary, develop an innovative framework, and discuss future developments, as well as present specific empirical studies. The first five volumes have featured some of the top researchers and scholars in our discipline who have reviewed an array of important topics. The response to the first four volumes has been truly gratifying, and we look forward to the impact of the fifth volume with great anticipation. PUBLICATION MISSION The purpose of this series is to provide current, comprehensive, state-of-the-art articles in review of marketing research. Wide-ranging paradigmatic, theoretical, or substantive agendas are appropriate for this publication. This includes a wide range of theoretical perspectives, paradigms, data (qualitative, survey, experimental, ethnographic, secondary, etc.), and topics related to the study and explanation of marketing-related phenomena. We reflect an eclectic mixture of theory, data, and research methods that is indicative of a publication driven by important theoretical and substantive problems. We seek studies that make important theoretical, substantive, empirical, methodological, measurement, and modeling contributions. Any topic that fits under the broad area of “marketing research” is relevant. In short, our mission is to publish the best reviews in the discipline. Thus, this publication bridges the gap left by current marketing research publications. Current marketing research publications such as the Journal of Marketing Research (USA), International Journal of Marketing Research (UK), and International Journal of Research in Marketing (Europe) publish academic articles with a major constraint on the length. In contrast, Review of Marketing Research will publish much longer articles that are not only theoretically rigorous but also more expository, with a focus on implementing new marketing research concepts and procedures. This will also serve to distinguish this publication from Marketing Research magazine, published by the American Marketing Association (AMA). ix

x

REVIEW OF MARKETING RESEARCH

Articles in Review of Marketing Research should encompass the following: ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀

฀ ฀ ฀ ฀ ฀ ฀ ฀

฀ ฀ ฀ ฀ ฀ ฀ ฀

฀ ฀ ฀ ฀ ฀



฀ ฀

฀ ฀

฀ ฀







฀ ฀

฀ ฀ ฀ ฀



฀ ฀

฀ ฀

฀ ฀



฀ ฀







฀ ฀



฀ ฀



ARTICLES IN THE FIRST VOLUME The inaugural volume exemplified the broad scope of the Review of Marketing Research. It contained a diverse set of review articles covering areas such as emotions, beauty, information search, business and marketing strategy, organizational performance, reference scales, and correspondence analysis. These articles were contributed by some of the leading scholars in the field, five of them being former editors of major journals (Journal of Marketing and Journal of Consumer Research). Johnson and Stewart provided a review of traditional approaches to the analysis of emotion in the context of consumer behavior. They reviewed appraisal theory and discussed examples of its application in the contexts of advertising, customer satisfaction, product design, and retail shopping. Holbrook explored and reviewed the concept of beauty as experienced by ordinary consumers in their everyday lives. His typology conceptualized everyday usage of the term “beauty” as falling into eight categories distinguished on the basis of three dichotomies: (i) extrinsically/ intrinsically motivated; (ii) thing(s)-/person(s)-based; and (iii) concrete/abstract. Xia and Monroe first reviewed the literature on consumer information search, and then the literature on browsing. They proposed an extended consumer information acquisition framework and outlined relevant substantive and methodological issues for future research. Hunt and Morgan reviewed the progress and prospects of the “resource-advantage” (R-A) theory. They examined in detail the theory’s foundational premises, showed how R-A theory provides a theoretical foundation for business and marketing strategy, and discussed the theory’s future prospects. Bharadwaj and Varadarajan provided an interdisciplinary review and perspective on the determinants of organizational performance. They examined the classical industrial organization school, the efficiency/revisionist ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ school, and the resource-based view of the firm. They proposed an integrative model of business performance that modeled firm-specific intangibles, industry structure, and competitive strategy variables as the major determinants of business performance. Vargo and Lusch focused attention ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ and proposed social judgment-involvement (SJI) theory as a potential theoretical framework to augment, replace, and/or elaborate the disconfirmation model and latitude models associated with ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ the methodological perspectives, issues, and applications related to correspondence analysis. They concluded with a list of the creative applications and the technique’s limitations.

REVIEW OF MARKETING RESEARCH

xi

ARTICLES IN THE SECOND VOLUME The second volume continued the emphasis of the first by featuring a broad range of topics contributed by some of the top scholars in the discipline. The diverse articles in the second volume can all be grouped under the broad umbrella of consumer action. Bagozzi developed a detailed framework for consumer action in terms of automaticity, purposiveness, and self-regulation. Mac฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ Lee, and Talukdar reviewed the literature related to use of the Internet as a vehicle for information search. They developed and empirically tested a general model of the choice of information sources with encouraging results. Miller, Malhotra, and King reviewed the categorization literature and developed a categorization-based model of the product evaluation formation process, which assists ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ an integrated framework that incorporated a more comprehensive set of various individual-level determinants of technology adoption and usage. Recently, marketing has come under increased pressure to justify its budgets and activities. Lehmann developed a metrics value chain to capture the various levels of measurement employed in this respect. Finally, Oakley, Iacobucci, and ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ARTICLES IN THE THIRD VOLUME Bolton and Tarasi described how companies can effectively cultivate customer relationships and develop customer portfolios that increase shareholder value. They reviewed the extensive literature on customer relationship management (CRM), customer asset management, and customer portfolio management, and summarized key findings. They examined five organizational processes necessary for effective CRM: making strategic choices that foster organizational learning; creating value for customers and the firm; managing sources of value; investing resources across functions, organizational units, and channels; and globally optimizing product and customer portfolios. Chandrasekaran and Tellis critically reviewed research on the diffusion of new products primarily in the marketing literature and also in economics and geography. While other reviews on this topic are available, their review differs from prior ones in two important aspects. First, the prior reviews focused on the S-curve of cumulative sales of a new product, mostly covering growth. Chandrasekaran and Tellis focused on phenomena other than the S-curve, such as takeoff and slowdown. Second, while the previous reviews focused mainly on the Bass model, Chandrasekaran and Tellis also considered other models of diffusion and drivers of new product diffusion. Eckhardt and Houston reviewed, compared, and contrasted cultural and cross-cultural psychological methods. They presented the underlying conceptions of culture that underpin both streams, and discussed various methods associated with each approach. They identified the consumer research questions best answered using each approach and discussed how each approach informs the other. Finally, they examined how consumer research can benefit from understanding the differences in the two approaches. While cultural and cross-cultural perspectives adopt distinct views about culture and psychological processes, it is possible to view them as complementary rather than incompatible. Several suggestions by Malhotra and his colleagues can be useful in ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ Malhotra and McCort 2001; Malhotra et al. 2005). For example, one can start with an etic approach and then make emic modifications to adapt to the local cultures. Alternatively, one can start with an emic perspective and then make etic adaptations to get an understanding across cultures. This systematic theory building and testing process is illustrated by Kim and Malhotra (2005).

xii

REVIEW OF MARKETING RESEARCH

Grewal and Compeau synthesized research from consumer behavior, psychology, and applied economics to address how price as an information cue affects consumers’ responses in the context of other information cues. They developed a conceptual framework, using adaptation-level theory and transaction utility theory, which synthesized prior research on price, reference price, and other information cues and their effects on consumers’ price expectations, evaluations, and behavioral intentions. Their conceptual model contributes to our understanding of the way imperfect information affects consumers’ decision processes, goes well beyond the original price– perceived quality paradigm, and integrates knowledge from consumer research, psychology, and applied economics. Sayman and Raju provided a review of research on store brands. Their review focused on integrating research in key areas and identifying directions for future research. There is limited theoretical and empirical research regarding optimal counterstrategies of national brands against store brands; studies tend to focus on one aspect, and national brand quality is typically assumed to be exogenous. Researchers have, by and large, focused on me-too-type store brands. Future research should consider premium store brand products as well. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ structure of a language, per se, influences the thoughts of those who speak it. They reviewed empirical research conducted over the past half-century on the effects of language structure on a variety of mental activities. They found support for the weak form of the linguistic relativity hypothesis, the notion that the structure of a language does indeed influence (but not determine) cognition. The estimation of independent and joint effects of language is difficult at best. We need comprehensive studies that incorporate the order in which bilinguals acquire their respective languages, how they acquire their languages, and when they acquire their languages. Future research should also compare the possible influence of a single language on mental processing across different cultures. Belk discussed the implications of getting visual for research, teaching, and communicating. He identified basic opportunities, threats, and consequences of becoming visual. Several techniques for collecting visual data were discussed in the realm of interviewing as well as observation. We might well be entering a Golden Age of visual and multimedia marketing research, and Belk helps us to get a good handle on it. ARTICLES IN THE FOURTH VOLUME Consistent with the first three volumes, this fourth volume also features a broad array of topics with contributions from some of the top scholars in the field. These articles fall under the broad umbrella of the consumer and the firm. Louviere and Meyer consider the literature on behavioral, economic, and statistical approaches to modeling consumer choice behavior. They focus on descriptive models of choice in evolving markets, where consumers are likely to have poorly developed preferences and be influenced by beliefs about future market changes. They call for a better alliance among behavioral, economic, and statistical approaches to modeling consumer choice behavior. Economic and statistical modelers can constructively learn from behavioral researchers, and vice versa. Folkes and Matta identify factors that influence how much an individual consumes on a single usage occasion by drawing on research in consumer behavior as well as allied disciplines. They develop an integrated framework to understand how, and at what stage, various factors affect usage ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ price and social norms influence consumption-related goals and their perceived desirability and

REVIEW OF MARKETING RESEARCH

xiii

feasibility. In the next phase, factors such as self-control strategies and product instructions influence the implementation of the goal. Finally, the consumer’s motivation to use feedback, and the type of feedback about consumption, has an influence on subsequent goal setting. Kumar and Luo also examine consumption, but from a modeling perspective. In order to allocate scarce marketing resources efficiently and effectively, it is important for a firm to know what to sell, when to sell, and to whom. Kumar and Luo review how the purchase timing, brand choice, and purchase quantity decisions have been modeled historically, as well as the issues within each decision that have been addressed. A vast majority of these studies use scanner data or transaction data. Since recent research has shown that common method variance may not be ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ data and should be increasingly employed. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ markets remain very limited. The fact that so few studies exist limits our understanding of effective ฀ ฀ ฀ ฀฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀฀ ฀ conceptual framework and several propositions regarding effective global brand extension strategy in a cross-cultural context. In doing so, they first review more commonly examined antecedent variables of (national) brand extension evaluation. Then, they propose a definition of culture and subsequently review the existing cross-cultural brand extension research. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ tracking research in marketing and evaluate its effectiveness. Specifically, they review eye-tracking applications in advertising (print, TV, and banner), health and nutrition warnings, branding, and choice and shelf search behaviors. Finally, they discuss findings, identify current gaps in our knowledge, and provide an outlook on future research. Singh and Saatcioglu review different approaches for examining role theory implications for boundary spanners such as salespeople, frontline, and customer contact employees. They focus on universalistic and contingency approaches and develop the configural approach by extending configurational theory principles to role theory. They compare and contrast different approaches and review literature that has remained less accessible to marketing researchers. John considers price contract design templates governing procurement and marketing of industrial equipment. He argues that price format choices precede the selection of a price level. These price formats are an integral aspect of the institutional arrangement devised to govern an exchange. John reviews institutions, that is, rules of interaction that govern the behavior of actors in dealing with other actors, with a focus on their pricing elements. ARTICLES IN THIS VOLUME In keeping with the earlier four volumes, this one also reflects an eclectic mixture of theory, measurement, data, and research methods, reinforcing the mission of Review of Marketing Research. The existence of two discrete, parallel, interactive cognitive systems underlying human judgment and reasoning has been postulated in several psychological and behavioral disciplines (Agarwal and Malhotra 2005; Malhotra 2005). One system is relatively unconscious, based on associations, and tends to be rapid. The other system is consciously guided, based on symbolic manipulation, and tends to be slower. The two systems generally operate in parallel, contributing interdependently to decision outcomes. Bond, Bettman, and Luce review recent developments in consumer behavior in terms of this dual-systems paradigm. They first examine a variety of frameworks that have been proposed, focusing on both their commonalities and their application domains. Then, they apply these frameworks to review selected topics

xiv

REVIEW OF MARKETING RESEARCH

from the recent marketing literature, including persuasion, metacognition, and immersive experiences. Most real-world consumer behavior situations involve complex combinations of experiential and analytical processing. Any phenomenon that is examined in terms of either system alone is unlikely to provide a complete picture and an adequate representation of underlying mental activity. Bond, Bettman, and Luce provide an insightful discussion of some of the emerging issues regarding the role of affect, the existence of multiple attitudes, and the notion of unconscious thought. The Chasm is a well-accepted paradigm among new products marketing practitioners that has taken root in the last decade. According to this paradigm, the market for new products is composed of “early” and “mainstream” markets with a “chasm” in between them. A fundamental premise of such an approach is that there is a communication break, at least to some degree, between the consumers in the early adopters and the mainstream market segment. Libai, Mahajan, and Muller examine empirical support for the existence of a communication break in the diffusion of innovations using aggregate product growth data, typically used in the diffusion of innovation research. They review three alternative models due to Bass, Rogers, and Moore. These models are flexible enough to include a partial or complete chasm between the early and mainstream markets. They analyze the sales data for three durable goods: color TVs, room air conditioners, and citizen’s band (CB) radios. Their results provide some support for the dual-market phenomenon and show the existence of a partial communication break in all three markets. The existence of the Chasm raises several important questions. Can we find ways to better predict it? Can we empirically identify the product factors that most differentiate between early adopters and the mainstream market? What are the factors that affect the depth of the communications break between products? As the authors point out, aggregate adoption data are not sufficient for answering these questions. More in-depth and disaggregate investigation across various time points should be conducted (Kim and Malhotra 2005). Rajagopalan and Bayus explore two of Eric Raymond’s key open source product development principles embodied in the bazaar community development model involving developers and users. They empirically examine the relationships between project community size (“eyeballs”) and development activity, and between development activity and product adoption. Their analysis supports the premise that “developer eyeballs” are positively related to development activity and that product development activity is significantly related to the speed of product adoption. Thus, they find support for some key principles of the open source bazaar. In particular, attracting a large developer base is important to further product development, and the early market evolution and acceptance of open source products is driven by product development activity. However, some of their results are contrary to the bazaar model. Specifically, they find that “user eyeballs” do not significantly contribute to increased development activity, and that product success does not in turn drive additional product development. Therefore, Raymond’s bazaar community development model involving developers and users should be revised to accommodate the more typical open source development project. Future research should explore the applicability of different new product diffusion models to open source innovations. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ brand coordinates are a linear function of product characteristics. Their method simultaneously identifies consumer segments, derives a joint space of brand coordinates and segment-level ideal points, and creates a link between specified product attributes and brand locations in the derived joint

REVIEW OF MARKETING RESEARCH

xv

space. This later feature permits a variety of policy simulations by brand(s), as well as subsequent ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ concerning consumers’ intentions to buy various competitive brands of portable telephones. The results of their proposed methodology are compared to a naïve sequential application of multidimensional unfolding, clustering, and correlation/regression analyses performed on the same data. The proposed clusterwise model obtained much better predictive validation than the traditional multistep clustering, multidimensional unfolding, and regression procedures. Generalizing the proposed methodology to the analysis of nonmetric and three-way data would extend the range of applications for this approach. Conjoint analysis is one of the most versatile methods in marketing research. Although this method has been popular in practice, one serious constraint has been dealing with the large numbers of attributes that are normally encountered in many conjoint analysis studies. Rao, Kartono, and Su review thirteen methods for handling a large number of attributes that have been applied in various contexts. They discuss the advantages and disadvantages of these methods. Based on ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ single study has systematically evaluated these potential alternative methods in the context of a specific applied problem. It would be worthwhile to conduct large-scale empirical and simulation studies to compare the methods. Laddering is a qualitative research technique that has great potential to uncover the factors underlying consumer decision making. However, this potential has not been realized because the time and costs of this qualitative technique as well as the lack of standard statistical measures to ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ research practices of both professional and academic researchers. They propose a set of quality metrics, and demonstrate the use of these measures to empirically compare the traditional face-toface interviewing method to an online one-on-one interviewing approach. As the online approach demonstrated, technology can ease the time and cost burden, while also yielding high-quality, reliable, and valid results. The Internet provides marketers with an expanded set of communications vehicles for reaching customers (Kim and Malhotra 2005; Malhotra, Kim, and Agarwal 2004). Two of the important and fast-growing elements of this new communications mix are online advertising and electronic word-of-mouth. While these vehicles are gaining in popularity, there are also challenges that need to be overcome. In particular, there is a need for marketers to understand how consumers respond and to develop new metrics for assessing performance. Bucklin, Rutz, and Trusov review recent research developments in marketing that are most relevant to assessing the impact of these communications vehicles. They first discuss the two major forms of Internet advertising, ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ sponsored links by Internet search engines.) Online communities, social networking sites, online referral programs, product reviews, and blogs all allow word-of-mouth (WOM) to spread faster and farther than in the past. Research has shown how electronic records of online word-of-mouth (e.g., product reviews) can be connected, via models, to performance outcome variables such as product ratings and sales levels. The chapter by Bucklin, Rutz, and Trusov is useful in highlighting key findings in the existing and emerging literature and in providing a framework for organizing future thinking and research It is hoped that collectively the chapters in this volume will substantially aid our efforts to understand, model, and make predictions about both the firm and the consumer and provide fertile areas for future research.

xvi

REVIEW OF MARKETING RESEARCH

REFERENCES Agarwal, James, and Naresh K. Malhotra. 2005. “Integration of Attitude and Affect: An Integrated Model of ฀ ฀ ฀ ฀Journal of Business Research฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀In The Psychology of Action,฀ ฀ ฀ ฀ Gollwitzer and John A. Bargh.฀ ฀ ฀ ฀ Kim, Sung, and Naresh K. Malhotra. 2005. “A Longitudinal Model of Continued IS Use: An Integrative ฀ ฀ ฀ ฀ ฀ ฀ ฀Management Science 51 (5) (May): 741–755. Malhotra, Naresh K. 2001. “Cross-Cultural Marketing Research in the Twenty-First Century.” International Marketing Review 18 (3): 230–234. Malhotra, Naresh K. 2005. “Attitude & Affect: New Frontiers of Research in the Twenty-First Century.” Journal of Business Research 58 (4) (April): 477–482. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ odological Issues and Guidelines.” International Marketing Review 13 (5): 7–43. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ national Marketing Research by Using Nonattribute Based Correspondence Analysis.” International Marketing Review฀ ฀ ฀ Malhotra, Naresh K., Betsy Charles, and Can Uslay. 2005. “Correspondence Analysis: Methodological ฀ ฀ ฀ ฀Review of Marketing Research ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀Information Systems Research฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ Management Science ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ Models: Theoretical Consideration and an Empirical Investigation.” International Marketing Review 18 ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ International Marketing Review฀ ฀ ฀

CONTENTS, VOLUME 1

XVII

CONTENTS, VOLUME 1 Review of Marketing Research Naresh K. Malhotra

ix

1. A Reappraisal of the Role of Emotion in Consumer Behavior: Traditional and Contemporary Approaches Allison R. Johnson and David W. Stewart

3

2. The Eye of the Beholder: Beauty as a Concept in Everyday Discourse and the Collective Photographic Essay Morris B. Holbrook

35

3. Consumer Information Acquisition: A Review and an Extension Lan Xia and Kent B. Monroe

101

4. The Resource-Advantage Theory of Competition: A Review Shelby D. Hunt and Robert M. Morgan

153

5. Toward an Integrated Model of Business Performance Sundar G. Bharadwaj and Rajan Varadarajan

207

6. Consumers’ Evaluative Reference Scales and Social Judgment Theory: A Review and Exploratory Study Stephen L. Vargo and Robert F. Lusch

245

7. Correspondence Analysis: Methodological Perspectives, Issues, and Applications Naresh K. Malhotra, Betsy Rush Charles, and Can Uslay

285

About the Editor and Contributors Index

317 319

xvii

xviii

CONTENTS, VOLUME 2

CONTENTS, VOLUME 2 Review of Marketing Research: Some Reflections Naresh K. Malhotra

ix

1. Consumer Action: Automaticity, Purposiveness, and Self-Regulation Richard P. Bagozzi

3

2. Looking Through the Crystal Ball: Affective Forecasting and Misforecasting in Consumer Behavior Deborah J. MacInnis, Vanessa M. Patrick, and C. Whan Park

43

3. Consumer Use of the Internet in Search for Automobiles: Literature Review, a Conceptual Framework, and an Empirical Investigation Brian T. Ratchford, Myung-Soo Lee, and Debabrata Talukdar

81

4. Categorization: A Review and an Empirical Investigation of the Evaluation Formation Process Gina L. Miller, Naresh K. Malhotra, and Tracey M. King

109

5. Individual-Level Determinants of Consumers’ Adoption and Usage of Technological Innovations: A Propositional Inventory Shun Yin Lam and A. Parasuraman

151

6. The Metrics Imperative: Making Marketing Matter Donald R. Lehmann

177

7. Multilevel, Hierarchical Linear Models, and Marketing: This Is Not Your Adviser’s OLS Model James L. Oakley, Dawn Iacobucci, and Adam Duhachek

203

About the Editor and Contributors Index

229 231

xviii

CONTENTS

XIX

CONTENTS, VOLUME 3 Review of Marketing Research: A Look Ahead Naresh K. Malhotra Contents, Volume 1 Contents, Volume 2

ix xv xvii

1. Managing Customer Relationships Ruth N. Bolton and Crina O. Tarasi

3

2. A Critical Review of Marketing Research on Diffusion of New Products Deepa Chandrasekaran and Gerard J. Tellis

39

3. On the Distinction Between Cultural and Cross-Cultural Psychological Approaches and Its Significance for Consumer Psychology Giana M. Eckhardt and Michael J. Houston

81

4. Consumer Responses to Price and Its Contextual Information Cues: A Synthesis of Past Research, a Conceptual Framework, and Avenues for Further Research Dhruv Grewal and Larry D. Compeau

109

5. Store Brands: From Back to the Future Serdar Sayman and Jagmohan S. Raju

132

6. Language, Thought, and Consumer Research Dwight R. Merunka and Robert A. Peterson

152

7. You Ought to Be in Pictures: Envisioning Marketing Research Russell W. Belk

193

About the Editor and Contributors

207

xix

CONTENTS, VOLUME 4 Review of Marketing Research: Taking Stock Naresh K. Malhotra

ix

Contents, Volume 1 Contents, Volume 2 Contents, Volume 3

xv xvii xix

1. Formal Choice Models of Informal Choices: What Choice Modeling Research Can (and Can’t) Learn from Behavioral Theory Jordan J. Louviere and Robert J. Meyer

3

2. How Much to Use? An Action-Goal Approach to Understanding Factors Influencing Consumption Quantity Valerie S. Folkes and Shashi Matta

33

3. Integrating Purchase Timing, Choice, and Quantity Decisions Models: A Review of Model Specifications, Estimations, and Applications V. Kumar and Anita Man Luo

63

4. Brand Extension Research: A Cross-Cultural Perspective Michael A. Merz, Dana L. Alden, Wayne D. Hoyer, and Kalpesh Kaushik Desai

92

5. A Review of Eye-Tracking Research in Marketing Michel Wedel and Rik Pieters

123

6. Role Theory Approaches for Effectiveness of Marketing-Oriented Boundary Spanners: Comparative Review, Configural Extension, and Potential Contributions Jagdip Singh and Argun Saatcioglu

148

7. Designing Price Contracts for Procurement and Marketing of Industrial Equipment George John

183

About the Editor and Contributors

201

xx

Review of

Marketing Research

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

3

CHAPTER 1

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE Recent Evidence and Emerging Issues SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE Abstract Researchers across a variety of psychological disciplines have postulated the existence of two functional systems underlying human judgment and reasoning. One system is rapid, relatively unconscious, and based on associations; the other is slower, consciously guided, and based on symbolic manipulation. According to most conceptualizations, the two systems operate in parallel, contributing interdependently to decision outcomes. This chapter examines recent developments in consumer behavior in terms of the dual-systems paradigm. We first review a variety of proposed frameworks, focusing on both their commonalities and their domains of application. Next, we apply these frameworks to review selected topics from the recent marketing literature. Research on persuasion, metacognition, and immersive experiences is examined through the lens of experiential and analytical processing pathways. We close with a discussion of emerging questions regarding the role of affect, the existence of multiple attitudes, and the notion of unconscious thought. Truly successful decision making relies on a balance between deliberate and instinctive thinking. —Malcolm Gladwell Introduction Consider the following scenario: In the checkout line at the grocery store, you are contemplating various new brands of chewing gum. Based on price, packaging, and so forth, you have trouble really distinguishing among the brands. However, you notice during this cursory scanning process that you keep coming back to one brand in particular. Objectively, there is nothing superior about the brand, but it just seems to “feel right,” so you make the purchase based on this impression. The following day, you are talking to a good friend at work as she reaches into her purse and pulls out the exact same brand of chewing gum! Being a marketing scholar, you realize that, somehow, your friend undoubtedly played a role in your “intuition” concerning the brand. Or consider the following: During a shopping trip through the mall, you walk past a cookie and candy vendor. Faced with the vast array of sweets on display and the chocolate aroma, you make an impulse decision to stop for a moment and buy something for yourself. However, waiting in line, you begin to think about the implications of this decision for your upcoming dinner, the 3

4

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

possibility that the cookies have been sitting out all day, and the number of calories represented by even the smallest of your options. Realizing that you would not have been interested in the treats if you had not walked right past them, you decide that giving in to temptation does not “make sense,” leave the line, and continue with your shopping trip. Despite representing very different decision contexts, these scenarios share certain features that come up repeatedly in everyday consumer decisions. One way of understanding behavior in these contexts is to consider the effects of thoughts, beliefs, and ideas that “feel right” and those that “make sense,” on cognitive processes underlying choice. This topic has been gathering increased attention in recent years by investigators in the fields of social, cognitive, and consumer psychology. As a consequence, a variety of conceptual models have been developed to capture the constructs of intuition and analysis as distinct processes conducted by separate underlying cognitive systems. Although still in their infancy, these dual-systems models present compelling explanations for consumer processing in situations like the examples above, and they offer promising opportunities for research into real-world consumer phenomena. This chapter provides an introduction to the dual-systems concept and its application to consumer research. We first review some of the more prominent models that make a distinction between instinctual and analytical systems of reasoning. Having established this conceptual background, we then use the lens of a dual-systems framework to examine a variety of emerging topics in consumer behavior. The chapter concludes by addressing specific theoretical issues surrounding the intuition-reason distinction that are directly relevant to the study of marketing. Intuition and Analysis as Distinct Systems of Thought The notion that individuals possess two distinct modes of cognition is widespread. However, current understanding regarding the characteristics of these two modes remains limited, and the diverse theoretical frameworks that have been presented are not without disagreement. This section provides an overview of various approaches used to capture the functional mechanisms commonly known as system-1 and system-2 (Stanovich and West 1998). We discuss a number of popular conceptual models, highlighting their distinct areas of focus and points of agreement. Historical Background: Domain-Specific Dual-Processing Accounts Over the decades, psychologists from a wide variety of disciplines have utilized dual-process accounts to explain the phenomena they study. Most commonly, researchers have postulated that the effects of a particular variable on judgment are mediated by the cognitive involvement, ability, or effort of the individual. As a result, conceptual models were developed that involved two discrete levels or stages of rumination, such that individuals operating at a subordinate level produce responses that are distinct from (and usually inferior to) those operating at a higher level. In general, these models are characterized by specificity to the domains in which they were introduced and the tendency to classify constituent processes in a hierarchical manner. Probably the best known of all dual-process models are the heuristic-systematic model (HSM— Chaiken 1980) and elaboration likelihood model (ELM—Petty and Cacioppo 1981), which were introduced almost concurrently to explain the effects of persuasive communication on attitude change. The ELM proposes that a message can be processed via either a “central” route (based on strength of arguments) or a “peripheral” route (based on aspects of the source, the situation, etc.); the particular route that is taken depends on the elaboration likelihood of recipients, which is a function of their ability and motivation to process the message. The HSM bears strong resemblance

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

5

to the ELM in most respects, but the former places more emphasis on particular goals pursued by the individual in responding to the message, such as forming a valid attitude or defending current beliefs. Depending on the particular goals that are active, individuals may process by either a default, heuristic route following well-learned decision rules (e.g., “follow the majority”) or by a more effortful systematic route, carefully scrutinizing the elements of the message.1 The basic principles underlying the ELM and HSM have been supported by hundreds of studies since their introduction (Chaiken and Trope 1999). In other areas of psychological inquiry, numerous dual-process models have been proposed that share features similar to those above; here we highlight just a few. In the domain of person perception, Brewer (1988) and Fiske (Fiske and Nueberg 1988) offer similar models to represent social perception and categorization. Both models propose that upon receiving information about a target person, individuals process that information in one of two ways. One of these is a cursory screening process in which the target is categorized by age, gender, and so forth, and that category is used as a basis for their impression. However, provided that perceivers are motivated and able to do so, they may instead process more elaborately by attending to individuating information. In a sequential, two-stage model presented to explain stereotyping behavior, Devine (1989) suggests that merely encountering a stereotyped group member leads to the activation of strongly formed associations regarding that group; however, individuals low in prejudice may effortfully suppress these associations at a second stage by considering more favorable group associations and beliefs. The notion of stages is also integral to the attributional inference model developed by Gilbert to explain inference anomalies such as the fundamental attribution error (e.g., Gilbert, Pelham, and Krull 1988). According to the model, perceivers who are unmotivated or cognitively constrained will tend to infer that a target individual’s behavior is a result of stable dispositional tendencies. However, perceivers with sufficient ability and motivation enter a second stage in which they consider aspects of the situation that may be responsible for the target’s behavior. These sequential frameworks share the notion of a correction process whereby deeper processing may override more cursory first impressions. Hence their approach is distinct from the ELM, HSM, and related frameworks emphasizing the selection of a particular style of thought. In sum, the models reviewed so far represent an entire class of theories introduced in the 1980s and 1990s that explain the existence of a particular psychological phenomenon by the existence of two distinct pathways to judgment. Notably, these theories almost always present their proposed pathways as two mutually exclusive or sequential processing alternatives—a generally inferior method based on associations and a generally superior method based on elaboration. Dual-Systems Approaches: Instinct and Reason as Parallel Pathways Although each of the dual-process models above represents a valuable framework for the investigation of a particular area of psychological inquiry, a clear need has existed for more integrative theories that could tie together core concepts underlying these approaches into broader models of human cognition. Responding to this need, a number of more general dual-process models have recently been proposed (e.g., Epstein 1991; Kahneman 2003; Sloman 1996; Smith and DeCoster 2000); we describe these as dual-systems approaches. Generally speaking, these more integrative models postulate the existence of two distinct but parallel systems of human judgment. One system (system-1) is relatively automatic, rapid, affect-laden, and based on the recognition of patterns or associations; outputs of system-1 may be achieved without any understanding of the underlying process (as in the chewing gum example, where a particular option “just felt right”). Importantly, the rapidity of system-1 processing does not preclude its ability to follow higher-order rules

6

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

and contingencies, once these rules and contingencies are detected (Hogarth 2001; Kunda and Thagard 1996). The other system (system-2) is consciously directed, effortful, slow, and based on the application of rules or computation. Outputs of system-2 take the form of conclusions arrived at intentionally and with awareness of the underlying process (as in the cookie example, where yielding to temptation “just didn’t make sense”). We review some prominent dual-systems models below and refer the reader to Table 1.1 for a summary. General Models The most longstanding of the dual-systems models is the cognitive-experiential self-theory of Seymour Epstein (1991, 2003). According to his framework, individuals possess an intuitiveexperiential system-1 that is holistic, nonverbal, and based on images, feelings, and sensations; intuition is considered a function of the experiential system. In addition, individuals possess an analytical-rational system-2 that is logical, abstract, and based on the application of symbols, words, and numbers. In the development of the theory, Epstein and colleagues have focused on stable individual differences in the utilization of each system, resulting in the development of the rational-experiential inventory (REI) as a measure of trait experiential and analytical processing (Epstein et al. 1996; Pacini and Epstein 1999). An important contribution of the model is that the experiential and analytical systems are not mutually exclusive but rather operate in parallel; in fact, Epstein et al. (1996) present evidence that REI analytical and experiential scores are modestly correlated. In research based on the theory, a considerable amount of attention has been paid to perceptual illusions and judgment heuristics as manifestations of experiential processing (Epstein et al. 1992; Kirkpatrick and Epstein 1992). Importantly, however, the model also stresses the beneficial role of both systems. For example, individual differences in experiential and analytical processing appear to contribute independently to various measures of coping ability, achievement, and interpersonal relationships (Epstein et al. 1996). Applying these ideas to consumer and marketer activities, Novak and Hoffman (2007) argue that task performance will be enhanced by utilization of the appropriate processing mode (e.g., creative tasks benefit from experiential processing, while computational tasks benefit from analytical processing). Their research demonstrates a variety of advantages accruing to individuals actively engaging in the mode best suited to the task, such as enhanced performance, better mood, and subjective ease. In a highly influential paper, Smith and DeCoster (2000) utilize connectionist theories of memory and cognition as a starting point for the development of a dual-systems theory. They review compelling psychological and neurological evidence for the existence of two underlying memory systems: (1) a schematic memory that slowly updates according to the occurrence of general regularities among sensory, motor, and perceptual processes, and (2) a fast-learning memory that focuses on deviations from expectation, quickly creating new representations of an experience based on its context (McClelland, McNaughton, and O’Reilly 1995). These underlying memory systems form the structural basis for the associative and rule-based processing modes that are central to the model. Associative processing is based on the operation of schematic memory; that is, salient cues in the environment automatically elicit retrieval of relevant information. The associative mode thus serves as a rapid form pattern completion mechanism that can function without awareness (e.g., the automatic retrieval of gender stereotypes upon encountering an individual). In contrast, rule-based processing involves consciously directed manipulation of information according to symbolic rules contained in either the schematic or the fast-learning memory system (e.g., the use of individuating characteristics to infer unique traits of the same individual). Because rule-based

Consumer Focus

Decision/ Reasoning Focus

Broad/General Focus

Affective

Impulsive

Strack, Werth, Deutsch (2006)

Intuition

Kahneman (2003)

Shiv and Fedorikhin (1999)

Associative

Associative

Smith and DeCoster (2000)

Sloman (1996)

Experiential

System-1 Label

Epstein (1991)

Model (Representative Cite)

Summary of Selected Dual-Systems Models

Table 1.1

Reflective

Cognitive

Reasoning

Rule-based

Rule-based

Rational

System-2 Label

Primarily concerns self-control dilemmas; stresses joint operation and interaction of both systems

Primarily concerns self-control dilemmas; sequential; immediate affect suppressed by cognitive deliberation

Sequential model—all judgments go through both systems; system-2 as monitor of frequently errorful system-1 processing

Focuses on reasoning and inference tasks, cultural basis for system-2, cases where systems reach contradictory conclusions

Connectionist model linking thought to underlying memory systems; stresses intelligence of associative system

Situated within broader theory of personality; emphasizes individual differences; system-1 and system-2 positively correlated

Notes

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE 7

8

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

processing proceeds according to explicit, sequential representations based on specific conceptual principles rather than overall similarity, this mode is necessarily slow, effortful, and analytic. Like many other theorists, Smith and DeCoster stress that the discrete processing modes they propose are not inhibitory but rather operate in tandem. Although the speed of the associative system allows its conclusions to form quickly, this relative efficiency does not imply that the system stands idle as rule-based processing slowly progresses. Instead, the application of rules by the latter system (“When product knowledge is limited, seek expert advice”) can lead to new associations (“Jane and Bill are experts”) and inferences (“I should ask Jane and Bill about this product”). A similar overlap can be observed in the memory systems underlying the model. In one case, a conclusion that follows from the application of rule-based processes (e.g., the spelling of a word) is repeated so often that it becomes a part of schematic memory, retrievable without any effort by associative processing. In the converse case, an individual reflects upon repeated experiences built over time (e.g., repeated observation of Brand X releasing new products) to generate new rules and knowledge that can be flexibly used as the need arises (“Brand X is innovative”). Decision/Reasoning Models An influential model proposed by Sloman (1996) shares many similarities with the Smith and DeCoster (2000) framework and even uses the same labels (associative and rule-based) to identify its processing modes. However, Sloman focuses less on the underlying memory structures responsible for the operation of the two modes and instead concentrates on the use of each mode in inference and reasoning tasks. In this model, the associative and rule-based modes are conceptualized as alternative forms of computation. The former contains an intuitive processor that performs calculations by processing environmental regularities built up by repeated experience; thus the defining operations for the associative system are those based on similarity and contiguity. Akin to some other theorists, Sloman argues that representative functions of the associative system include not only visual recognition and intuition, but also broader cognitive operations such as fantasy, creativity, and imagination. In contrast, the rule-based system contains a rule interpreter that allows for the manipulation of abstract concepts and symbols via principles of logic, causation, and so on. An important aspect of Sloman’s model is that these principles are acquired from the sociocultural milieu of the individual and, by virtue of their abstraction, are applicable to a wide variety of reasoning tasks: explanation, formal analysis, determination of causation etc. Like the Epstein and Smith/Decoster approaches, Sloman’s model stresses the simultaneous operation of the associative and rule-based systems: operating jointly but independently, the systems reach a steady state by a process of parallel constraint satisfaction. However, this resolution is obstructed when the two systems reach incompatible conclusions, a situation of simultaneous contradictory belief in which “both systems try, at least some of the time, to generate a response” (1996, p. 15). According to the model, most classic examples of decision heuristics (the “Linda” problem, base-rate neglect, etc.) are in fact instantiations of this contradiction between rules and associations. Notably, Sloman proposes that because of its origin in social learning, the rule-based system tends to be more subjectively compelling as a basis for judgment. Therefore, logical analysis tends to suppress instinctive responses, but in some cases the latter continue to “intrude” on the process. The model proposes that certain task features may influence the outcome of this conflict, and we return to this issue later. In an integration of multiple contributions to the study of decision making, Kahneman (2003) has advanced a model whose centerpiece is the notion of accessibility. Drawing largely from research involving judgment under uncertainty (Stanovich and West 2000), Kahneman argues for

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

9

a system-1 intuition that is fast, effortless, associative, emotional, and governed by habit, as well as a system-2 reasoning that is slow, effortful, flexible, and dependent on conscious monitoring and control. According to the model, system-1 operates jointly with perceptual processes during an evaluation task to present implicit impressions of the particular stimuli, not actual judgments, which necessarily involve system-2. A distinct feature of this framework is its emphasis on the dual roles played by accessibility in judgment tasks. First, the particular intuitions that come to mind during a task are a direct function of the accessibility of various mental representations evoked by that task. Second, the extent to which intuitions are corrected by reasoning is a direct function of the accessibility of various rules that might be used to do so. This formulation is more sequential than parallel; indeed, an important element of Kahneman’s formulation is the notion of system-2 as a monitoring mechanism. Furthermore, although the value of intuitive expertise in certain circumstances is acknowledged, Kahneman, much like Sloman (1996), focuses on heuristics and biases as cases in which flawed intuitive judgments go uncorrected by the monitoring process. When based on highly accessible associations, intuitive impressions can feel extremely compelling (Kahneman and Frederick 2002; Simmons and Nelson 2006) and therefore pass through the reasoning system unmodified. Consumer Models Recently, the field of consumer research has itself fostered the development of theories making a dual-systems distinction. In particular, a number of theories have been developed to explain processes underlying self-control and impulsive behavior. The most prominent of these theories is the affective-cognitive framework of Shiv and Fedorikhin (1999, 2002). Their model presents self-control dilemmas as the result of a conflict between spontaneously evoked, intense affective reactions to a desired stimulus and more controlled, deliberative, cognitive reactions that only occur when task conditions are amenable to extensive processing. In the well-known “cake versus fruit salad” task (Shiv and Fedorikhin 1999), participants given a choice between these two options were much more likely to choose the cake when placed under cognitive load, and these constrained participants tended to attribute their choice to reliance on the “heart” versus the “head.” Although the affective-cognitive model was originally presented in terms of sequential stages, later modifications relaxed this restriction somewhat and acknowledged the interactive nature of the two systems (Shiv and Fedorikhin 2002). More recently, a different theory of consumer impulsivity has been advanced by Strack, Werth, and Deutsch (2006). In their reflective-impulsive model (RIM), the impulsive system represents the encoding of environmental regularities as patterns of association in an associative network. Processing by the impulsive system leads to three kinds of feelings: physical sensation, positive or negative affect, and “cognitive feelings” such as familiarity. The reflective system complements the impulsive system by performing various representational and self-regulatory functions. Consistent with other models, the RIM suggests that the reflective system operates according to the manipulation of symbolic representations, and that the flexibility of this system comes at the cost of inefficiency and instability of representations. Whereas perceptual input is by itself the primary basis for the impulsive system, the behavioral implications of the reflective system result from a process of reasoning about the desirability and feasibility of an action. Compared to the other models cited, the RIM is notable for its emphasis on interaction between the two processing mechanisms. The impulsive system provides content for the reflective system in multiple ways: for example, the accessibility of concepts and schemata, the production of affective and cognitive feelings, and observation of one’s own behaviors. Conversely, reflective

10

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

processing may influence the impulsive system passively (thinking of a behavior may activate related schemata), actively (through the conscious “intention” to do something), or through the regulation of perceptual inputs. Even in situations of self-control failure, both systems are assumed to contribute: for example, in order to make an “impulsive” in-store purchase, a buyer must still locate the checkout line, perform a monetary transaction, and so on, and these are hardly “impulsive” actions. For these situations, Strack, Werth, and Deutsch (2006) present various factors that moderate the relative influence of the reflective system (decision importance, preexisting habits, the need to justify choices, etc.). Summary The models reviewed above represent only a sample of the various integrative dual-systems approaches; nonetheless, they possess several theoretical distinctions. As noted earlier, the models differ markedly in the extent to which system-1 and system-2 are presumed to influence one another functionally and structurally, the degree to which either system is perceived as more subjectively compelling, and the extent to which system-2 acts as a corrective mechanism for errorful system-1 responses. Later sections of this chapter will address other important areas of disagreement, notably the role of each system in the production and utilization of affect (Epstein 1991; Shiv and Fedorikhin 2002; c.f. Strack, Werth, and Deutsch 2006). More important for present purposes, however, are the basic principles that are mostly shared by these and other models postulating dual processing systems. There exists general agreement that system-1 requires little or no effort to operate, is extremely rapid and efficient, is based on the accessibility of well-formed declarative or procedural memories, and operates according to associative principles that may not be accessible to conscious awareness. Thus, outputs of this system may be described as “feeling right,” even though individuals have no introspective access to the system’s underlying operation. Furthermore, there exists general agreement that system-2 is comparatively slow, proceeds according to logical principles or symbolic manipulation, is subject to conscious awareness, and is under the control of the individual. Because its conclusions can be introspectively verified and justified in terms of reasons, outputs of this system may be described as “making sense.” (As we discuss later, the distinction between “feeling right” and “making sense” may play a pivotal role in subjective reliance on either system.) Keeping in mind these similarities across various conceptualizations of the system-1/system-2 distinction, we next turn to the issue of factors affecting differential engagement of the two systems. Determinants of System Engagement and Utilization Although all the frameworks reviewed above acknowledge that system-1 and system-2 are continuously active, the extent to which each system is involved in a particular processing task will vary dramatically across situations. Below, we review a diverse literature concerning various factors that may influence the extent to which activity of either system impacts judgment and behavior. We begin with classical approaches based on motivation and ability and then review recent consumer behavior findings that relate to this issue. Finally, we discuss some of our own work on the manner in which conflict among the systems is resolved. As noted above, each of the different theoretical approaches to the dual-systems distinction utilizes a distinct set of labels to identify each system. For the sake of clarity, we adopt terminology used by Epstein (1991) and refer to system-1 and system-2 as experiential and analytical, respectively.

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

11

“Triggers” of Experiential and Analytical Processing Much of the empirical research taking a dual-systems perspective has been characterized by two general approaches. One approach has examined processing style as an individual difference reflecting stable underlying characteristics, and the other approach has examined the extent to which processing style is moderated by task-specific motivation and ability. Individual Differences In typical investigations using the stable-characteristics approach, individuals are presented tasks for which experiential and analytical processing modes favor distinct responses, and their behavior is correlated with scales such as Need for Cognition (Cacioppo and Petty 1982), Faith in Intuition (Epstein et al. 1996), or the Rational-Experiential Inventory (Pacini and Epstein 1999). For example, relevant work has documented that individuals high in need for cognition are more likely to be persuaded by central than peripheral cues (Cacioppo, Petty, and Morris 1983), to correct for biases when forewarned (Martin, Seta, and Crelia 1990), and to resist conformity (Areni, Ferrell, and Wilcox 2000). Individuals high in Faith in Intuition succumb more readily to framing effects (Levin et al. 2002) and respond more heuristically across a variety of decision tasks (Epstein et al. 1996). Similar research has investigated the extent to which experiential and rational processing tendencies are associated with other individual difference measures. For example, Stanovich and West (1998) demonstrate that a “rational thinking style” (open-mindedness, consideration of counterfactuals, etc.) is correlated with performance on the Raven Progressive Matrices (Raven 1962), Scholastic Aptitude Test scores, and reading comprehension. High scores on the rationality subscale of the REI are associated with conscientiousness, ego strength, and belief in a meaningful world, while high scores on the experiential subscale are associated with extraversion, trust in others, and emotional expressivity (Pacini and Epstein 1999). A number of recent applications have explored individual differences of particular relevance to consumer research. For example, research indicates that veteran users of a brand, despite their enhanced factual knowledge, will engage more heavily in gestalt, experiential processing during brand-based decisions. Among other results, Jewell and Unnava (2004) find brand recall to be more correlated with brand “feelings” for heavy users but more correlated with brand “thoughts” for light users. As a consequence, advertisements based on feelings were most effective at improving recall among heavy users, while advertisements based on attributes were most effective for light users. Other investigations have focused on cultural and personality differences. For example, Aggarwal and Law (2005) demonstrate a substantial effect of relationship norms on the processing style used by consumers in forming brand evaluations. Across multiple experiments, communal relationship norms led participants to process brand information more abstractly and holistically, while exchange relationship norms led to processing that was less abstract. Task-Specific Ability and Motivation A second approach is to view engagement of either system as the result of variables specific to the processing task. Most often, a particular decision-making bias is assumed to result from flawed experiential processing, and this claim is buttressed by showing elevation of the bias when either ability or motivation to process is constrained. Concerning ability, empirical investigations have linked processing constraints to the fundamental attribution error (Gilbert and Malone 1995; Nisbett and Ross 1980), persuasion by heuristic versus systematic message cues (Eagly and Chaiken

12

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

1993; Petty, Cacioppo, and Schumann 1983), and stereotyping (Gilbert and Hixon 1991). In the consumer behavior literature, cognitive resource limitations are associated with diminished selfcontrol (Shiv and Fedorikhin 1999), susceptibility to perceptual illusions (Raghubir and Krishna 1996), and numerous context effects (Dhar, Nowlis, and Sherman 2000). Concerning the motivation variable, studies of forewarning and correction show that individuals can be remarkably effective in reducing the impact of extraneous information (e.g., an attractive endorser) on their judgment, so long as they possess an appropriate “naïve theory” regarding the undue influence (e.g., Schwarz and Bless 1992; Wegener and Petty 1995). In addition, high involvement with an advertisement is linked to greater message elaboration (Petty, Cacioppo, and Schumann 1983) and reduced susceptibility to irrelevant or deceptive information (Johar 1995; Park and Young 1986). Consumer research using an integrated ability/motivation framework has tended to conceptualize processing style along a single continuum. For example, Kardes et al. (2004) suggest that because most consumers have formed strong associations between price and quality (Broniarczyk and Alba 1994), price-based inferences should be particularly compelling to the extent that individuals are unmotivated or unable to engage in more rigorous analytical processing. In their experiments, valuations revealed a stronger price-quality inference when participants’ need for closure was enhanced by task instructions, but only when the product information under consideration was complex and unstructured. Some recent applications in consumer research have incorporated the notion of simultaneous operation of underlying systems. For example, Meyers-Levy and Maheswaran (2004) use the heuristic-systematic model (Eagly and Chaiken 1993) as a backdrop to investigate the role of positive and negative framing effects in the evaluation of advocacy messages. Replicating past results (Maheswaran and Meyers-Levy 1990), they show that when processing motivation is low (i.e., messages of low personal relevance with few risky implications), positively framed messages are more effective (presumably because the heuristic cue of message valence is utilized), but when motivation is high, negatively framed messages are more effective (because negative information is considered diagnostic). Most interestingly, at moderate levels of motivation (few risky implications but high personal relevance), framing effects were nearly eliminated. The authors suggest that in this case, both heuristic and systematic processing are engaged, to approximately equal degrees. The Resolution of System Conflict Many important topics in consumer research can be understood as cases where the experiential and analytical systems generate opposing outputs. Self-control dilemmas, perceptual illusions, choice conflict, and so forth present consumers with situations where contradictory conclusions are derived from “gut instincts” and “thoughtful analysis.” Therefore, an understanding of the extent to which consumers rely on the outputs of either system is fundamental to predicting behavior in these situations. Our own recent research has focused on individuals’ lay theories regarding the use of intuition versus analysis (Bond 2008; Bond, Bettman, and Luce 2008). We see prior approaches as focused primarily on system output generation, defined as the extent to which experiential and analytical processing are engaged, either chronically or during a particular decision task. Instead, we focus on output selection, defined as the resolution of conflict pitting outputs of each system against one another. We argue that an assessment of processing mode suitability (“How should I think about it?”) plays a pivotal role in the resolution process. Our model assumes that an individual facing a novel decision is relatively free of cognitive constraints and sufficiently motivated to engage in effortful processing (i.e., experiential and analytical

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

13

processing modes are both engaged). For example, an individual who is favorably predisposed toward one of two brands in a consideration set may be asked to read objective information about the brands and choose between them. We focus on situations in which the output of experiential processing is distinct from that of analytical processing, so that the two processing modes imply different choices; for example, the objective information may reveal that the favorably conditioned brand is inferior. In keeping with the notion of experiential and analytical processing as distinct, parallel cognitive pathways, we propose that this scenario evokes competing processing outputs that are mutually accessible to the decision maker, and we use the term simultaneous conflicting preferences to capture the tension inherent in these situations. Our model is predicated on the precept that higher-order cognition (“thinking about thinking”) moderates the influence of retrieved content on a decision (Schwarz 2004). We suggest that in most situations, individuals possess a naïve theory regarding the extent to which each processing route is an appropriate basis for deciding. Therefore, to the extent that individuals believe a particular cognition to be the result of experiential versus analytical processing, their theory about the suitability of that processing mode for the present decision will impact the weight assigned to the cognition. We use the term perceived suitability to capture this notion of lay theories regarding the appropriateness of experiential and analytical processing. Our primary hypothesis is that certain characteristics of the decision setting will systematically influence perceived suitability and, as a result, the judgment process (see also Alter et al. 2007). As an illustration of these principles, consider the following exploratory study (Bond 2008), in which we presented a student sample with a broad set of common consumer decisions (see Table 1.2 for sample items). Our goal was twofold: (1) to obtain a direct assessment of lay theories regarding the perceived suitability of either processing system across different situations, and (2) to identify variables that influence this perceived suitability. Therefore, for each decision, we measured participants’ beliefs about the appropriateness of experiential versus analytical processing as a basis for deciding, and we also measured perceptions of the decision on a variety of underlying dimensions. We focus first on lay theories of perceived suitability for each decision. We asked participants to indicate the output they would select if different conclusions were reached by “following your gut” and “using your head.” Analyses revealed that, as expected, certain decision items tended to be judged as more experientially suited and others as more analytically suited (average ratings are given in Table 1.2). We next examined whether the variance in perceived suitability could be attributed to decision characteristics. Indeed, analyses revealed that ratings of decision dimensions were systematically and sizably correlated with the forced tradeoff measure. When a decision presented conflicting experiential and analytical outputs, participants were more likely to “go with their gut” to the extent that the decision was short-term, hedonic, concrete, unimportant, emotionally positive, and gain-focused. If lay theories of perceived suitability appear to vary with certain decision characteristics, an obvious question is whether these differences affect choices. In a controlled follow-up experiment, we focused on one of the dimensions found to be most impactful in the initial survey—hedonic/ functional. Participants were given a conditioning task in which two hypothetical brand names were conditioned either positively or negatively via their association with pleasant or unpleasant images. Next, participants were given verbal information about the two brands that indicated one was objectively superior. Importantly, the product category represented by the two brands was manipulated to be either hedonic (chocolate bar) or functional (alkaline batteries). We predicted that for hedonic categories, participants would perceive that instinctual processing is a reasonable

14

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

Table 1.2

Appropriateness of Experiential Versus Analytical Processing Across Decision Scenarios

Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding Deciding

which friends to spend time with whether to go to an upcoming concert on a birthday gift for a close friend what clothes to wear to a party which brand of cereal to buy where to go on an outdoor day trip whether to return an item of clothing to the store which days to visit an amusement park whether to help a friend with money problems what major to pursue which digital camera to buy what kind of career to pursue where to buy gasoline which credit card to use for an expensive purchase whether to invest money in the stock market

Mean (higher = “follow my gut”)

Std. deviation

6.54 6.29 6.14 6.11 5.93 5.46 5.18 4.89 4.39 3.93 3.82 3.29 3.11 2.46 2.32

1.6 1.4 2.2 1.8 1.8 1.8 2.0 2.1 2.3 2.4 2.1 2.6 1.7 1.5 1.8

Source: Sample items from Bond, Bettman, and Luce 2008.

basis for deciding, and therefore base their evaluations primarily on their conditioned associations with each brand; however, for functional categories, participants would largely neglect their prior associations, focusing instead on deliberative processing of brand information. Results for both brand evaluation and choice measures supported these predictions, thus providing experimental evidence that the hedonic/functional nature of a product category systematically influences evaluations by varying the willingness of consumers to “follow their gut” instead of ignoring (or correcting for) the influence of instinctual responses. Looking forward, it will be fruitful to apply this methodology to other dimensions of common consumer decisions. Application to Topics in Consumer Behavior In the following section, the experiential/analytical distinction is utilized to examine recent developments in selected areas of consumer behavior. We have chosen to focus on domains where activity has been abundant and the distinction seems particularly relevant. We first review recent advances in the study of persuasion, then turn to the emerging topic of consumer metacognition, and finally discuss a growing body of research on “immersive” experiences. Persuasion and Related Topics One of the most active research streams applying dual-systems models to consumer phenomena is that dealing with persuasion, trust, and consumer defensiveness. Here, we will discuss applica-

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

15

tions falling under the broad topic of source effects and then relate these ideas to one of the more prominent theories in this stream, the persuasion knowledge model (Friestad and Wright 1994). Source Effects Much of the literature regarding source effects in persuasion has been predicated on concepts advanced by the elaboration-likelihood model (Petty and Cacioppo 1981) and heuristic-systematic model (Chaiken 1980). For example, researchers have long assumed that trust in a message source operates as a cue during heuristic processing (Petty, Cacioppo, and Schumann 1983). Later evidence indicated that in addition, lack of trust can serve a motivational purpose, enhancing message scrutiny and differentiation of strong and weak arguments (Priester and Petty 1995). Within marketing, an immense stream of research on advertising, communications, and promotion has addressed the topic of source effects on attitude change, but various studies have obtained seemingly disparate findings (Wilson and Sherrell 1993). In a recent attempt to integrate these findings, Kang and Herr (2006) provide a model linking source characteristics to various downstream consequences. Their model, rooted in the HSM and ELM, suggests that source effects may transpire via three distinct routes. When cognitive resources are low, attitudes will be impacted directly by source characteristics (e.g., an attractive endorser). At a higher level of resources, some analytical processing is initiated and transfer will only occur if the relevant source characteristics are applicable to the product category (e.g., a beauty product). Most interesting is the third case, where resources are “excessively high,” and thus sensitivity to source biases is particularly salient. In this case, a positive source can actually inhibit persuasion as consumers overcorrect for what they perceive as a pernicious influence (Petty and Wegener 1999). In terms of the dual-systems distinction, the third case described by Kang and Herr (2006) illustrates the means by which defensive processing can be driven by both analytical cognition (motivated deliberative processing) and experiential cognition (enhanced sensitivity to influences seen as persuasive). The latter pathway has been a focus of copious research in recent years. For example, Main, Dahl, and Darke (2007) describe a sinister attribution error whereby consumers become irrationally distrustful and misattribute flattery expressed by a salesperson to persuasive intent (see also Kramer 1994). Importantly, this attribution appears to be generated at least partly within the experiential system: distrust is increased by flattery even when no plausible persuasive intent is available, and consumers cannot consciously identify the reason for their distrust. At the brand level, a notion similar to trust is brand credibility, which is generally conceptualized as a combination of trustworthiness and expertise. Across a variety of product categories, Erdem and Swait (2004) observed that a brand’s credibility has positive effects on consideration and choice (especially in categories marked by high uncertainty), and follow-up analyses indicated that the benefits of credibility are predominately due to the trustworthiness dimension. As described by the authors, the trustworthiness dimension appears to tap into highly accessible brand associations, which are likely a product of experiential processing. Persuasion Knowledge The persuasion knowledge model (PKM—Friestad and Wright 1994) argues that over time, individuals acquire a set of beliefs about the psychological mediators of persuasion and also about the effectiveness and appropriateness of persuasive tactics. Faced with a particular communication, individuals are assumed to hold the goal to form accurate attitudes toward both the agent and the communication topic. As a result, they attempt to make inferences about the motivations underly-

16

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

ing the persuasive communication by applying their accumulated persuasion knowledge. Once a particular action is recognized as a persuasive tactic, a “change-of-meaning” occurs whereby the individual attempts to identify effective and appropriate strategies to cope with the attempt (e.g., withdrawing, actively resisting). As is the case with trust and credibility, elements within the persuasion knowledge model appear to involve both experiential and analytical processing. Concerning the former, persuasion knowledge itself can be understood as a set of stored procedural knowledge regarding target-agent interactions. Frequently, this knowledge is activated without conscious reflection, as when an agent’s behavior activates highly accessible scripts portraying persuasive episodes (e.g., during a discussion with customers regarding price, a car salesman says that he needs to “speak to the manager”). Nonetheless, the attributional process described by Friestad and Wright (1994) is concerned mostly with attributions resulting from analytical, rule-based processing. The model assumes that a target’s goal of holding valid topic and agent attitudes is moderated by motivation and ability, and the identification of persuasion attempts requires thoughtful elaboration on either message content, its source, or the persuasive tactics being utilized. Evidence for persuasive recognition as a deliberative process is provided by Ahluwalia and Burnkrant (2004), who focus on the use of rhetorical questions in advertising. The authors conceptualize rhetorical questions as an “artful deviation in style” that, by virtue of being unexpected, may focus attention and elaboration on the source of that deviation rather than the message itself. A series of ad-based empirical studies confirmed this possibility: when the rhetorical format was highly salient (e.g., for ads containing many rhetorical questions), participants tended to base their attitudes more on their perceptions of the source than the content of the ad. For example, the perceived appropriateness of sales tactics used by a shoe manufacturer affected evaluations of its brand more strongly when ads for the brand contained many rhetoricals (e.g., “[Brand X] shoes are beneficial for you, aren’t they?”). In a similar vein, Williams, Fitzsimons, and Block (2004) show that the effect of asking questions on behavior interacts with perceptions of the party responsible for the questions. A long line of research has established that asking individuals to state their intentions can lead them to behave in line with their stated intentions (Morwitz, Johnson, and Schmittlein 1993; Sherman 1980). However, Williams, Fitzsimons, and Block (2004) demonstrate that this “mere-measurement effect” is attenuated or even reversed when individuals perceive that persuasive intent underlies the questions: participants in their studies did not change their behavior to match stated intentions for questions asked by an obviously self-interested source. Most relevant for present purposes, cognitive capacity was required for a “change of meaning” to occur: when participants were placed under cognitive load during the intention questions, mere-measurement effects were observed regardless of source motivations. To the extent that analytical processing depends on the accessibility of task-relevant rules (e.g., Kahneman 2003), heightening the salience of principles and tactics associated with persuasion should increase the likelihood of a communication being identified as a persuasive attempt. This notion is illustrated by Brown and Krishna (2004) in an examination of the effect of default options on choice. Numerous prior studies have documented that consumers tend to anchor their purchase decisions on whichever option is presented as the default, so that different defaults lead to different decisions. Brown and Krishna replicate this result but also demonstrate that when consumers’ social intelligence about the behavior of the marketplace (marketplace metacognition) is invoked, they become more skeptical about defaults and attempt to correct for the unwanted influence. Typically, the persuasion coping process is also cast in analytical terms, as consumers seek to derive response strategies that are justifiable, productive, and appropriate to the situation. In the Brown and Krishna (2004) studies, for example, participants who recognized default options as

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

17

manipulative tended to determine that a counteracting response was appropriate. The result was a “backfire effect” by which these participants actually chose less expensive options when presented with more expensive defaults. In an extensive overview of contingent responses to persuasion, Kirmani and Campbell (2004) describe consumers as goal-directed individuals who attempt to manage exchanges with a marketing agent in order to serve those goals. Through the use of qualitative surveys, the authors identify fifteen different response strategies reflecting a consumer’s dual roles of goal seeker (asking for information, testing an agent’s expertise) and persuasion sentry (resisting assertively, withdrawing from an interaction). In sum, the work reviewed here supports a general consensus that after recognizing a persuasion attempt, consumers engage in a deliberative process of matching their coping responses with personal goals and situation-specific factors. Summary The preceding paragraphs have highlighted the operation of system-1 and system-2 in the occurrence of source effects, the development and activation of persuasion knowledge, and persuasion coping. Hopefully, future empirical work will more clearly integrate the notion of dual processing systems into frameworks traditionally used in this area. A promising illustration is provided by Darke and Ritchie (2007) in their research into consumer skepticism. The authors demonstrate that after an initial experience with deceptive advertising, consumers form a bias against subsequent advertising, becoming defensive and distrustful of future claims (even claims from a different source). The model they present to explain these results incorporates properties of both processing systems: after an initially deceptive encounter induces negative associations, subsequent encounters lead to the automatic activation of these associations via defensive stereotyping, which in turn leads to biased deliberative processing of message contents: the consumer actively counterargues in order to avoid being fooled again. This example illustrates the utility of a dual-systems perspective for thinking about the psychological determinants of persuasion, and other scholars would benefit greatly from considering such a perspective. Metacognition Over the last decade, the topic of metacognitive influences on judgment and choice has emerged as an abundant research stream. Below, we consider the ways in which experiential and analytical processing are distinctly and simultaneously integral to the experience of metacognition. After introducing the topic, we concentrate on applications involving fluency, ease-of-retrieval effects, and thought confidence. Thinking About Thinking The traditional cognitive perspective tended to assume that conscious judgment and attitude formation are a function of specific thoughts regarding the focal stimulus (e.g., Wyer and Srull 1989). Supplementing this view, however, a vast body of recent research has demonstrated that thought processes can be as important as thought content. Various theories fall under the “metacognitive” umbrella, but all share the notion that our subjective experience of thought generation, or “thinking about thinking,” plays a critical role in the judgment process (e.g., Novemsky et al. 2007; Schwarz 2004; Wanke, Bohner, and Jurkowitsch 1997). From a dual-systems perspective, this research exemplifies the profound insights that can be gained by considering experiential and analytical processing as intertwined (rather than distinct) cognitive pathways.

18

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

The combined function of both processing systems in metacognitive experiences is perhaps best illustrated by Schwarz (2004), who reviews the topic and presents a model of metacognition as a theorydriven process. Almost every judgment and decision situation requires the processing or recruitment of declarative information (e.g., reasons that one product is superior to another). Schwarz argues that this cognitive activity produces not only the declarative content itself but also a subjective feeling regarding the accessibility of that content. The latter constitutes a form of experiential information (e.g., “coming up with those reasons was hard”), similar to moods or bodily sensations. Unless it is deemed irrelevant to the task, this experiential information will serve as a source of judgment-related inferences. However, the particular inferences that are made (“the product must be inferior,” “the products are complex,” etc.) depend on the naïve theory that is brought to bear by the consumer to interpret that experience; different naïve theories may be more accessible during different judgment tasks. The Schwarz (2004) model clearly illustrates how multiple roles may be enacted by each system during the processing of metacognitive experiences. Next, we use this perspective to consider recent applications in consumer research involving fluency, retrieval ease, and thought confidence. Fluency One major subtopic within the metacognitive arena deals with the notion of fluency, or the subjective ease of processing information. The general hypothesis is that individuals tend to misattribute this subjective ease to their internal evaluation of the stimulus. The best known example of the fluency principle is the “mere exposure effect,” in which repeated exposure to a stimulus leads to enhanced liking (Zajonc 1968). The relevance of mere exposure to the marketing discipline is readily apparent, and indeed, the phenomenon has often been examined as a conduit to advertising effectiveness. In a recent example, Fang, Singh, and Ahluwalia (2007) created banner ads that rotated on a computer screen while participants performed an unrelated task. The number of times that each ad appeared on the screen was varied as a manipulation of exposure. In line with a fluency account, results indicated that the number of exposures affected liking for the advertisements; however, in conditions where the positive feeling of fluency could be misattributed (to music playing in the background), the effect of exposure on liking was eliminated. The effects of fluency have even been observed for a single ad exposure (Lee and Labroo 2004): in a set of experiments, advertised products were rated more favorably when ad storyboards created an expectation that the target product appear (e.g., a bar scene as the background for a beer product) than when no such expectation was created (e.g., a bar scene as a background for a vitamin product). Processing fluency has also been used as an explanation for the aforementioned mere measurement effect. Across a series of experiments, Janiszewski and Chandon (2007) demonstrate that contrary to most existing models (e.g., Morwitz, Johnson, and Schmittlein 1993), questions can affect behavior without increasing the accessibility of preexisting attitudes toward the behavior. When presented with two brands of a novel product but asked about purchase intentions for only one of the brands, participants’ actual likelihood of purchase increased for only the measured brand. The authors argue that the effect was a result of fluency—stemming from redundancy in the measurement task and actual behavioral decision—that was then misattributed to liking for the product. Corroborating this account, the mere measurement effect was not induced by questions about purchase intent for the general product category or about the product’s general appeal. Two general accounts have been proposed to explain the process by which fluency affects other cognitions, and the accounts are notable for their distinct emphasis on experiential versus analytical processing. In the perceptual fluency account, (e.g, Janiszewski and Meyvis 2001), the assumption is that people make inferences based on their experience of fluency that are in turn

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

19

based on logical cognitive principles (e.g., “thoughts that are easy to process are more certain”). At times, for example, the experience of difficulty processing a stimulus can itself invoke a system-2 correction mechanism that discounts impulsive judgments, presumably because of the inference that rigorous analysis is needed (Alter et al. 2007). Note, however, that although the experience of subjective ease falls clearly within the experiential system, resulting metacognitive inferences are based on acquired rules or principles and thus constitute an output of analytical processing; that is, two individuals may share the same metacognitive experience but make different inferences due to different theories of mental activity. In the affective fluency account (e.g., Winkielman et al. 2003), fluent experiences generate positive affect directly, due to either stimulus familiarity or their signal of progress toward a goal. In this case, metacognitive inferences are based on the misattribution of irrelevant positive affect to liking for a stimulus (Schwarz and Clore 1983). Note that this process of affective generalization may occur without conscious inference (in fact, a failure in inference-making underlies the misattribution). Therefore, the affective fluency account requires limited operation of the analytical system. Ease of Retrieval Another class of metacognitive phenomena attracting a great deal of recent attention is illustrated by the aforementioned “ease-of-retrieval” effect. When asked to generate reasons for a particular attitude or behavior, individuals focus not only on the content of the reasons that they generate, but also on the ease with which that generation occurs (Schwarz et al. 1991). This ease or difficulty in producing reasons then leads to inferences regarding one’s attitude. In a well-known example (Wanke, Bohner, and Jurkowitsch 1997), individuals rated BMW automobiles more favorably when asked to generate one rather then ten reasons supporting that brand. In a more recent application, Novemsky et al. (2007) show how ease of retrieval can exacerbate choice conflict: participants in their studies were given descriptions of specific consumer products and asked how difficult it would be to generate either two or ten reasons for choosing one. Compared to participants in the two-reasons condition, those in the ten-reasons condition were more likely to defer their choice until later or to choose a compromise option.2 Other work has demonstrated that the accessibility of task-relevant knowledge moderates the effect of retrieval ease. Using automobile stimuli similar to those of Wanke, Bohner, and Jurkowitsch (1997), Tybout et al. (2005) show that ease of retrieval does not affect ratings when relevant content is either highly accessible (e.g., after viewing an ad discussing automobile attributes) or inaccessible (e.g., when the target brand is unfamiliar). The authors suggest that in these cases, subjective ease or difficulty is discounted because it is expected. This intriguing result is relevant to the present framework in at least two respects: first, expectancies undoubtedly moderate the phenomenological experience of retrieval difficulty; second, expectancies also influence the inferences drawn by the decision maker regarding that experience. In some cases, the notion of discrete processing systems has been explicitly integrated with the ease-of-retrieval phenomenon. After asking participants to generate two or eight reasons for a position and then state their agreement with that position, Danziger, Moran, and Rafaely (2006) administered the Rational-Experiential Inventory (Pacini and Epstein 1999), a trait measure of processing style. The classic ease-of-retrieval effect was observed for participants high in experiential processing, suggesting that experiential engagement is indeed fundamental to the utilization of subjective difficulty as a cue to judgment. However, the effect was also observed among individuals low in experiential processing who were first asked to evaluate the perceived difficulty of the generation task. Presumably, individuals who do not normally recognize (or make use of) the experiential information provided by subjective ease will do so when this information is made

20

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

salient. Interestingly, other results suggest that a tendency toward analytical processing mitigates the ease-of-retrieval effect. Silvera et al. (2005, study 2) asked participants to conjecture about the cause of a negative event (e.g., a product selling poorly) by rating the likelihood of various causes on a fault tree (typically, individuals exhibit an “omission bias” by overestimating the likelihood of causes that are explicitly represented in the tree—Fischhoff, Slovic, and Lichtenstein 1978). Prior to this assessment, however, the subjective difficulty of generating causes was manipulated by asking participants to generate their own fault trees (for a different event) containing either two branches (low difficulty) or eight branches (high difficulty). Results indicated that this subjective difficulty manipulation interacted with participants’ need for cognitive closure (NCC). For participants low in NCC, the content of the generation task was paramount: larger self-generated trees led to a reduction of the omission bias. However, for participants high in NCC, larger self-generated trees actually enhanced the bias. The authors cast their findings in analytical terms, arguing that high-NCC individuals focus “mindlessly” on the subjective experience of generating examples. Thought Confidence A distinct approach to the study of metacognition has been taken by Petty and colleagues (e.g., Petty, Briñol, and Tormala 2002) in their notion of thought confidence. Briefly, this theory states that the impact of specific thoughts on attitude change will be greater when an individual has greater confidence in those thoughts; however, these effects should be restricted to tasks involving “relatively extensive information processing” (high elaboration in the ELM sense). Applying the idea to persuasive communication, Briñol, Petty, and Tormola (2004) asked participants to generate a list of thoughts about an advertisement before providing their attitudes toward the focal product. In separate studies, confidence in listed thoughts was either measured or manipulated (by credibility of the ad source), and elaboration was measured indirectly via need for cognition (NFC). Results indicated that low thought confidence weakened the effect of thought valence on resulting attitudes, but only for participants high in NFC. The possibility that thought confidence matters only under high elaboration appears to link the phenomenon to the analytical system. However, as with fluency and ease-of-retrieval effects, it should be noted that the subjective feeling of thought confidence might be relatively automatic and effortless, acting as an output of the experiential system that is subject to subsequent analytical processing. Immersive Experiences Given its roots in attitude and measurement theory, the classical focus of marketing research on more “cognitive” topics is understandable. However, an enduring trend toward diversification has opened the field to numerous subjects outside of this classical focus. In the present section, we address emerging research on immersive consumer experiences, focusing in particular on the topics of narrative transportation, imagery, interactivity, and embodied knowledge. A common theme across these topics is the extent to which system-1 operates not only as a “quick-and-dirty” alternative to rule-based processing, but also as a complex, dynamic approach to judgment and evaluation. Transportation Marketing scholars have recently shown an increasing interest in the notion of transportation experiences. This topic is an outgrowth of the idea that people generate stories to organize their experiences, take various perspectives, and formulate goals (Pennington and Hastie 1986). The

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

21

narrative transportation approach to marketing communication argues that product and brand meanings are created, organized, and brought to mind via stories involving brand-user associations. In a seminal work on the topic, Escalas (2004) suggests that individuals form self-brand connections (SBCs) by processing brand-relevant experiences in a narrative manner. An implication is that marketers may help consumers to form favorable SBCs by managing the stories embedded in messages about the brand. As evidence, Escalas presents an experiment in which participants were given an advertisement presenting the same information in either a story-based or a vignette-based manner. Compared to vignette-based ads, story-based ads were shown to evoke more narrative processing, stronger SBCs, and more favorable attitudes toward the brand. For the present discussion, it is worthwhile to consider narrative processing of brand experiences in terms of the experiential-analytical distinction. Processing by stories, while involving a degree of conscious control, is fundamentally about creating schemas or scripts by which personal experience is structured—that is, forming a coherent mental representation. Generally, therefore, narrative construction is considered a matching process by which incoming information is paired with episodes stored in memory, such that existing stories are reinforced or reinterpreted in some way (Shank and Abelson 1995). The extent to which this matching process is associative versus rule-based is a complex issue (and probably dependent on other variables). One approach is to conceptualize story construction as a synergistic process by which brand experiences are integrated into existing schemas in a rule-based manner. The end result is an updating of one’s associative network for the brand and, in the case of a strong SBC, the incorporation of the brand into an individual’s self-concept. Thus, the brand becomes embedded in a set of associations representing the self, and SBCs can be automatically activated and retrieved in future contexts. Taken this way, narrative construction illustrates one mechanism by which utilization of system-2 processing alters or strengthens the associative network underlying system-1 (Smith and DeCoster 2000). A related phenomenon is that of media transportation, by which an individual is absorbed into the narrative flow of a story. Wang and Calder (2006) suggest that media transportation can be considered a type of pleasurable flow in which consumers are engaged and absorbed by their experience (e.g., Csikszentmihalyi 2000). Extending previous research showing that media transportation is most important for story-relevant evaluations (Green and Brock 2000), Wang and Calder suggest that this immersion can produce either positive or negative effects on advertisements presented temporally close to the transportation experience. In a series of experiments, ads were found more effective to the extent that they did not intrude on a highly immersive transportation experience (in one example of intrusion, sections of a highly engrossing story were broken apart by a fast food ad). Importantly, manipulations of involvement with the story (in terms of personal consequences) did not moderate the effect of ad intrusiveness. Given that involvement is traditionally associated with analytical processing (e.g., Petty, Cacioppo, and Schumann 1983), this finding reinforces the concept of story immersion as an experiential, system-1 phenomenon involving absorptive perception of story elements and rapid activation of related scripts or schemas. Imagery and Interactivity Emerging research on product imagery, experience, and interactivity further illustrates the extent to which experiential processing is becoming a focus in the study of consumer behavior. For example, Schlosser (2003, 2006) provides a fascinating discussion of the ways in which interacting with a virtual object can affect mental representations and subsequent behavior. The author focuses on object interactivity, defined as the ability to manipulate objects in a virtual world as if they were real (e.g., clicking a shutter to take a virtual picture, opening a box to reveal product information).

22

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

Compared to static interactions via pictures or text, object interactivity elicits a “virtual experience” with a product, producing mental images that are rich and conceptually detailed. Mental imagery, which involves the encoding of information in terms of concrete, nonverbal, sensory representations, stands in direct contrast with discursive processing, which is dependent on symbolic, linguistic, and numeric manipulation (MacInnis and Price 1987); not surprisingly, the two are typically identified with operation of the experiential and analytical systems, respectively (e.g., Sloman 1996). With its basis in mental imagery rather than logical analysis, experiential processing of interactive information formats has been shown to affect consumer cognition in a variety of interesting ways. For example, Schlosser (2003) presents a series of Web-based experiments comparing the passive presentation of information via text and graphics to interactive presentation of the same information. The interactive presentation format led to a robust positive effect on purchase intentions, apparently by evoking mental simulation of actual product use. Importantly, this beneficial effect of interactivity was mediated by the amount of mental imagery processing that was evoked, but it was independent of both cognitive elaboration during the task and trait need for cognition. Together, these findings indicate that object interactivity facilitates the induction of visualization-based experiential processing.3 In addition to positive effects on intention, however, Schlosser (2006) has shown that interactivity—and the vivid imagery it produces—can actually lead to decrements in memory for a product. In a series of studies, participants were presented with product information on websites that were either static or interactive in format. When tested on their memory for product features, participants exposed to the interactive websites were more accurate overall, but they were also more subject to false positives, incorrectly ascribing specific features to products. Notably, false memories were amplified when instructions encouraged the use of imagery processing. In most applications, the use of imagery appeals has been shown to increase the effectiveness of persuasive messages; for example, individuals estimate the likelihood of getting a disease to be higher when symptoms are more easily imagined (Sherman et al. 1985). However, this imagery effect may depend on the ease with which message recipients can engage in task-specific imagery processing, a capacity that Petrova and Cialdini (2005) label imagery fluency. In a series of experiments, the authors demonstrate that ad-based product evaluations can be enhanced by a specific appeal to use imagery processing. However, imagery fluency did not moderate preferences for individuals low in dispositional imagery skill or for ad contexts that were not vivid (concrete, pictorial, easy-to-imagine). In addition, the effects were only observed for participants highly attuned to their internal experiences, as measured by the Private Self-Consciousness Scale (Fenigstein, Scheier, and Buss 1975). These findings may be cast in terms of the affective fluency account (described earlier), in which fluent experiences generate affect that is transferred directly to a stimulus (Winkielman et al. 2003). If so, then imagery fluency requires limited operation of the analytical system and should be relatively unaffected by processing depth; supporting this argument, the observed effects were independent of cognitive elaboration and recall. On the other hand, it appears that three forms of accessibility may be critical to the operation of the experiential system in this context (see also MacInnis and Price 1987). The formation of affective, imagined representations requires both vivid cues (to stimulate visual processing) and dispositional imagery skill (to generate vivid mental representations), while the incorporation of these imagined representations into subsequent judgments requires sensitivity to subjective experience. Flow and Play Activities The heightened levels of interactivity and perceptual stimulation characterizing online consumer behavior have made this topic amenable to theories emphasizing immersion and psychological

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

23

engagement. In particular, recent research has examined goal-directed Web behavior as a form of “flow” generating intrinsic pleasure and arousal. According to the popular four-channel model (Csikszentmihalyi 1990), flow results from a balance between task challenges and individual skills, such that states of mind may vary from apathy (skill and challenge levels are both low) to boredom (skill exceeds challenges) to anxiety (challenge exceeds skill) to flow—an optimal balance of challenge and skill above some critical level. Research by Novak, Hoffman, and Duhacheck (2003) suggests that although flow is a common occurrence during online activity, the condition is more often reached during task-oriented than recreational activities. A related but distinct concept is captured by the notion of “play” behaviors (Mathwick, Malhotra, and Rigdon 2001), in which the search for product information serves as a leisure activity providing pleasure and diversion. In online shopping contexts, recent research has demonstrated that play can have direct effects on attitude formation, in that the experience of enjoyment and escape during a search experience correlates with positive attitudes toward both the website and the brand being considered (Mathwick and Rigdon 2004). From a dual-systems perspective, the phenomenology of play and flow activities as absorptive, immersive engagement is clearly a result of experiential processing. However, both concepts also imply a critical role for system-2: experiential engagement occurs in conjunction with focused, task-directed processing, so that, for example, the attitudinal effects of play are more prominent under high-involvement conditions (Mathwick and Rigdon 2004). Together, these research streams reveal a fascinating connection between experiential and analytical processing during the conduct of engrossing consumer activities. Embodied Knowledge The emerging emphasis on the experiential underpinnings of consumer behavior is exemplified further by research into embodied knowledge. The concept of embodied knowledge is intended to represent “information elements that are generated and maintained outside the brain cavity and that are incorporated into consumer assessments of products and services” (Rosa and Malter 2003, p. 63). For products whose consumption involves a high degree of sensory and somatic activity (the smell of a perfume, the feel of clothing), embodied knowledge is assumed to play a critical role in directing consumer thought processes. Rosa and Malter (2003) identify three distinct elements of embodied knowledge—body mapping and monitoring, proprioceptive knowledge, and body boundaries—and propose that these three elements are combined with conceptual knowledge to form mental simulations of product experience. During the new car shopping process, for example, consumers will compare various models not only on attributes such as acceleration, safety, and cargo capacity, but also on the “feel” of the driving experience, the extent to which body boundaries are enhanced or diminished by the cabin design, and the way in which the steering wheel and pedals serve as an extension of the limbs. Although still in its infancy, the embodied knowledge perspective is notable for considering aspects of sensory experience (olfactory, haptic, etc.) that have traditionally received little attention in marketing research. Future exploration could yield insightful contributions to models of sensation, perception, and the broader experiential system. Emerging Issues for Marketing Research Having discussed a variety of recent developments from a dual-systems perspective, we now shift attention to unresolved conceptual issues that bear directly on topics relevant to marketing research. Here we will focus on three particular concerns: the role of affect, the question of multiple attitudes, and the notion of corrective processes.

24

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

The Role of Affect in a Dual-Systems Framework During the introduction to this chapter, we noted that there is considerable disagreement over the role played by affect in the operation of either system. This section will extend that discussion by first reviewing the traditional identification of affective experience with system-1, then highlighting various potential roles for system-2, and finally considering relevant findings from recent marketing research. Affect as a Product of System-1 In the diverse literature on affect, a common approach is to conceptualize emotions and feelings as a defining property of the experiential system. For example, Peters and colleagues (2007) argue that system-1 processing is fundamentally directed toward the production of evaluative, affective feelings. In literature on risk and uncertainty, the “affect heuristic” (Slovic et al. 2002) describes a process by which individuals consult an overall affective impression of a decision rather than weighing its specific attributes. More broadly, Loewenstein et al. (2001) have proposed a “risk-as-feelings” hypothesis, in which individuals evaluate risky situations not only in terms of cognitions about the likelihood and value of possible outcomes, but also in terms of “feeling states” such as worry, dread, or anxiety, which will frequently diverge from the cognitive reactions (e.g., emotions are largely insensitive to probability).4 Of the various models presented earlier, the one most congruent with these ideas is the affective-cognitive approach of Shiv and Fedorhikin (1999). As described before, their framework proposes that affective reactions tend to occur more rapidly than cognitive deliberations, and thus the former will dictate decision making when individuals are limited in cognitive capacity, leading to impulsive choices. Extending beyond the self-control domain, Nowlis and Shiv (2005) apply this model to the effect of cognitive distraction on perceptual experiences. Across multiple experiments involving the sampling of food products, the authors demonstrate that being distracted during the consumption of a sample actually increases liking for the product. Results are explained in dual-process terms: ultimate liking for a somatosensory experience consists of an affective component (emotional reactions such as delight) and an informational component (qualities such as the texture or composition of the food), and cognitive distraction serves to increase the influence of the former component relative to the latter. To a certain extent, these results parallel a stream of evidence collected by Wilson and colleagues regarding the effect of analyzing reasons on choices and satisfaction (Wilson et al. 1993; Wilson and Schooler 1991). In decisions where ultimate satisfaction is likely to be driven by attributes that are difficult to verbalize (e.g., posters, strawberry jams), individuals who are forced to deliberate on their preferences tend to make “worse” decisions and ultimately be less satisfied. Some (but not all) of these findings can be explained by the overweighting of cognitive judgments versus affective reactions. Possible Roles for System-2 Nevertheless, a wide body of research suggests that the role of emotions in dual-systems processing may be more complex than previously thought. Although few models directly emphasize the role of affect in the operation of system-2, many acknowledge the use of affective experience as an input to analytical processing, and many others allow for the production of affect as an output. An example of the latter case is provided by Shiv and Fedorikhin (2002) in a follow-up to their well-known (1999) “cake versus fruit salad” demonstration. Participants in the follow-up who

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

25

were exposed to both alternatives under no cognitive constraints became more likely to succumb to the tempting option when given longer to deliberate. The authors suggest that in contrast to “low-order affect” that occurs spontaneously, the behavior of these participants was driven by “high-order affect” brought about by elaboration of various sensory qualities of the options. Research on decision stress and coping also implicates the role of high-order affect. For example, Drolet and Luce (2004) manipulated cognitive load and then presented participants with decisions requiring trade-offs between emotionally laden attributes (e.g., safety and price). Results indicated that unconstrained participants avoided these emotional trade-offs and relied instead on noncompensatory processing; in contrast, constrained participants were willing to form trade-offs and thus made normatively better decisions. Basic research on affect and emotions also points to the role of controlled, deliberative processing. For example, appraisal theories of emotion divide discrete affective experiences in terms of underlying cognitive dimensions (Lazarus 1982; Scherer, Schorr, and Johnstone 2001). “Selfconscious” emotions such as pride, shame, and embarrassment seem to be predicated on analysis of attributions underlying a situation (Lewis 1997); as opposed to more “basic” emotions such as fear and anger, self-conscious emotions require time for elicitation and are less common in children than adults. In discussing the role of feelings in consumer behavior, Cohen and Areni (1991) distinguish three “types” of affect, the first two of which implicate associative processing alone. However, “Type III affect” is slow to develop, is cognitively mediated, and involves scrutiny of the stimulus-provoking event. In sum, the broad evidence for deliberative “high-road affect” exemplifies the parallel and reciprocal relationship between experiential and analytical pathways. Other frameworks capture the interaction between affect and deliberative processing by positioning the former as an input to the latter. This principle is incorporated in a wide variety of theories postulating multiple routes of affect in judgment (e.g., Forgas 1995). The idea has long been a basis of the “affect-as-information” hypothesis (Martin et al. 1993; Schwarz and Clore 1983), which argues that individuals use their feeling states (sometimes erroneously) as a source of data to make inferences about their present situation or attitudes. Peters (2006) describes four ways that affect may be utilized during an emotional decision: as information in itself, as a source of motivation, as a “common currency” by which to integrate diverse attributes, or as a spotlight on certain information. Although these approaches have tended to emphasize the ways that affect causes bias in processing strategies, few scholars would deny that feelings are sometimes pivotal to good decision making. For example, in the popular risk-as-feelings model (Loewenstein et al. 2001), anticipated emotional reactions are considered to be a logical basis for decisions. If so, then rule-based, analytical processing must be able to incorporate these emotional reactions into its operation. In a unique approach highlighting the usefulness of feelings, Pham (2004) describes affect as a system unto itself, distinct from (but overlapping with) traditional conceptualizations of system-1 and system-2. The model stresses metacognitive capabilities of the affective system and argues that above and beyond simple judgments of liking, feelings play numerous adaptive roles by serving as a source of inference for assessments of preference strength, situational requirements, causal attributions, and so forth. In recent years, one of the most prominent frameworks to incorporate affect as a component of deliberative decision making has been the somatic marker hypothesis (Damasio 1994). Based on a program of neuroscientific research, Damasio and colleagues argue that decision makers encode the emotional consequences of various alternatives in the form of “somatic markers” via specific areas of the prefrontal cortex. In a series of gambling experiments pitting low-risk options with a modest expected value against high-risk options with potentially catastrophic losses, Bechara and colleagues (1997) demonstrate that individuals whose prefrontal cortex is impaired perform

26

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

much worse than individuals without prefrontal damage. Despite learning and acknowledging the riskiness of the unsafe option, impaired patients return to these options quickly after encountering a loss, seemingly failing to incorporate the momentary fear that they experience into their decisions. From a dual-systems perspective, somatic markers may represent immediate, association-based outputs of the experiential system that provide critical sources of information for the conduct of slower, rule-based processing of alternatives. Evidence from Consumer Behavior A series of recent investigations in the marketing literature support the broad conjecture that affect is a result of experiential processing that nonetheless influences—and is influenced by—the analytical system. Using the heuristic-systematic model as a backdrop, Darke, Chattopadhyay, and Ashworth (2006) presented low- and high-elaboration groups with product judgments involving both affective (irrelevant) and informational (relevant) cues. Results indicated that affective cues were utilized in both elaboration conditions: for the low-elaboration group, affect exerted a direct heuristic effect on overall judgments, while for the high-elaboration group, affect exerted both a heuristic effect and a more systematic effect through biased evaluation of product information. When the irrelevance of affect was pointed out, participants were able to correct for its influence on their evaluative thoughts but not for its direct, heuristic influence.5 In conceptually related research, Bakamitsos (2006) demonstrates the multiple roles played by irrelevant mood in the formation of advertising reactions. After being induced with positive mood, participants evaluated advertisements more favorably, but only if: (1) the potential biasing affects of mood were not salient, and (2) the ads contained no attribute information. In addition to this heuristic function, however, mood appeared to serve as a resource for analysis (Isen, Daubman, and Nowicki 1987): when target ads were paired with related contextual ads, participants in positive moods engaged in more relational processing, focusing on commonalities and hierarchical categories. Together, these examples present compelling evidence that the influence of affect on consumer processing occurs through both experiential and analytical pathways. Additional evidence is provided by Yeung and Wyer (2004), who investigate the diverse effects caused by “initial affective impressions” of a product. Participants in their studies were given a mood induction and then presented with verbal product information. Happy participants evaluated this information more positively only if the product domain was hedonic in nature, so that feelings would be deemed relevant to the evaluation (see also Pham 1998). However, when a picture of the product was presented alongside the verbal information, evaluations were correlated with mood regardless of the domain. Pictures appeared to catalyze the formation of affective impressions (which were biased by prior mood), and these impressions were incorporated into subsequent judgment. Considerably less consumer research has focused on the affect formation process itself. In a provocative exception, Herr and Page (2004) uncover systematic asymmetry in the methods by which people generate judgments of liking versus disliking for a product. Reviewing prior research on “positivity” and “negativity” effects, the authors suggest that although individuals are predisposed to retrieve positive information, negative information retrieval has a stronger effect on subsequent processing. Participants in their experiments first viewed pictures of consumer products on a computer screen, then provided separate liking and disliking judgments for every product. Across a number of studies and conditions, participants were consistently faster to report their liking than their disliking. Furthermore, an asymmetric priming effect was observed: participants were much faster to give liking judgments when first asked for disliking judgments, but in the reverse order this facilitation effect was minimal. These intriguing results imply different

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

27

roles for experiential and analytical processing during affect retrieval, suggesting that formation of liking judgments is a primary, spontaneous process, but formation of disliking judgments is secondary, cognitive, and easily interrupted. Multiple Systems = Multiple Attitudes? The notion that an individual may hold multiple attitudes toward the same object has recently taken root in the social cognitive arena, and applications of this idea in consumer behavior have attracted considerable attention. Below, we briefly review the distinction between implicit and explicit attitudes and then consider the utility of a dual-systems perspective for conceptualizing this distinction. As an illustration, we present an example from our own recent work. Implicit and Explicit Attitudes According to classical approaches, attitudes can be conceptualized as “enduring propensities to respond,” and attitude change is assumed to involve the overwriting of a prior propensity with its replacement (Anderson 1971). However, contemporary researchers have instead suggested that multiple attitudes may be held simultaneously (Petty et al. 2006; Wilson, Lindsey, and Schooler 2000). Although the various paradigms differ considerably, they share the notion that an existing attitude need not be overwritten with newer associations or beliefs toward the attitude object. Typical of the current approaches is the notion that a prior-formed implicit attitude may continue to exist in conjunction with a newly formed explicit attitude, and this implicit attitude may still be brought to bear in a task where conditions are favorable to its recruitment. An interesting question, therefore, is the relationship of disparate attitudes to the underlying operation of experiential and analytical processing systems. Researchers making the implicit-explicit distinction have supported their thesis by presenting evidence that explicit attitudes correlate highly with overt judgments, ratings, and choices, whereas implicit attitudes are more predictive of subtler, indirect judgments or behaviors (e.g., the Implicit Association Test of Greenwald and Banaji 1995, the sequential priming measure of Fazio et al. 1986). In a prototypical example from consumer research, Brunel and colleagues (2004) compare implicit and explicit measures in cases where they are expected to either converge or diverge. The authors first demonstrate that implicit associations to the PC and Macintosh are positively correlated with explicit measures of brand attitudes, ownership, and usage. Next, they compare the usefulness of these measures for assessing attitudes subject to self-presentational concerns: after participants evaluated ads containing either white or black spokespersons, implicit (but not explicit) measures showed a favorable bias for same-race spokespersons. Maison, Greenwald, and Bruin (2004) provide further validation in the context of brand loyalty. Across a variety of products (yogurt, fast food, soft drinks), participants’ implicit attitudes revealed a preference for their own brand over a competitor. In addition, implicit measures predicted consumers’ ability to distinguish the brands in a blind taste test. Overlap with a Dual-Systems Framework Despite such promising applications, the attitude literature has been subject to considerable disagreement regarding the defining characteristics (and even the existence) of implicit attitudes (Fazio and Olson 2003; Gawronski, Hofmann, and Wilbur 2006). In particular, an assumption underlying implicit approaches is that the associations they capture are held beneath conscious

28

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

awareness (e.g., Banaji, Lemm, and Carpenter 2004), but a growing body of research has questioned the extent to which attitudes labeled “implicit” are in fact inaccessible to conscious introspection (Hofmann et al. 2005). Based on this concern, Gawronski, LeBel, and Peters (2007) propose that implicit attitudes are best understood as stimulus-relevant associations that are immediately accessible but also subject to validation processes that may reduce their impact on reported attitudes. In consumer research, validation is a critical step in the multiple-attitude model advocated by Cohen and Reed (2006): highly accessible attitudes are recruited and then subjected to a variety of representativeness checks and sufficiency checks before they are permitted to guide behavior. Conceptually similar ideas are presented by Wilson, Lindsey, and Schooler (2000), who use the term motivated overriding to describe cases in which individuals are at least somewhat aware of implicitly held attitudes but choose consciously to override them. Wilson presents a variety of settings where judgments are typically guided by explicit attitudes, but cognitive load induces individuals to respond in line with implicit attitudes. Applying these ideas to the dual-systems paradigm, we share the conjecture of other scholars that implicit attitudes operate primarily within the experiential system, but explicit attitudes operate mainly within the analytical system (Gawronski and Bodenhausen 2006). For example, Rydell and McConnell (2006) demonstrate that implicit attitudes are shaped by slowly acquired associations, unaffected by explicit processing goals, and predictive of spontaneous behaviors, whereas explicit attitudes are shaped by the application of rules, affected by explicit processing goals, and predictive of deliberate intentions. In a series of person perception experiments, participants were first exposed to a large body of valenced information about a target person (stage-1), then a smaller body of information that confirmed or contradicted the initial statements (stage-2). As expected, self-reported preferences aligned strongly with the valence of stage-2 information; however, implicit measures were completely unaffected by this information, reflecting only the valence of stage-1. An Example We have applied a similar methodology to the acquisition of new product information by consumers (Bond, Bettman, and Luce 2008). In keeping with others who question the conscious inaccessibility of implicit attitudes (Gawronski, LeBel, and Peters 2007), we speculate that implicit attitudes formed during an initial learning stage resemble “gut instincts” that may be evoked under proper conditions. Using a design similar to that of Rydell and McConnell (2006), we exposed participants to two stages of information about a new consumer product. All participants received positively valenced information at stage-1, but stage-2 was manipulated so that participants received either neutral (nonvalenced) or negative information. Before asking for product evaluations, we manipulated the extent to which participants thought that following “gut instincts” was appropriate (adopted from Avnet and Pham 2007). Results indicated that although stage-2 information did, in fact, influence evaluations, its influence was far less pronounced for participants induced to follow their instincts. Our findings offer an important caveat to the notion that implicit attitudes exert their influence primarily on subtle, indirect measures. Moreover, they reinforce the emerging argument that distinct, coexisting attitudes can be evoked by the experiential and analytical systems. Correction Processes Although rapid, association-based processing is an efficient means of directing behavior in many typical situations, this mode of thought will frequently lead individuals astray. Hence, the traditional view of correction processes has centered on the overriding of system-1 impulses by more rational deliberation.

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

29

In this section, we first review the traditional perspective and then consider a markedly different view by which experiential processing may outperform (or even correct for) flawed analytical processing. The Traditional Perspective In social psychology, the notion of correction was an outgrowth of early dual-process models, notably the elaboration-likelihood model (Petty and Cacioppo 1981) and heuristic-systematic model (Chaiken 1980). In keeping with the idea that elaboration and systematic thinking are more conducive to enduring attitude change, psychologists have tended to conceptualize processing biases as the result of flawed system-1 processing. A variety of formal models of correction have been introduced, including the flexible correction model (Wegener and Petty 1997), set-reset model (Martin 1986), and inclusion-exclusion model (Schwarz and Bless 1992). Although they possess important distinctions, these models share the assumption that various “irrelevant” factors operate by systematically biasing heuristic processing, so that correction involves both the recognition of these influences as irrelevant and an attempt to remove their influence via more thorough deliberation (although overcorrection can sometimes result—see Wegener and Petty 1997). Characteristic features of many dual-systems frameworks imply a view of correction similar to the one just described. For example, given that experiential processing is far less time-consuming than analytical processing, it is reasonable to assume that the latter would bear responsibility for adjusting faulty outputs of the former, while the reverse would not be the case. This viewpoint is most easily identified with the intuition/reason dichotomy of Kahneman and Frederick (2002), in which system-2 operates (in part) as a monitoring mechanism that assesses the outputs of system-1 and then modifies them accordingly. Under this framework, cognitive capacity is an important determinant of correction, but the monitoring process can break down for a variety of other reasons. One prominent reason is the phenomenon of “intuitive confidence” (Simmons and Nelson 2006); that is, as a result of their accessibility, intuitions are often perceived as extremely compelling by the individual and therefore pass through the reasoning system unmodified. Correction Versus Selection In contrast to the traditional perspective, numerous emerging research streams have presented domains in which choices, judgments, and behaviors resulting from experiential processing produce outcomes superior to those resulting from deliberative analysis. Within social and cognitive psychology, accumulated evidence to this effect underlies broad theories of intuition and “unconscious intelligence” (Gigerenzer 2007; Hogarth 2001). Such evidence is also apparent in the aforementioned research on debilitating effects of analyzing reasons for a choice (Wilson et al. 1993), and in the recently popularized notion of “unconscious thought” (Dijksterhuis 2004; Dijksterhuis et al. 2005), whereby individuals who are distracted for a period of time before choosing are observed to make normatively better choices. Within consumer behavior, the question of when intuitive versus analytical responding is superior has been directly addressed by Kardes (2006) in the form of a prescriptive model. Incorporating different informational variables, his model suggests that intuition will tend to outperform analysis when reliable feedback is available and inferential errors are easily detected. In sum, these diverse research areas present a view of correction that is quite distinct from the traditional approach. By focusing on the capacity of experiential processing to be complex, goal-directed, and adaptive, their findings call into question the portrayal of system-1 as a faulty, easily biased mechanism that is highly dependent on the corrective monitoring of system-2.

30

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

One way of recasting the conflict in these approaches is to consider that traditional research on correction processes has focused on domains where intuitive processing is likely to be flawed (e.g., the heuristics and biases, stereotyping). Therefore, the experiential system has acted primarily as a source of inferential biases, but rarely as an adaptive processor or adjustment mechanism. However, in other domains, such as those mentioned above or the various “immersive experiences” described earlier, the experiential system may be better suited and less subject to bias than its analytical counterpart. Earlier in the chapter, we presented evidence that individuals possess lay theories about the perceived suitability of “going with their gut” or “using their head” in different situations (Bond, Bettman, and Luce 2008), and researchers themselves might be well served by this perspective. The issue then becomes one of output selection rather than correction; that is, optimal processing involves a match between the task domain and the system whose operation is best suited to that domain. However, given the acknowledged interactivity of the two systems (with outputs of each serving as inputs to the other), more nuanced frameworks for correction may also be plausible. For example, one might conceive of situations where biased analytical processing is received as an input by the experiential system and “corrected” according to highly accessible rules, in a manner requiring little effort or intention. Regardless of the particular frameworks developed, it is clear that for researchers to fully understand the operation of correction processes in a particular domain, they must consider the domain-relevant functions performed by each underlying system. Conclusions Psychologists across a variety of disciplines have come to a consensus that human thought and reasoning can best be represented by two discrete, parallel, interactive cognitive systems, and the field of marketing is beginning to incorporate this emerging view. The present chapter has investigated the role of dual-systems models as they pertain to various recent findings within consumer behavior. In place of a comprehensive summary, we offer the following list of principles that we consider especially relevant to marketing research: ฀

฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ analytical processing. Research casting a phenomenon in terms of either system alone is unlikely to provide an adequate representation of underlying mental activity. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ search on the experiential system should encompass not only the “intuitive” use of errorful heuristics, but also the adaptive role of this system in mental imagery, perception, metacognition, and other complex processes. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ systems regularly provide inputs to (and receive outputs from) one another. Greater exploration of this dynamic interplay is sorely needed. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ to the experiential system, affect is frequently dependent on analytical processing, and analytical processing often incorporates affective information. ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ ฀ processing mode for different domains, and these perceptions will frequently be critical determinants of behavior. We close by considering implications of the present perspective for marketers in the field. Any new approach brings more questions than answers, and this certainly holds true for the applica-

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

31

tion of dual-systems models to marketing practice. For example, one set of questions concerns the use of these ideas for designing effective communications. How might communication tools be utilized to “resonate” at an instinctive level, to “make sense” at a rational level, or to do both simultaneously? What role do brand names, logos, and so forth play in “priming” the use of either system? Can oft-repeated taglines (“Just Do It,” “Image means nothing,” etc.) be utilized for this purpose? A common concern in designing marketing communications, mentioned earlier, is that consumers feel that they are being manipulated (e.g., Wright 2002). To the extent that resultant consumer skepticism is a form of system-1, gut-reaction against the message and its source, what tools can marketers use to overcome this obstacle? Another set of questions relates to brand equity and positioning. To the extent that brand equity represents a set of instinctive, positive associations with the brand (Keller 1993), what methods are most effective for creating these associations? How might a firm enjoying strong brand equity leverage features of the experiential system when introducing new offerings? Given that the memory structures underlying experiential processing change very slowly, what actions can be used by competitors to displace entrenched leaders? Two other applications merit particular attention. One is the extent to which many consumer “vices” (drug and alcohol abuse, impulsive purchasing, etc.) can be seen as a self-control failure involving overreliance on the experiential system: doing what “feels good” even when you “know better” (Baumeister 2002). Conversely, an opposite situation sometimes occurs whereby consumers rely too heavily on rational analysis, to the detriment of long-term satisfaction. This outcome is especially common when individuals are concerned with justifying their decision (Wilson et al. 1993) or avoiding future regret (Kivetz and Keinan 2006). Both categories of problems describe situations in which either experiential or analytical processing is excessively engaged (or weighted), to the detriment of consumer welfare. In dealing with these dilemmas, therefore, policy makers and responsible marketers should be encouraged to think creatively about ways of fostering reliance on the appropriate system. Notes 1. Current formulations of the HSM and ELM allow for simultaneous operation of heuristic and systematic processing (Chen and Chaiken 1999; Petty and Wegener 1999); however, the overwhelming majority of applications have represented the two routes as a continuum. 2. In discussing these findings, Novemsky et al. (2007) go considerably further, suggesting that the subjective experience of difficulty generating reasons can suitably account for prior work explaining choice conflict in terms of the decision context or attributes of the options. 3. Although the effects of interactivity on brand intentions were uniformly positive in the Schlosser (2003) study, brand attitudes were affected by the fit between interactivity information and users’ goals (browsing versus searching). Fit appears to induce greater cognitive elaboration, enhancing the scrutiny of information (which was generally favorable in this study). In contrast, interactivity appears to affect intentions more directly by allowing users to simulate themselves using the product. 4. Affect is not necessarily maladaptive in the risk-as-feelings framework. The model distinguishes between anticipatory emotions, which are experienced during the decision process, and anticipated emotions, which capture expected feelings at the culmination of a decision; only the former are deemed inappropriate as a basis for deciding. 5. Intriguingly, Darke, Chattopadhyay, and Ashworth (2006) use a longitudinal approach to demonstrate that for important (presumably high-elaboration) purchases, affective responses immediately after purchase predict both immediate and long-term satisfaction.

References Aggarwal, Pankaj, and Sharmistha Law. 2005. “Role of Relationship Norms in Processing Brand Information.” Journal of Consumer Research 32 (3), 453–64.

32

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

Ahluwalia, Rohini, and Robert E. Burnkrant. 2004. “Answering Questions About Questions: A Persuasion Knowledge Perspective for Understanding the Effects of Rhetorical Questions.” Journal of Consumer Research 31 (1), 26–42. Alter, Adam L., Daniel M. Oppenheimer, Nicholas Epley, and Rebecca N. Eyre. 2007. “Overcoming Intuition: Metacognitive Difficulty Activates Analytic Reasoning.” Journal of Experimental Psychology: General 136 (4), 569–76. Anderson, Norman H. 1971. “Integration Theory and Attitude Change.” Psychological Review 78 (3), 171–206. Areni, Charles S., M. Elizabeth Ferrell, and James B. Wilcox. 2000. “The Persuasive Impact of Reported Group Opinions on Individuals Low vs. High in Need for Cognition: Rationalization vs. Biased Elaboration?” Psychology & Marketing 17 (10), 855–75. Avnet, Tamar, and Michel Tuan Pham. 2007. “Metacognitive and Nonmetacognitive Reliance on Affect as Information in Judgment.” Working paper. Bakamitsos, Georgios A. 2006. “A Cue Alone or a Probe to Think? The Dual Role of Affect in Product Evaluations.” Journal of Consumer Research 33 (3), 403–12. Banaji, Mahzarin R., Kristi M. Lemm, and Siri J. Carpenter. 2004. “The Social Unconscious.” In Social Cognition, ed. Marilynn B. Brewer and Miles Hewstone. Malden, MA: Blackwell, 28–53. Baumeister, Roy F. 2002. “Yielding to Temptation: Self-Control Failure, Impulsive Purchasing, and Consumer Behavior.” Journal of Consumer Research 28 (4), 670–76. Bechara, Antoine, Hanna Damasio, Daniel Tranel, and Antonio R. Damasio. 1997. “Deciding Advantageously Before Knowing the Advantageous Strategy.” Science 275 (5304), 1293–94. Bond, Samuel D. 2008. “Intuition vs. Reason in the Evaluation of Hedonic and Functional Products.” Working paper. Bond, Samuel D., James R. Bettman, and Mary Frances Luce. 2008. “‘Follow Your Gut’ or ‘Use Your Head’? Malleability in Lay Theories of Dual-Systems Processing.” Working paper. Brewer, Marilynn B. 1988. “A Dual Process Model of Impression Formation.” In Advances in Social Cognition, ed. Thomas K. Srull and Robert S. Wyer, Jr., 1–36. Hillsdale, NJ; England: Lawrence Erlbaum. Briñol, Pablo, Richard E. Petty, and Zakary L. Tormala. 2004. “Self-Validation of Cognitive Responses to Advertisements.” Journal of Consumer Research 30 (4), 559–73. Broniarczyk, Susan M., and Joseph W. Alba. 1994. “Theory Versus Data in Prediction and Correlation Tasks.” Organizational Behavior and Human Decision Processes 57 (1), 117–39. Brown, Christina L., and Aradhna Krishna. 2004. “The Skeptical Shopper: A Metacognitive Account for the Effects of Default Options on Choice.” Journal of Consumer Research 31 (3), 529–39. Brunel, Frederic F., Brian C. Tietje, and Anthony G. Greenwald. 2004. “Is the Implicit Association Test a Valid and Valuable Measure of Implicit Consumer Social Cognition?” Journal of Consumer Psychology 14 (4), 385–404. Cacioppo, John T., and Richard E. Petty. 1982. “The Need for Cognition.” Journal of Personality and Social Psychology 42 (1), 116–31. Cacioppo, John T., Richard E. Petty, and Katherine J. Morris. 1983. “Effects of Need for Cognition on Message Evaluation, Recall, and Persuasion.” Journal of Personality and Social Psychology 45 (4), 805–18. Chaiken, Shelly. 1980. “Heuristic Versus Systematic Information Processing and the Use of Source Versus Message Cues in Persuasion.” Journal of Personality and Social Psychology 39 (5), 752–66. Chaiken, Shelly, and Yaacov Trope. 1999. Dual-Process Theories in Social Psychology. New York: Guilford Press. Chen, Serena, and Shelly Chaiken. 1999. “The Heuristic-Systematic Model in its Broader Context.” In DualProcess Theories in Social Psychology, ed. S. Chaiken and Y. Trope, 73-96. New York: Guilford Press. Cohen, Joel B., and C.S. Areni. 1991. “Affect and Consumer Behavior.” In Handbook of Consumer Behavior, ed. T.S. Robertson and H.H. Dassarjian. Englewood Cliffs, NJ: Prentice Hall. Cohen, Joel B., and Americus Reed II. 2006. “A Multiple Pathway Anchoring and Adjustment (MPAA) Model of Attitude Generation and Recruitment.” Journal of Consumer Research 33 (1), 1–15. Csikszentmihalyi, Mihaly. 1990. Flow: The Psychology of Optimal Experience. New York: Harper & Row. ———. 2000. “The Costs and Benefits of Consuming.” Journal of Consumer Research 27 (2), 267–72. Damasio, A.R. 1994. Descartes’ Error: Emotion, Reason, and the Human Brain. New York: Putnam. Danziger, Shai, Simone Moran, and Vered Rafaely. 2006. “The Influence of Ease of Retrieval on Judgment as a Function of Attention to Subjective Experience.” Journal of Consumer Psychology 16 (2), 191–95.

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

33

Darke, Peter R., Amitava Chattopadhyay, and Laurence Ashworth. 2006. “The Importance and Functional Significance of Affective Cues in Consumer Choice.” Journal of Consumer Research 33 (3), 322–28. Darke, Peter R., and Robin J.B. Ritchie. 2007. “The Defensive Consumer: Advertising Deception, Defensive Processing, and Distrust.” Journal of Marketing Research (JMR) 44 (1), 114–27. Devine, Patricia G. 1989. “Stereotypes and Prejudice: Their Automatic and Controlled Components.” Journal of Personality and Social Psychology 56 (1), 5–18. Dhar, Ravi, Stephen M. Nowlis, and Steven J. Sherman. 2000. “Trying Hard or Hardly Trying: An Analysis of Context Effects in Choice.” Journal of Consumer Psychology 9 (4), 189–200. Dijksterhuis, Ap. 2004. “Think Different: The Merits of Unconscious Thought in Preference Development and Decision Making.” Journal of Personality and Social Psychology 87 (5), 586–98. Dijksterhuis, Ap, Pamela K. Smith, Rick B. Van Baaren, and Daniel H. J. Wigboldus. 2005. “The Unconscious Consumer: Effects of Environment on Consumer Behavior.” Journal of Consumer Psychology 15 (3), 193–202. Drolet, Aimee, and Mary Frances Luce. 2004. “The Rationalizing Effects of Cognitive Load on EmotionBased Trade-Off Avoidance.” Journal of Consumer Research 31 (1), 63–77. Eagly, Alice H., and Shelly Chaiken. 1993. The Psychology of Attitudes. Orlando, FL: Harcourt Brace Jovanovich College Publishers. Epstein, Seymour. 1991. “Cognitive-Experiential Self-Theory: An Integrative Theory of Personality.” In The Relational Self: Theoretical Convergences in Psychoanalysis and Social Psychology, ed. Rebecca C. Curtis, 111–37. New York: Guilford Press. ———. (2003). “Cognitive-Experiential Self-Theory of Personality.” In Handbook of Psychology: Personality and Social Psychology, Vol. 5, eds. T. Millon & M. J. Lerner, 159–84. Hoboken, NJ: John Wiley & Sons. Epstein, Seymour, Abigail Lipson, Carolyn Holstein, and Eileen Huh. 1992. “Irrational Reactions to Negative Outcomes: Evidence for Two Conceptual Systems.” Journal of Personality and Social Psychology, 62 (2), 328–339. Epstein, Seymour, Rosemary Pacini, Veronika Denes-Raj, and Harriet Heier. 1996. “Individual Differences in Intuitive-Experiential and Analytical-Rational Thinking Styles.” Journal of Personality and Social Psychology 71 (2), 390–405. Erdem, Tulin, and Joffre Swait. 2004. “Brand Credibility, Brand Consideration, and Choice.” Journal of Consumer Research 31 (1), 191–98. Escalas, Jennifer Edson. 2004. “Narrative Processing: Building Consumer Connections to Brands.” Journal of Consumer Psychology 14 (1/2), 168–80. Fang, Xiang, Surendra Singh, and Rohini Ahluwalia. 2007. “An Examination of Different Explanations for the Mere Exposure Effect.” Journal of Consumer Research 34 (1), 97–103. Fazio, Russel H., David M. Sanbonmatsu, Mark C. Powell, and Frank R. Kardes. 1986. “On the Automatic Activation of Attitudes.” Journal of Personality and Social Psychology, 50 (2), 229–38. Fazio, Russel H., and Michael A. Olson. 2003. “Implicit Measures in Social Cognition Research: Their Meaning and Uses.” Annual Review of Psychology 54, 297–27. Fenigstein, Allan, Michael F. Scheier, and Arnold H. Buss. 1975. “Public and Private Self-Consciousness: Assessment and Theory.” Journal of Consulting and Clinical Psychology 43 (4), 522–27. Fischhoff, Baruch, Paul Slovic, and Sarah Lichtenstein. 1978. “Fault Trees: Sensitivity of Estimated Failure Probabilities to Problem Representation.” Journal of Experimental Psychology: Human Perception and Performance 4 (2), 330–44. Fiske, S.T., and S.L. Nueberg. 1988. “A Continuum Model of Impression Formation: From Category-Based to Individuating Processes as a Function of Information, Motivation, and Attention.” Advances in Experimental Social Psychology 23, 1–108. Forgas, Joseph P. 1995. “Mood and Judgment: The Affect Infusion Model (AIM).” Psychological Bulletin 117 (1), 39–66. Friestad, Marian, and Peter Wright. 1994. “The Persuasion Knowledge Model: How People Cope with Persuasion Attempts.” Journal of Consumer Research 21 (1), 1–31. Gawronski, Bertram, and Galen V. Bodenhausen. 2006. “Associative and Propositional Processes in Evaluation: An Integrative Review of Implicit and Explicit Attitude Change.” Psychological Bulletin 132 (5), 692–731. Gawronski, Bertram, Wilhelm Hofmann, and Christopher J. Wilbur. 2006. “Are ‘Implicit’ Attitudes Unconscious?” Consciousness and Cognition: An International Journal 15 (3), 485–99.

34

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

Gawronski, Bertram, Etienne P. LeBel, and Kurt R. Peters. 2007. “What Do Implicit Measures Tell Us? Scrutinizing the Validity of Three Common Assumptions.” Perspectives on Psychological Science 2 (2), 181–93. Gigerenzer, Gerd. 2007. Gut Feelings: The Intelligence of the Unconscious. New York: Penguin Group. Gilbert, Daniel T., and J. Gregory Hixon. 1991. “The Trouble of Thinking: Activation and Application of Stereotypic Beliefs.” Journal of Personality and Social Psychology 60 (4), 509–17. Gilbert, Daniel T., and Patrick S. Malone. 1995. “The Correspondence Bias.” Psychological Bulletin 117 (1), 21–38. Gilbert, Daniel T., Brett W. Pelham, and Douglas S. Krull. 1988. “On Cognitive Busyness: When Person Perceivers Meet Persons Perceived.” Journal of Personality and Social Psychology 54 (5), 733–40. Gladwell, Malcolm. 2005. Blink: The Power of Thinking Without Thinking. New York: Little, Brown. Green, Melanie C., and Timothy C. Brock. 2000. “The Role of Transportation in the Persuasiveness of Public Narratives.” Journal of Personality and Social Psychology 79 (5), 701–21. Greenwald, Anthony G., and Mahzarin R. Banaji. 1995. “Implicit Social Cognition: Attitudes, Self-Esteem, and Stereotypes.” Psychological Review, 102(1), 4-27. Herr, Paul M., and Christine M. Page. 2004. “Asymmetric Association of Liking and Disliking Judgments: So What’s Not to Like?” Journal of Consumer Research 30 (4), 588–601. Hofmann, Wilhelm, Bertram Gawronski, Tobias Gschwendner, Huy Le, and Manfred Schmitt. 2005. “A MetaAnalysis on the Correlation Between the Implicit Association Test and Explicit Self-Report Measures.” Personality and Social Psychology Bulletin 31 (10), 1369–85. Hogarth, Robin M. 2001. Educating Intuition. Chicago: University of Chicago Press. Isen, Alice M., Kimberly A. Daubman, and Gary P. Nowicki. 1987. “Positive Affect Facilitates Creative Problem Solving.” Journal of Personality and Social Psychology 52 (6), 1122–31. Janiszewski, Chris, and Elise Chandon. 2007. “Transfer-Appropriate Processing, Response Fluency, and the Mere Measurement Effect.” Journal of Marketing Research (JMR) 44 (2), 309–23. Janiszewski, Chris, and Tom Meyvis. 2001. “Effects of Brand Logo Complexity, Repetition, and Spacing on Processing Fluency and Judgment.” Journal of Consumer Research 28 (1), 18–32. Jewell, Robert D., and H. Rao Unnava. 2004. “Exploring Differences in Attitudes Between Light and Heavy Brand Users.” Journal of Consumer Psychology 14 (1/2), 75–80. Johar, Gita V. 1995. “Consumer Involvement and Deception from Implied Advertising Claims.” Journal of Marketing Research 32 (3), 267–79. Kahneman, Daniel. 2003. “A Perspective on Judgment and Choice: Mapping Bounded Rationality.” American Psychologist 58 (9), 697–720. Kahneman, Daniel, and Shane Frederick. 2002. “Representativeness Revisited: Attribute Substitution in Intuitive Judgment.” In Heuristics and Biases: The Psychology of Intuitive Judgment, ed. Thomas Gilovich, Dale Griffin, and Daniel Kahneman, 49–81. New York: Cambridge University Press. Kang, Yong-Soon, and Paul M. Herr. 2006. “Beauty and the Beholder: Toward an Integrative Model of Communication Source Effects.” Journal of Consumer Research 33 (1), 123–30. Kardes, Frank R. 2006. “When Should Consumers and Managers Trust Their Intuition?” Journal of Consumer Psychology 16 (1), 20–24. Kardes, Frank R., Maria L. Cronley, James J. Kellaris, and Steven S. Posavac. 2004. “The Role of Selective Information Processing in Price-Quality Inference.” Journal of Consumer Research 31 (2), 368–74. Keller, Kevin L. 1993. “Conceptualizing, Measuring, and Managing Customer-Based Brand Equity.” Journal of Marketing 57 (1), 1–22. Kirmani, Amna, and Margaret C. Campbell. 2004. “Goal Seeker and Persuasion Sentry: How Consumer Targets Respond to Interpersonal Marketing Persuasion.” Journal of Consumer Research 31 (3), 573–82. Kirkpatrick, Lee A., and Seymour Epstein. 1992. “Cognitive-Experiential Self-Theory and Subjective Probability: Further Evidence for Two Conceptual Systems.” Journal of Personality and Social Psychology, 63 (4), 534–44. Kivetz, Ran, and Anat Keinan. 2006. “Repenting Hyperopia: An Analysis of Self-Control Regrets.” Journal of Consumer Research 33 (2), 273–82. Kramer, Roderick M. 1994. “The Sinister Attribution Error: Paranoid Cognition and Collective Distrust in Organizations.” Motivation and Emotion 18 (2), 199–230. Kunda, Ziva, and Paul Thagard. 1996. “Forming Impressions from Stereotypes, Traits, and Behaviors: A Parallel-Constraint-Satisfaction Theory.” Psychological Review, 103 (2), 284–308. Lazarus, Richard S. 1982. “Thoughts on the Relations Between Emotion and Cognition.” American Psychologist 37 (9), 1019–24.

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

35

Lee, Angela Y., and Aparna A. Labroo. 2004. “The Effect of Conceptual and Perceptual Fluency on Brand Evaluation.” Journal of Marketing Research (JMR) 41 (2), 151–65. Levin, Irwin P., Gary J. Gaeth, Judy Schreiber, and Marco Lauriola. 2002. “A New Look at Framing Effects: Distribution of Effect Sizes, Individual Differences, and Independence of Types of Effects.” Organizational Behavior and Human Decision Processes 88 (1), 411–29. Lewis, Michael. 1997. “The Self in Self-Conscious Emotions.” In The Self Across Psychology: Self-Recognition, Self-Awareness, and the Self Concept, ed. Joan Gay Snodgrass and Robert L. Thompson, 119–42. New York: New York Academy of Sciences. Loewenstein, George F., Elke U. Weber, Christopher K. Hsee, and Ned Welch. 2001. “Risk as Feelings.” Psychological Bulletin 127 (2), 267–86. MacInnis, Deborah J., and Linda L. Price. 1987. “The Role of Imagery in Information Processing: Review and Extensions.” Journal of Consumer Research 13 (4), 473–91. Maheswaran, Durairaj, and Joan Meyers-Levy. 1990. “The Influence of Message Framing and Issue Involvement.” Journal of Marketing Research 27 (3), 361–67. Main, Kelley J., Darren W. Dahl, and Peter R. Darke. 2007. “Deliberative and Automatic Bases of Suspicion: Empirical Evidence of the Sinister Attribution Error.” Journal of Consumer Psychology 17 (1), 59–69. Maison, Dominika, Anthony G. Greenwald, and Ralph H. Bruin. 2004. “Predictive Validity of the Implicit Association Test in Studies of Brands, Consumer Attitudes, and Behavior.” Journal of Consumer Psychology 14 (4), 405–15. Martin, Leonard L. 1986. “Set/Reset: Use and Disuse of Concepts in Impression Formation.” Journal of Personality and Social Psychology 51 (3), 493–504. Martin, Leonard L., John J. Seta, and Rick A. Crelia. 1990. “Assimilation and Contrast as a Function of People’s Willingness and Ability to Expend Effort in Forming an Impression.” Journal of Personality and Social Psychology 59 (1), 27–37. Martin, Leonard L., David W. Ward, John W. Achee, and Robert S. Wyer. 1993. “Mood as Input: People Have to Interpret the Motivational Implications of Their Moods.” Journal of Personality and Social Psychology 64 (3), 317–26. Mathwick, Charla, Naresh Malhotra, and Edward Rigdon. 2001. “Experiential Value: Conceptualization, Measurement and Application in the Catalog and Internet Shopping Environment.” Journal of Retailing 77 (1), 39–56. Mathwick, Charla, and Edward Rigdon. 2004. “Play, Flow, and the Online Search Experience.” Journal of Consumer Research 31 (2), 324–32. McClelland, James L., Bruce L. McNaughton, and Randall C. O’Reilly. 1995. “Why There Are Complementary Learning Systems in the Hippocampus and Neocortex: Insights from the Successes and Failures of Connectionist Models of Learning and Memory.” Psychological Review 102 (3), 419–57. Meyers-Levy, Joan, and Durairaj Maheswaran. 2004. “Exploring Message Framing Outcomes When Systematic, Heuristic, or Both Types of Processing Occur.” Journal of Consumer Psychology 14 (1/2), 159–67. Morwitz, Vicki G., Eric Johnson, and David Schmittlein. 1993. “Does Measuring Intent Change Behavior?” Journal of Consumer Research 20 (1), 46–61. Nisbett, R., and L. Ross. 1980. Human Inference: Strategies and Shortcomings of Social Judgment. Englewood Cliffs, NJ: Prentice Hall. Novak, Thomas P., and Donna L. Hoffman. 2007. “Consumer Thinking Style, Task Congruence, and Performance: New Measures of Task-Specific Experiential and Rational Cognition.” Working paper. Novak, Thomas P., Donna L. Hoffman, and Adam Duhachek. 2003. “The Influence of Goal-Directed and Experiential Activities on Online Flow Experiences.” Journal of Consumer Psychology 13 (1), 3–16. Novemsky, Nathan, Ravi Dhar, Norbert Schwarz, and Itamar Simonson. 2007. “Preference Fluency in Choice.” Journal of Marketing Research (JMR) 44 (3), 347–56. Nowlis, Stephen M., and Baba Shiv. 2005. “The Influence of Consumer Distractions on the Effectiveness of Food-Sampling Programs.” Journal of Marketing Research (JMR) 42 (2), 157–68. Pacini, Rosemary, and Seymour Epstein. 1999. “The Relation of Rational and Experiential Information Processing Styles to Personality, Basic Beliefs, and the Ratio-Bias Phenomenon.” Journal of Personality and Social Psychology 76 (6), 972–87. Park, C. Whan, and S. Mark Young. 1986. “Consumer Response to Television Commercials: The Impact of Involvement and Background Music on Brand Attitude Formation.” Journal of Marketing Research 23 (1), 11–24. Pennington, Nancy, and Reid Hastie. 1986. “Evidence Evaluation in Complex Decision Making.” Journal of Personality and Social Psychology 51 (2), 242–58.

36

SAMUEL D. BOND, JAMES R. BETTMAN, AND MARY FRANCES LUCE

Peters, Ellen. 2006. “The Functions of Affect in the Construction of Preferences.” In The Construction of Preference, ed. Sarah Lichtenstein and Paul Slovic. New York: Cambridge University Press, 454–63. Peters, Ellen, Thomas M. Hess, Daniel Vastfjall, and Corinne Auman. 2007. “Adult Age Differences in Dual Information Processes: Implications for the Role of Affective and Deliberative Processes in Older Adults’ Decision Making.” Perspectives on Psychological Science 2 (1), 1–23. Petrova, Petia K., and Robert B. Cialdini. 2005. “Fluency of Consumption Imagery and the Backfire Effects of Imagery Appeals.” Journal of Consumer Research 32 (3), 442–52. Petty, Richard E., and John T. Cacioppo. 1981. Attitudes and Persuasion: Classic and Contemporary Approaches. Dubuque, IA: Brown. Petty, Richard E., John T. Cacioppo, and David Schumann. 1983. “Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement.” Journal of Consumer Research 10 (2), 135–46. Petty, Richard E., and Duane T. Wegener. 1999. “The Elaboration Likelihood Model: Current Status and Controversies.” In Dual-Process Theories in Social Psychology, ed. Shelly Chaiken and Yaacov Trope, 37–72. New York: Guilford Press. Petty, Richard E., Pablo Briñol, and Zakary L. Tormala. 2002. “Thought Confidence as a Determinant of Persuasion: The Self-Validation Hypothesis.” Journal of Personality and Social Psychology 82 (5), 722–41. Petty, Richard E., Zachary L. Tormala, Pablo Briñol, and W. Blair G. Jarvis. 2006. “Implicit Ambivalence from Attitude Change: An Exploration of the PAST Model.” Journal of Personality and Social Psychology 90 (1), 21–41. Pham, Michel Tuan. 1998. “Representativeness, Relevance, and the Use of Feelings in Decision Making.” Journal of Consumer Research 25 (2), 144–59. ———. 2004. “The Logic of Feeling.” Journal of Consumer Psychology 14 (4), 360–69. Priester, Joseph R., and Richard E. Petty. 1995. “Source Attributions and Persuasion: Perceived Honesty as a Determinant of Message Scrutiny.” Personality and Social Psychology Bulletin 21 (6), 637–54. Raghubir, Priya, and Aradhna Krishna. 1996. “As the Crow Flies: Bias in Consumers’ Map-Based Distance Judgments.” Journal of Consumer Research 23 (1), 26–39. Raven, J.C. 1962. Advanced Progressive Matrices (Set II). London: H.K. Lewis. Rosa, José Antonio, and Alan J. Malter. 2003. “E-(Embodied) Knowledge and E-Commerce: How Physiological Factors Affect Online Sales of Experiential Products.” Journal of Consumer Psychology 13 (1/2), 63. Rydell, Robert J., and Allen R. McConnell. 2006. “Understanding Implicit and Explicit Attitude Change: A Systems of Reasoning Analysis.” Journal of Personality and Social Psychology 91 (6), 995–1008. Scherer, Klaus R., Angela Schorr, and Tom Johnstone. 2001. Appraisal Processes in Emotion: Theory, Methods, Research. New York: Oxford University Press. Schlosser, Ann E. 2003. “Experiencing Products in the Virtual World: The Role of Goal and Imagery in Influencing Attitudes Versus Purchase Intentions.” Journal of Consumer Research 30 (2), 184–98. ———. 2006. “Learning Through Virtual Product Experience: The Role of Imagery on True Versus False Memories.” Journal of Consumer Research 33 (3), 377–83. Schwarz, Norbert. 2004. “Metacognitive Experiences in Consumer Judgment and Decision Making.” Journal of Consumer Psychology 14 (4), 332–48. Schwarz, Norbert, and Herbert Bless. 1992. “Constructing Reality and Its Alternatives: An Inclusion/Exclusion Model of Assimilation and Contrast Effects in Social Judgment.” In The Construction of Social Judgments, ed. Leonard L. Martin and Abraham Tesser, 217–45. Hillsdale, NJ; England: Lawrence Erlbaum Associates. Schwarz, Norbert, Herbert Bless, Fritz Strack, Gisela Klumpp, Helga Rittenauer-Schatka, and Annette Simons. 1991. “Ease of Retrieval as Information: Another Look at the Availability Heuristic.” Journal of Personality and Social Psychology 61 (2), 195–202. Schwarz, Norbert, and Gerald L. Clore. 1983. “Mood, Misattribution, and Judgments of Well-Being: Informative and Directive Functions of Affective States.” Journal of Personality and Social Psychology 45 (3), 513–23. Shank, R.C., and R.P. Abelson. 1995. “Knowledge and Memory: The Real Story.” In Knowledge and Memory: The Real Story, ed. R.S. Wyer, 1–85. Hillsdale, NJ: Lawrence Erlbaum Associates. Sherman, Steven J. 1980. “On the Self-Erasing Nature of Errors of Prediction.” Journal of Personality and Social Psychology 39 (2), 211–21. Sherman, Steven J., Robert B. Cialdini, Donna F. Schwartzman, and Kim D. Reynolds. 1985. “Imagining Can Heighten or Lower the Perceived Likelihood of Contracting a Disease: The Mediating Effect of Ease of Imagery.” Personality and Social Psychology Bulletin 11 (1), 118–27.

CONSUMER JUDGMENT FROM A DUAL-SYSTEMS PERSPECTIVE

37

Shiv, Baba, and Alexander Fedorikhin. 1999. “Heart and Mind in Conflict: The Interplay of Affect and Cognition in Consumer Decision Making.” Journal of Consumer Research 26 (3), 278–92. ———. 2002. “Spontaneous Versus Controlled Influences of Stimulus-Based Affect on Choice Behavior.” Organizational Behavior and Human Decision Processes 87 (2), 342–70. Silvera, David H., Frank R. Kardes, Nigel Harvey, Maria L. Cronley, and David C. Houghton. 2005. “Contextual Influences on Omission Neglect in the Fault Tree Paradigm.” Journal of Consumer Psychology 15 (2), 117–26. Simmons, Joseph P., and Leif D. Nelson. 2006. “Intuitive Confidence: Choosing Between Intuitive and Nonintuitive Alternatives.” Journal of Experimental Psychology: General 135 (3), 409–28. Sloman, Steven A. 1996. “The Empirical Case for Two Systems of Reasoning.” Psychological Bulletin 119 (1), 3–22. Slovic, Paul, Melissa Finucane, Ellen Peters, and Donald G. MacGregor. 2002. “The Affect Heuristic.” In Heuristics and Biases: The Psychology of Intuitive Judgment, ed. Thomas Gilovich, Dale Griffin, and Daniel Kahneman, 397–420. New York: Cambridge University Press. Smith, Eliot R., and Jamie DeCoster. 2000. “Dual-Process Models in Social and Cognitive Psychology: Conceptual Integration and Links to Underlying Memory Systems.” Personality and Social Psychology Review 4 (2), 108–31. Stanovich, Keith E., and Richard F. West. 1998. “Individual Differences in Rational Thought.” Journal of Experimental Psychology: General 127 (2), 161–88. ———. 2000. “Individual Differences in Reasoning: Implications for the Rationality Debate?” Behavioral and Brain Sciences 23 (5), 645–726. Strack, Fritz, Lioba Werth, and Roland Deutsch. 2006. “Reflective and Impulsive Determinants of Consumer Behavior.” Journal of Consumer Psychology 16 (3), 205–16. Tybout, Alice M., Brian Sternthal, Prashant Malaviya, Georgios A. Bakamitsos, Se-Bum Park, Dawn Iacobucci, and Frank Kardes. 2005. “Information Accessibility as a Moderator of Judgments: The Role of Content Versus Retrieval Ease.” Journal of Consumer Research 32 (1), 76–85. Wang, Jing, and Bobby J. Calder. 2006. “Media Transportation and Advertising.” Journal of Consumer Research 33 (2), 151–62. Wanke, Michaela, Gerd Bohner, and Andreas Jurkowitsch. 1997. “There Are Many Reasons to Drive a BMW: Does Imagined Ease of Argument Generation Influence Attitudes?” Journal of Consumer Research 24 (2), 170–77. Wegener, Duane T., and Richard E. Petty. 1995. “Flexible Correction Processes in Social Judgment: The Role of Naive Theories in Corrections for Perceived Bias.” Journal of Personality and Social Psychology 68 (1), 36-51. ———. 1997. “The Flexible Correction Model: The Role of Naive Theories of Bias in Bias Correction.” In Advances in Experimental Social Psychology, vol. 29, ed. Mark P. Zanna, 141–208. San Diego: Academic Press. Williams, Patti, Gavan J. Fitzsimons, and Lauren G. Block. 2004. “When Consumers Do Not Recognize ‘Benign’ Intention Questions as Persuasion Attempts.” Journal of Consumer Research 31 (3), 540–50. Wilson, Elizabeth J., and Daniel L. Sherrell. 1993. “Source Effects in Communication and Persuasion Research: A Meta-Analysis of Effect Size.” Journal of the Academy of Marketing Science 21 (2), 101–12. Wilson, Timothy D., Samuel Lindsey, and Tonya Y. Schooler. 2000. “A Model of Dual Attitudes.” Psychological Review 107 (1), 101–26. Wilson, Timothy D., Douglas J. Lisle, Jonathan W. Schooler, and Sara D. Hodges. 1993. “Introspecting About Reasons Can Reduce Post-Choice Satisfaction.” Personality and Social Psychology Bulletin 19 (3), 331–39. Winkielman, Piotr, Norbert Schwarz, Tedra A. Fazendeiro, and Rolf Reber. 2003. “The Hedonic Marking of Processing Fluency: Implications for Evaluative Judgment.” In The Psychology of Evaluation: Affective Processes in Cognition and Emotion, ed. Jochen Musch and Karl Christoph Klauer, 189–217. Mahwah, NJ: Lawrence Erlbaum Associates. Wright, Peter. 2002. “Marketplace Metacognition and Social Intelligence.” Journal of Consumer Research 28 (4), 677–82. Wyer, Robert S., Jr., and Thomas K. Srull. 1989. Memory and Cognition in Its Social Context. Hillsdale, NJ; England: Lawrence Erlbaum Associates. Yeung, Catherine W.M., and Robert S. Wyer, Jr. 2004. “Affect, Appraisal, and Consumer Judgment.” Journal of Consumer Research 31 (2), 412–24. Zajonc, Robert B. 1968. “Attitudinal Effects of Mere Exposure.” Journal of Personality and Social Psychology 9 (2, Pt. 2), 1–27.

38

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

CHAPTER 2

CAN YOU SEE THE CHASM? Innovation Diffusion According to Rogers, Bass, and Moore BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Abstract A well-accepted idea among new-products marketing practitioners in the last decade is that the market for new products should be viewed as composed of “early” and “mainstream” markets with a “chasm” in between them. A fundamental premise of such an approach is that there is a communication break, at least to some degree, between the consumers in the early market segment and the mainstream adopters. The aim of this study is to examine to what extent aggregate product growth data, typically used in the diffusion of innovation paradigm, can provide empirical support for the existence of a communication break in the diffusion of innovations. We present three alternative models (Bass, Rogers, and Moore) that are flexible enough to include a partial or complete chasm between the early and mainstream markets, and analyze the sales data for three durable goods: color TVs, room air conditioners, and citizen’s band (CB) radios. We show the existence of a partial communication break in all three markets. Introduction One of the most accepted ideas among new-products marketing practitioners in the last decade is that the market for new products should be viewed as composed of “early” and “mainstream” markets with a “chasm” in between them. The idea was popularized by the Silicon Valley consultant Geoffrey Moore in his two books (Moore 1991, 1995). The basis of Moore’s approach is the premise that the early market adopters are so different from the mainstream market adopters that, without a significant change in marketing strategy, including product offering, the new product has a good chance of falling into the “chasm.” A second important premise is that mainstream market consumers do not communicate with early market adopters, or that there is a discontinuity in the diffusion process. The latter premise contrasts with diffusion theory, which views diffusion as a continuous communication process driving the sales of the new product (Rogers 1995). Moore’s idea of a communications break is the opposite of what can be considered the conventional wisdom in marketing: that earlier adopters are often also opinion leaders and thus are a worthwhile investment in terms of acquisition costs. Not only they are not opinion leaders, claims Moore, but they do not affect the later (mainstream) market at all. If acquiring them is worthwhile, it will be only because of the revenues they generate directly and not because of the chain reaction that they will generate. 38

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

39

The chasm phenomenon was later formalized and explored by Mahajan and Muller (1998), Goldenberg, Libai, and Muller (2002), Golder and Tellis (2004), Muller and Yogev (2006), Lehmann and Esteban-Bravo (2006), Karmeshu and Goswami (2001), and Van den Bulte and Joshi (2007). Goldenberg, Libai, and Muller (2002) referred to this phenomenon as a saddle, and defined it as a pattern in which an initial peak predates a trough of sufficient depth and duration, followed by sales that eventually exceed the initial peak. These works did find some support for partial communication breaks between the innovators and mainstream adopters by presenting aggregate as well as agent-based models. The considerable interest in the chasm phenomenon reflects the realization that such a pattern may have a substantial effect on the profitable strategy of firms in its presence, whether it is which customers to target (Mahajan and Muller 1998) or how to optimally plan the marketing mix (Lehmann and Esteban-Bravo 2006). Clearly, more explorations of this intriguing topic are needed to enable marketers better decision making in this regard. The aim of this study is to contrast directly the Moore paradigm with Rogers’s more traditional model of innovation diffusion, and to compare them with the benchmark model of Bass. This comparison will enable us to directly examine the role of innovators as being influential in passing information about the new product. It will also help us to better understand the chasm phenomenon. The rest of the chapter continues as follows: In the next section, we provide the theoretical background for the chasm, and the relationship between early adoption and opinion leadership. In Section 3, we present three alternative models (Bass, Rogers, and Moore), which are flexible enough to include a partial or complete communication break between the early and mainstream markets. In Section 4, we analyze the sales data for three durable goods: color TVs (1954–1963), room air conditioners (1936–1961), and citizen’s band (CB) radios (1958–1982). Section 5 presents the results of the empirical study, and in particular it lends support to the chasm phenomenon, rather than the Bass model or Rogers’s formulation. Section 6 discusses the results and concludes the chapter. Innovators as Influential Rogers’s well-known classification uses the term “innovators” to describe the first 2.5 percent of the population that adopts a given product. The term “innovators” has also been used in a more general sense to describe an early market that includes the first 16 percent of adopters, a group that includes Rogers’s “innovators” and “early adopters” (Summers 1971; Midgley 1977). For his generalizations regarding the nature of adopters, Rogers himself used the term “earlier adopters” (in contrast to “later adopters”), which had been viewed as the early market that combines Rogers’s innovators and early adopters (Midgley 1977). It should be also noted that while Rogers defined innovativeness as the tendency to adopt innovations earlier, a different approach suggests that innovativeness is the degree to which an individual makes an innovation decision independently of the communicated experience of others (Midgley and Dowling 1978). Thus one can think of Rogers’s approach as an ex-post definition, while Midgley and Dowling or Tanny and Derzko (1988) define it ex-ante. Taking the latter approach, while one can expect that there will be a relatively high proportion of consumers who are mainly affected by external influence (or by innovators in this sense) in the early stages of adoption, these “communication innovators” can be found in all stages of the diffusion process. Marketing writings promote the idea that Rogers’s early adopters have a tendency to also be opinion leaders (Solomon 2006). Opinion leaders can be viewed as those who exert disproportionate

40

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

influence on others (Summers 1971). Their influence stems from their status as individuals who are highly informed, respected, or very connected (Watts and Dodds 2007). The major importance of this group is attributed to the fact that they tend to spread information by word of mouth, and that the early majority looks to the early adopters for guidance (Perreault and McCarthy 1996). This view is supported by Rogers, who suggests that early adopters, more than any other group, have a high degree of opinion leadership (Rogers 1995, p. 274). However, Rogers also notes that when social system norms do not favor change, opinion leaders are not especially innovative (Rogers 1995, p. 295). For example, when the norms in systems such as public health care organizations did not favor taking risks, opinion leaders were not earlier adopters of high-uncertainty innovations such as diabetic screening programs (Becker 1970). Empirical marketing findings on this subject, most of them from the 1960s and the 1970s, tend to provide support for the idea that earlier adopters tend to be also opinion leaders more than others (e.g., Bell 1963; Engel, Blackwell, and Kegerreis 1969; Midgley 1977; Gatignon and Robertson 1985). This tendency was found to differ among various product categories, and many of the findings relate to innovations such as supermarket goods and fashion products, which are not considered discontinuous innovations (e.g., Summers 1971; Midgley 1977). More recent work, however, supports this pattern also for durables such as computers and cars (Grewal and Mehta 2000). Diffusion models following the Bass model generally did not attribute a higher level of opinion leadership to any group in the adoption process. Mahajan, Muller, and Srivastava (1990) demonstrate that the effect of external influence is expected to be higher in early parts of the diffusion process. However, this is not attributed to a high degree of opinion leadership by individual adopters, but to the small number of previous adopters that can affect later adopters via word of mouth. Otherwise, the Bass model assumes a homogeneous influence level of all adopters over each other. Thus, diffusion theory studies traditionally support the idea of opinion leadership by innovators. In contrast, recent writings suggest that at least regarding high-tech products, not only are innovators not opinion leaders, but they do not have personal influence at all over later adopters (Moore 1991, 1995). Moore builds on Rogers’s normal diffusion curve and his division into adopter categories along the curve to explain how new products spread in the market. However, he sees a discontinuity in the process after about 16 percent of the population adopts the innovation. The social process of contagion is broken at this point, because the later adopters (mainstream market) refuse to rely for information on the earlier adopters (the early market). Moore’s explanation for this phenomenon centers on the difference between adopters in the early market and those in the mainstream market—especially that of the group preceding the chasm—the early adopters (“visionaries”) versus the group that adopts after the chasm, or the early majority (“pragmatists”). Early adopters are interested more in the new technology itself and the way they can use it as a window of opportunity for their company or themselves. They are influenced by people with the same interest in technology across other industries, and they agree to purchase products that may be expensive and may not be complete in terms of support, compatibility with existing infrastructure, and reliability. The pragmatists, which are the bulk of all technology infrastructure purchasers, are not especially interested in the technology itself and are not willing to “bet on the outcome.” They expect to get a 100 percent solution that is reasonably priced and already reliable and effective, has support, and is compatible with their infrastructure. Because of the difference between the solution they demand compared to that of the visionaries, pragmatists tend to be influenced only by other pragmatists, or in Moore’s words (1995, p. 18), “these two groups, although adjacent on

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

41

the adoption life cycle, are so different in terms of underlying values as to make communication between them almost impossible.” Thus, adoptions of new products can “fall into the chasm” between the visionaries and the pragmatists, causing the product to fail to enter the mainstream market. This failure can be avoided by a considerable change in the marketing strategy that will focus on tailoring the product attributes to the needs of the mainstream market, or reallocating the amount of resources that should be invested in innovators versus the mainstream market (Mahajan and Muller 1998). A “dual market” view has been suggested by academic researchers in various disciplines (for behavioral theories that support a dual market view, see a review in Van den Bulte and Joshi 2007). In communications, Rogers (1986) suggests that for interactive innovations (innovations with which people communicate), diffusion can be analyzed as a two-phase process, that is, before and after a critical mass is reached. In the organizational science literature, Cool, Dierickx, and Szulanski (1997) broadened the idea of markets before and after critical mass is reached to include diffusion of innovations within organizations. The technology management literature (e.g., Anderson and Tushman 1990; Utterback 1994) has focused on the effect of a standard (or a “dominant design”) for the evolution of technologies, and it views various sales processes before and after the dominant design has been established. Regarding geography, Brown (1981) links the early market to the time it takes to set outlets for the distribution of a new innovation. A common idea behind all of these approaches is that the initial product offered to consumers is quite different from that offered in the later phase, and thus we might need to examine two markets instead of one. It is clear that the dual-markets approach to diffusion has important implications for our understanding of new-product marketing. For example, it has been suggested that, following a dual-markets approach to diffusion, major changes should be made in the way new-product management is practiced (Mahajan and Muller 1998; Moore 1991). Yet important questions are still left open after taking current knowledge into account. For example, Moore’s theory of the dual market applies mostly to what he defines as “high-tech” (especially computer-related) products and to business-to-business sales situations. Yet the rationale presented for the dual-market view—that of changing product benefits that fit the “mainstream market” only at a later stage—may apply to many durables, including those sold to individuals. Another major question that stems from the dual-market approach relates to the communication patterns between the two groups. Almost by definition, if there are two markets, then the communication between members of the different markets should be different from that among each market’s members. A communication break between the two markets can clearly lead to a “slowdown” of the diffusion pattern, and the extent of the break can thus have a strong effect on the diffusion pattern. According to Moore, there is a complete communications break between the early and mainstream markets. Is a complete break the norm, or can there be only a “partial break” in communications? That is, might the early market have some limited influence on the main market rather than no influence at all? A third issue that relates to Moore’s theory deals with the role of the “innovators”—those who were willing to adopt the product in its early version. What happened to those who were willing to adopt initially but did not do so? Because the diffusion process takes its time, some members of the early market might not adopt the innovation before the mainstream market begins to adopt it. Would these early market members join the communication pattern of the mainstream? Would they be left out as nonpurchasers? Lately there have been some efforts to address some of these issues by modeling the chasm phenomenon directly. Thus Goldenberg, Libai, and Muller (2002), Golder and Tellis (2004), Muller

42

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

and Yogev (2006), and Van den Bulte and Joshi (2007) indeed found empirical support for a saddle occurring in many consumer durable products, mainly but not exclusively consumer electronics. Note that an alternative explanation as to the appearance of saddles was given by Golder and Tellis (2004) as well as Chandrasekaran and Tellis (2006), who showed that the saddle phenomenon could be explained using the informational cascade theory, where the term “informational cascade” is defined as the tendency of individuals to adopt a behavior based on the value of the signal they derive from the behavior of previous adopters (Golder and Tellis 2004). In this chapter we examine the above questions using data on the adoption of a number of durable goods. Specifically, we want to investigate if diffusion models that are tailored to the dual- market approach can help us answer the following: 1. Would we be able to identify the chasm phenomenon for durable goods sold to final consumers, including those that were not traditionally considered “high-tech”? 2. Can we determine the “opinion leadership” effect of the early market on the mainstream one? That is, would there be a complete communications break between the early and mainstream markets, or we would we observe only a partial break? 3. What happens to members of the early market who have not yet adopted before the mainstream market entry? Would they continue to adopt as a separate segment, or would they then join the mainstream? We examine these main research issues using data on the diffusion of three product categories: color televisions, room air conditioners, and citizen’s band (CB) radios. In the next section, we develop three different models that correspond to three different points of view regarding the effect of the early adopters on the late adopters. Life According to Rogers, Bass, and Moore Before presenting the three models, we first present the Bass model, as it constitutes the main building block for all three models. The Bass model (Bass 1969) is the best-known model in diffusion research in marketing. Since its publication in Management Science, it has been cited over 600 times, and it forms the basis for nearly all the models reviewed in this paper. See also the review papers by Mahajan, Muller, and Bass (1990, 1995), Meade and Islam (1998, 2006), Hauser, Tellis and Griffin (2006), Chandrasekaran and Tellis (2007), and a recent review paper by Muller, Peres, and Mahajan (2007). Assume a market with potential N. At each point in time, new adopters join the market as a result of external influence (p) and internal influence (q). The parameter p captures the activities of firms in the market, mainly advertising, and any other time-invariant element affecting the diffusion, such as the attractiveness of the innovation. The parameter q refers to the magnitude of influence of another single adopter. In the original paper by Bass, as well as later interpretations that regarded diffusion as a theory of interpersonal communications, q represented word of mouth. If A(t) is the cumulative number of adopters at time t, the total probability of adopting at that time point is qA(t)/N. The number of new adopters at time t can be described by the following differential equations:

G$ T$  S  1  $  1  $ GW 1

(1)

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

43

Equation (1) is a nonlinear first-order differential equation, and can be solved analytically using separation of variables. If A(0) = 0, the solution for Equation (1) is given by:

$ W = 1

 − H − S + T W  + T  S H − S + T W

(2)

The Bass model parameters p, q, and m can be estimated from adoption data, usually by using nonlinear least squares (Srinivasan and Mason 1986; Schmittlein and Mahajan 1982). Though numerous studies have estimated the parameters in various industries, the average values of q and p for durable goods were found to be p = 0.03, q = 0.38 (Sultan, Farley, and Lehmann 1990). Estimation issues are also discussed in Jiang, Bass, and Bass (2006), Stremersch and Van den Bulte (2007), Boswijk and Franses (2005), and Van den Bulte and Stremersch (2004). There has been a shift in the previous decades in the emphasis in diffusion research (see Muller, Peres, and Mahajan 2007). Previous research generally described the basic diffusion scenario: a monopoly of a durable good in a single market. These studies laid the foundations by developing tools for parameter estimation, exploring the effect of marketing mix variables, or checking the fit of the Bass model to various product categories. Extensions to the basic diffusion model were proposed, still in the framework of the basic scenario. Current research relies on these foundations to explore market phenomena that go beyond the classical diffusion scenario—service innovations, brand-level diffusion, multinational diffusion, takeoffs and saddles, and complex marketing mix decisions. While past research considered a fully connected, uniform social system, current studies look at cases of heterogeneous social systems. Past research was generally centered at the category level; current research ventures into brand level. Previous research was generally aimed at forecasting; current research places more emphasis on managerial diagnostics. The Bass model, for our purposes, serves as the main building block for the three different models. In order to formulate the three models, we first assume that there are two major groups of adopters: innovators and majority (the terms “early market” for the innovators and “mainstream market” for the majority are also used throughout this chapter). If we consider the well-known classification of Rogers, then we combine the first two adopter categories of innovators and early adopters, denoting them both as innovators. We also combine the early majority and late majority into one group: the majority. Since the laggards segment is usually not a viable marketing target, we drop this group without loss of generality. This is also the approach taken in Muller and Yogev (2006) and Van den Bulte and Joshi (2007). Let the index p denote the innovators segment, and q, the majority. Consider the following definitions: Np is the market potential of the innovators. Nq is the market potential of the majority. p and q are the coefficients of innovation and imitation, where p1 and q1 represent these coefficients for the innovators, and p2 and q2 are the coefficients of innovation and imitation for the majority. I(t) is the number of adopters out of the innovators population. M(t) is the number of adopters out of the majority population.

44

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Figure 2.1

Diffusion According to Rogers

The adoption process in the innovator group progresses through a Bass process as follows:

G,  GW = S + T ,  1 S 1 S − ,

(3)

The initial condition is I(t) = 0. In the majority segment, we have three competing models that depend on the manner in which we treat the innovators once the majority segment commences with purchasing the product. We denote by I* the cumulative number of innovators that have already adopted at the time at which the majority segment begins to adopt. Model 1: Diffusion According to Rogers As we noted earlier, marketing writings promote the idea that Rogers’s early adopters have a tendency to also be opinion leaders in that opinion leaders can be viewed as those who exert disproportionate influence on others. The major importance of this group is attributed to the fact that they tend to spread information by word of mouth, and that the early majority looks to the early adopters for guidance. Thus the view of Rogers is that innovation is a time-dependent construct. Hence, innovators are those customers that adopted the innovation first (the first 16 percent of the population), customers belonging to the early majority segment are the next one-third of the market to adopt, and so on. The full segmentation is given in Figure 2.1. First note that in order to compare the model with the ones by Bass and Moore, we combine the early and late majority segments into one. Under this assumption, the model according to Rogers assumes that at the time the majority segment starts adopting the product, the innovators have terminated purchasing altogether. The segments are defined in terms of their time of adoption, and not in terms of some innate characteristic of innovation tendency. The innovators, as well as previous adopters of the majority segment, influence the consumers in the majority segment. The rate at which the majority interacts with the innovators is given by Aq1. The parameter A is a key parameter in this formulation, which yields a gap or chasm. It is defined as an opinion

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE Figure 2.2

45

Diffusion According to Bass

leadership parameter that measures the degree to which the experience of the innovators carries over to the majority group. If A is zero, no experience is carried over, and the product stops its diffusion with the innovators. This absence of carryover corresponds to a discontinuous gap in the process. If the opinion leadership parameter is relatively large, much of the experience is indeed transferred, and the process might progress in a continuous manner from the innovators to the majority. In Model 1, the equation that governs the diffusion of the product among the majority is thus the following:

G0GW = {T  0  , + 1 T + α T ,  , + 1T } 1T − 0

(4)

Model 2: Diffusion According to Bass While the Bass model was described in detail in the previous section, we emphasize here the difference in the assumptions of the Bass model with that of Moore and Rogers. This is best done with the help of Figure 2.2. The Bass model assumes that at the time the majority segment starts adopting the product, the innovators continue purchasing the product independently of the majority group. However, while not affected by the majority, innovators can themselves affect the majority. In this respect, the model is the closest to the original interpretation of the diffusion process offered by Bass (1969). It also has some similarities to a model suggested by Tanny and Derzko (1988). The segments are defined in terms of the source of information that caused the adoption of the product, either internal or external. As in Model 1, the innovators, as well as previous adopters of the majority segment, influence the consumers in the majority segment. The rate at which the majority interacts with the innovators is given by Aq1, where A is the opinion leadership parameter that measures the degree to which the experience of the innovators carries over to the majority group.

46

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Figure 2.3

Diffusion According to Moore

In Model 2, the equation that governs the diffusion of the product among the majority is thus the following:

G0  GW = { S  + T  0  1 S + 1T + α T ,  1 S + 1T } 1T − 0

(5)

Note that this model has one additional parameter as compared to Model 1, which is the external parameter of the majority. If we were to introduce this parameter in Model 1 as well, a regression analysis performed on Model 1 could not have differentiated this parameter from the term that measures the influence of the innovators on the majority, as both of these terms are constants. One should note that this particular model is precisely equivalent to the models by Muller and Yogev (2006), and Van den Bulte and Joshi (2007) (see Muller, Peres, and Mahajan review paper, 2007). Model 3: Diffusion According to Moore In Model 1, innovators who did not purchase the product by the time the consumers of the majority segment started purchasing the product are still labeled as innovators. They do not participate in the diffusion process, as they have not purchased, nor will they purchase, the product. Model 3 is a variant of this model in which the innovators who, by the time the majority adopts the product, have not yet purchased the product, so are not labeled as innovators any longer, but rather as majority. To describe it more dramatically, by not purchasing by the time majority commences adopting, they have forfeited their status as innovators, and are now indistinguishable from the majority. Another difference from Model 1 is that the diffusion within the majority can start independently of an innovators segment influence. This is allowed because the majority can also be influenced by independent external influence p2. Thus, this model allows a situation of a complete communications break, and comes closest to the original interpretation of Moore. (Model can be described by Figure 2.3.)

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

47

In Model 3, the equation that governs the diffusion of the product among the majority is thus the following:

G0  GW = S  + T  0  1T 1T − 0

(6)

0 7   ,

where A is the opinion leadership parameter that measures the degree to which the experience of the innovators carries over to the majority. Note that M(T*) = !I*, since by the time the majority adopts the product, those innovators who have not yet purchased the product are not labeled as innovators any longer, but rather as majority. Data Analysis In order to find out to what extent aggregate product-growth data, typically used in the diffusion of innovation paradigm, can provide empirical support for the existence of a communication break in the diffusion of innovations, we analyzed the growth data for three durable goods: color TVs (1954–1963), room air conditioners (1936–1961), and citizen’s band (CB) radios (1958–1982). The first two products have been extensively analyzed in diffusion literature, though often using only a restricted time series that did not include early years (Bass 1969; Mahajan, Mason, and Srinivasan 1986; Jain and Rao 1990). Because of the emphasis on early rather than later market sales, we use a longer time series that includes the early diffusion period, based on data from the Electronics Industries Association and Merchandising. Data for CB radio sales were obtained from FCC records. For each product, one early market and three alternative mainstream market models were examined. In addition, we examined four alternatives to the year of the chasm, or the time during which the mainstream market begins to adopt. Thus, overall we conducted 4 × 4 = 16 estimations for each durable good examined. Regardless of the model used for the mainstream market, our approach models the early period using the basic Bass model (see Equation 1). Thus, using NLS, the Bass model has been estimated for the early period, changing the number of data points as the year of the chasm changes. Early period estimation demands caution, since in some cases, unit sales has not reached a clear “peak” prior to the suggested year of the chasm. In the absence of an exogenously defined market potential, the lack of peak is associated with unstable results (Srinivasan and Mason 1986). Possible suggestions for the potential number of users in the early period vary from around 2 percent of market potential (Mahajan, Muller, and Srivastava 1990) to 25 percent (Mahajan, Muller, and Bass 1995). Thus, in cases of unstable results, we restricted the market potential for the early period to no more than 25 percent of the total market potential for the product (derived with the full time series). There is no agreed-upon way to empirically derive the year when main markets begin to adopt (“year of the chasm”) from the sales data. We follow theoretical guidelines regarding the mainstream market entrance, which suggest that the end of the early period market is followed by a strong increase in demand (Moore 1995). An idea similar to Moore’s regarding an early adoption period followed by a discontinuity in demand, which is labeled the “takeoff,” has been suggested in the academic marketing literature (Golder and Tellis 1997; Stremersch and Tellis 2004; Tellis, Stremersch, and Yin 2003). Thus, the takeoff of the diffusion curve may constitute a good candidate for the time of the chasm. Golder and Tellis (1997) have suggested an operationalized method to determine what may constitute a takeoff in the case of durable sales, and their guidelines are used here to select the possible period for the chasm.

48

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Table 2.1

Estimation Results for Color TVs Innovators

1961

1962

Chasm at: p1 q1 Np MSE MPE

0.03 0.45 0.84 0.0005 0.323

0.02 0.33 1.70 0.0005 0.280

1963

1964

0.0005 0.76 10 0.0024 0.441

0.0003 0.91 10 0.0022 0.435

Model 1—Rogers q2 Nq  MSE MPE

0.68 37.46 1.95 0.176 0.383

0.67 37.2 3.47 0.186 0.219

0.65 36.7 1.28 0.198 0.172

0.62 36.2 2.57 0.191 0.118

Model 2—Bass p2 q2 Nq  MSE MPE

0.2 0.99 39.5 8.74 6.373 2.447

0.003 0.99 39.5 10.63 4.335 1.095

0 0.99 38 1.30 0.321 0.219

0 0.99 38 0.97 0.730 0.259

Model 3—Moore p2 q2 Nq  MSE MPE

0.011 0.67 37.46 0 0.176 0.383

0.019 0.65 37.22 0 0.186 0.219

0.032 0.63 36.76 0 0.198 0.172

0.055 0.59 36.21 0 0.191 0.118

While the Golder and Tellis guidelines allow us to choose a specific year for the takeoff, in this study we consider years around it as possible candidates for takeoff. One reason is that it is not clear if the takeoff in sales happens immediately after the mainstream market begins to adopt, or sometime afterward. A sensitivity analysis can help us to support our choice of the year of the chasm: we expect a better fit of the model in the case of the “right” year being chosen for the chasm. Thus, candidates for the chasm were as follows: for color TVs, 1961–1964 (1962 by the Golder and Tellis criteria); for room air conditioners, 1951–1954 (1953 by the Golder and Tellis criteria); and for CB radios, 1973–1976 (1974 by the Golder and Tellis criteria). Thus, regarding each product, we analyzed the three mainstream market models, varying the possible year of the chasm. Results for color TVs are presented in Table 2.1, air conditioners in Table 2.2, and CB radios in Table 2.3. All analyses used the discrete version of the diffusion models and NLS. We present two measures

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

49

Table 2.2

Estimation Results for Room Air Conditioners Innovators

1951

1952

1953

1954

Chasm at: p1 q1 Np MSE MPE

0.00031 0.70 2.17 0.0002 0.468

0.00049 0.57 3.00 0.0002 0.365

0.00031 0.58 4.5 0.0004 0.373

2E-05 0.99 4.5 0.005 0.621

Model 1—Rogers q2 Nq  MSE MPE

0.99 18 3.10 0.335 0.779

0.99 18 3.10 0.258 0.480

0.99 17.8 2.43 0.143 0.248

0.99 16.8 0.87 0.119 0.217

Model 2—Bass p2 q2 Nq  MSE MPE

0 0.99 18 1.25 0.117 0.247

0 0.99 18.2 1.29 0.051 0.193

0 0.99 17.8 0.85 0.033 0.292

0 0.99 16.8 0.45 0.220 0.191

Model 3—Moore p2 q2 Nq  MSE MPE

0.021 0.36 18 0.56 0.033 0.242

0.035 0.32 18.14 0.30 0.028 0.165

0.055 0.27 17.8 0 0.014 0.163

0.070 0.26 16.8 0 0.015 0.065

of fit: Mean Square Error (MSE) and Mean Percent Error (MPE). The latter has an advantage, since MPE, unlike MSE, is a relative measure that should not be influenced by the number of adoptions. In comparison, we can expect a higher square error when sales are very high. Thus, when we compare estimations using time series of different lengths, MPE might be a better error measure. Empirical Results and Implications The Year of the Chasm A better model fit associated with a certain year as the year of the chasm may indicate support for this year as the end of the early market. We note, however, that changing the year of the chasm also

50

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Table 2.3

Estimation Results for CB Radios Innovators Chasm at: p1 q1 Np MSE MPE

1973

1974

1975

0.023 0.28 3.00 0.0015 0.286

0.023 0.23 3.42 0.002 0.310

0.018 0.14 5.12 0.0035 0.374

1976 2.8E-06 0.91 6.73 0.028 0.772

Model 1—Rogers q2 Nq  MSE MPE

0.99 16.92 1.14 0.577 0.927

0.99 16.16 2.32 0.595 0.795

0.99 16.2 6.39 0. 594 0.574

0.67 14.4 1.41 0.369 0.447

Model 2—Bass p2 q2 Nq  MSE MPE

0 0.99 39.61 2.44 2.44 1.63

0 0.99 20.81 3.84 2.32 1.24

0 0.99 17.76 10.19 1.69 0.75

0 0.99 15.48 5.60 0.8 0.33

Model 3—Moore p2 q2 Nq  MSE MPE

0.036 0.99 15.98 0 0.478 0.810

0.070 0.99 15.28 0 0.50 0.738

0.054 0.99 16.2 0.47 0.496 0.563

0.114 0.80 17.27 0.946 0.345 0.457

changes the length of the time series analyzed, both for innovators and for mainstream market models. As it is easier to fit a smaller number of data points, adding data points to the time series may result in a higher mean error. Thus, as the proposed year of chasm is increased, early market error measures may be somehow higher, and mainstream market error may be lower, and vice versa. Therefore, in order to find the year of the chasm, we look for a sharp drop of the error term from one year to the next. This drop should be considerably larger than the drop in other years, making it harder to explain in terms of simply the effect of increasing or decreasing time series length. We first examine the change in error in the case of the innovators market that is similar to all cases. Looking at the innovators’ model for color TVs, 1962 as the year of the chasm clearly gives the best fit in terms of MPE. Using the same criteria, 1952 was chosen as the chasm year for room air conditioners, and 1975 for CB radios.

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

51

We also note that our chasm choice is generally supported by results for the three mainstream models that we examine. In the case of color TVs, the drop in error is sharpest for 1962 for both Model 1 and Model 3. In the case of room air conditioners, 1952 is the year of strongest drop in error for both Model 2 and Model 3. Examining the CB radio data, 1975 is the year of the largest drop in error, regarding all models. This convergence between the various models provides support for the choice of chasm year. Overall, the various models can be seen as generally consistent with certain years (1962 for color TVs, 1952 for air conditioners, 1975 for CB radios) chosen as the chasm year. In the case of some divergence from it (Model 2 for color TVs and Model 1 for air conditioners), the divergence is of one year only. It seems, therefore, that the models presented help to point to which year may be considered “best” for the chasm. Is There a “Best” Model Among the Mainstream Market Models? Taking MSE and MPE as the criteria, and considering the years chosen as the best fit for the chasm as the base for the comparison, Model 3—the Moore formulation—emerges as giving a better fit than the others. Regarding all three products, Model 3 MSE is the lowest. In two out of the three products, the MPE for Model 3 is the lowest (in the case of CB radios, the MPE for Model 1 is lower than that of Model 3, though the difference is not great). Model 1 (with one less parameter) fits better than Model 2 regarding two out of the three products: CB radios and colors TVs. In addition to the MSE and MPE measures, we conducted two other sets of tests to determine which of the three competing models for the majority (using the year of the chasm as indicated above) might be the best model. Our tests were conducted on the data for the three product models that we analyzed. We started with the P test proposed by Davidson and MacKinnon (1981). The P test has the advantage of being a relatively simple method to compare non-nested nonlinear models, and thus has been recommended for use in evaluating competing non-nested models in marketing (Balasubramanian and Jain 1994). However, our analysis results determine that generally the P test could not discriminate among the models, as none of the models was rejected. The only exception was for Model 2 for room air conditioners, which was rejected. The second approach is the “one item out” cross-validation method, which has been shown to be a powerful method for model selection (e.g., Stone 1974; Linhart and Zucchini 1986). In this method, a case (i.e., observation for a single year in our case) is taken out of the data set, and the model is estimated without that case. Then, based on estimation results, a prediction is made regarding the specific year, and the error is measured as the difference between the real value and the predicted value. The procedure is repeated for all cases. The sum of all square errors is a good measure of the fit of the model. Generally, models with a lower sum of square error are considered to better describe the process. The cross-validation method was performed for the durables analyzed and for the three competing models. Results are presented in Table 2.4. As can be seen, Model 3 outperforms Models 1 and 2 in all three cases. This is also consistent with the MSE and MPE results, and can be likewise seen as supporting Model 3 as best describing the diffusion process of the three durables. Thus, the rest of the analysis will focus on the results of Model 3. Model 3 supports the case in which the innovator market ceases to exist with the entry of the mainstream, but the mainstream market might be affected by previous innovators. Also, mainstream market adopters could be influenced by an independent external influence, which is represented by the parameter p2.

52

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Table 2.4

Cross-Validation Results—Sum of Square Error Model 1 Color TVs Room Air Conditioners CB Radios

4.55 3.29 10.3

Model 2 43.87 0.7842 32.80

Model 3 4.45 0.481 7.84

The Magnitude of the Communications Break Moore (1991, 1995) suggests that there is a complete break in communications between early and mainstream markets. If this is the case, then parameter A, which represents the effect of the innovators on mainstream market adopters, should be close to zero. As indicated in Tables 2.1–2.3 for Model 3, the results support the existence of a chasm as described by Moore, though the communications break may not be complete. For color TVs, A is zero, for room air conditioners it is 0.3, and for CB radios it is 0.47. According to this, and in line with Goldenberg, Libai, and Muller (2002), the “opinion leadership” effect of the early market on the mainstream one is low. Thus, to answer the question of what drives the process in the beginning—innovators’ influence or independent external influence—estimation results show that the value of  is close to zero, while p is not, so in this case the mainstream market “takeoff” is attributed to the independent external influence effect, and not to the innovators effect. Discussion Closing the Gap Between Chasm and Diffusion Theories A motivation for our research has been the gap between the chasm theory and the diffusion of innovation paradigm in marketing. We raised three questions, which are the basis for the empirical examination: 1. Would we be able to identify the chasm phenomenon for durable goods sold to final consumers, including those goods that were not traditionally considered “high-tech”? Our results support the chasm phenomenon in the three product categories chosen, which are not associated with the diffusion of computer-related high-tech products in the 1980s and 1990s of the type described by Moore (1991, 1995). We also note that most of Moore’s examples related to products sold in a business-to-business setting. Here we examined products that penetrated the final consumer market. Thus, it may be that the chasm phenomenon is a more general one than the one Moore has described. 2. Can we determine the “opinion leadership” effect of the early market on the mainstream one? That is, would there be a complete communications break between the early and mainstream markets, or we would we observe only a partial break?

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

53

Our results support a communications break between the early and mainstream market. However, this might not necessarily be a complete communications break. The magnitude of the break seems to depend on the specific product category. An interesting implication of these results relates to the connection we see between the move from the early to the mainstream market adoptions, and the takeoff phenomenon. As can be seen, the chasm is near the time of takeoff (as operationalized by Golder and Tellis 1997). However, diffusion theory (Rogers 1995) attributes the takeoff to the rising word-of-mouth (WOM) effect of previous adopters. If previous adopters from the early market do not affect, or have little effect on, the mainstream market, then why is there a takeoff at that point? The answer could lie in the change of market potential that occurs at this point. From a relatively small market potential that includes only the early market, we move to a much larger market potential (which can easily be five to forty times as large). At this time, mainstream market adoption is driven mostly by external influence from independent sources such as advertising and mass media, and “external” influence from the early market. However, even if external influence from the early market is quite small, the independent external influence on a large group can drive a fast rise in sales. As more mainstream market adopters begin to spread WOM of their own, this can become a quite large sales increase as compared to the early market period. 3. What happens to members of the early market who have not yet adopted before the mainstream market entry? Would they continue to adopt as a separate segment, or would they join the mainstream? The support for Model 3, compared to the other two models, implies a case in which, following the chasm, we can no longer distinguish between the part of the early market that has yet not adopted, and the “common” mainstream adopter. What can the above results tell us about the theory of diffusion? Going back to the role of innovators in the diffusion process, the importance of innovators has often been attributed to their opinion leadership effect. However, empirical results regarding innovators as opinion leadership are mixed (Midgley 1977). The Bass model assumed homogenous effect of adopters throughout the adoption process. Moore (1991) suggested that at least for high-tech products, not only are innovators not opinion leaders, but they also do not influence at all. Our results suggest that this phenomenon may exist for consumer durables, though there may be some effect of the early market on the later ones. The results also lead to some interesting observations regarding the conventional wisdom on the diffusion of new products. What Is the Source of Difference Between Early and Mainstream Markets? An interesting question relates to the source of the difference between the early and mainstream markets. Why don’t majority market members start to adopt earlier, and why aren’t they affected by the early market? It may be that the answer Moore gives regarding high-tech products in a business-to-business setting also applies to many other durables—that is, the needs of the early market adopters are so different from those of the majority that, to some extent, they cannot be considered the same market. During the stage of its introduction to the market, the product may not offer high enough value for many customers to adopt it, for example, by being expensive, unreliable, or mediocre (Rink and Swan 1979; Moore and Pessemier 1993). In recent writings Moore, Johnson, and

54

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Kippola (1997) emphasize the need for a product standard or a “dominant design” (Anderson and Tushman 1990) as a factor that the mainstream market considers a must before starting to adopt. Early market members who are interested in the technology or are willing to take the risk earlier may adopt, but at the beginning the product is of little interest to many, in fact to the majority, even if they are aware of it. Only after the product changes enough will the takeoff that symbolizes the acceptance by a larger number of consumers occur (Moore and Pessemier 1993). Take, for example, the case the case of the color TV, one of the durables analyzed here. In the 1950s following its introduction, the color TV suffered from many “childhood diseases” typical of similar products: quality was low, not many stations were broadcasting in color, and it was expensive to purchase. In fact, during the 1950s, color TV was chosen as the “worst durable of the year” by Consumer Reports. It is no wonder that it was not of interest to many. Only after it improved in quality, increased program offerings, and dropped in price at the beginning of the 1960s did the color TV take off. Are Single-Phase Diffusion Models Still Applicable? We note that for two of the three models examined—color TVs and room air conditioners—the single-phase Bass model has been successfully applied in a number of diffusion studies. However, in general most marketing studies have not included early market data, beginning with the commencement of diffusion. Thus, if one is interested mostly in a post-takeoff examination, the single-phase Bass model may be sufficient. However, as also implied by Mahajan, Muller, and Bass (1990), a single-phase model may not be sufficient for a period that includes the early market. For an investigation that includes the early market, a two-phase examination as presented here might be needed. Is the Early Market Still Important? If early market members have little effect on the mainstream, does it mean that they are of little value to marketers? This is not necessarily implied by our results. If the product needs to be changed in order to fit mainstream market needs, the early market needs to serve as a “large beta site” in which standards are defined, a learning curve enables price reduction, and product bugs are eliminated. Thus, the early market may still play an important role, though perhaps different from the traditional one attributed to it. The Cost of the Chasm Another interesting question is whether we can estimate the “cost” in terms of loss in unit adopters caused by the chasm. In order to examine this point, we should compare a case in which A, the opinion leadership parameter, is 1 (that is, there is no chasm), to a case in which the chasm exists. An example of this is given in Figure 2.4 in which actual unit sales in the years 1962 (chasm) to 1967 (peak of sales) are compared to the results we obtained with  = 0 and a case of  = 1. As can be seen from Figure 2.4, the difference in the number of adoptions between a case of a full chasm and a case of no chasm can be considerable. In the early years, a no-chasm scenario would lead to more than 50 percent more units sold per year.

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE Figure 2.4

55

Cost of the Chasm for Color TV

7 6

Units (millions)

5 4 3 2 1 0 1962

1963

1964

1965

1966

1967

Year Unit

Alpha = 0

Alpha = 1

Conclusions While our work provides some support for the dual market phenomenon, many important questions are still left open. In our case, we conducted an ad hoc examination of the chasm phenomenon. Can managers find ways to better predict it? Can we find empirical evidence for the product factors that most differentiate between early and mainstream market adoptions? What are the factors that affect the depth of the communications break between products? Aggregate adoption data is not sufficient for answering these questions, and more in-depth investigation, possibly across various time points, should be conducted. It is clear that if the chasm phenomenon is indeed widespread, then important changes should be made in the way that we teach and research the diffusion of innovations. This study is only a step in this direction. Acknowledgments The authors would like to thank Renana Peres and Ashutosh Prasad for a number of thoughtful comments and suggestions. References Anderson, Philip, and Michael L. Tushman. 1990. “Technological Discontinuities and Dominant Designs: A Cyclical Model of Technological Change.” Administrative Science Quarterly 35 (December), 604–33. Balasubramanian, Siva, and Dipak C. Jain. 1994. “Simple Approaches to Evaluate Competing Non-Nested Models in Marketing.” International Journal of Research in Marketing 11, 53–72. Bass, Frank M. 1969. “A New Product Growth Model for Consumer Durables.” Management Science 15 (January), 215–27. Becker, Marshal H. 1970. “Sociometric Location and Innovativeness: Reformulation and Extension of the Diffusion Model.” American Sociological Review 35, 262–82.

56

BARAK LIBAI, VIJAY MAHAJAN, AND EITAN MULLER

Bell 1963. “Consumer Innovators: A Unique Market for Newness.” In Proceedings of the American Marketing Association, ed. S. Greyser. Chicago: American Marketing Association. Boswijk, Peter H., and Philip Hans Franses. 2005. “On the Econometrics of the Bass Diffusion Model.” Journal of Business & Economic Statistics 23 (3), 255–88. Brown, Lawrence. 1981. Innovation Diffusion: A New Perspective. London and New York: Methuen. Chandrasekaran, Deepa, and Gerard J. Tellis. 2006. “Getting a Grip on the Saddle: Cycles, Chasms, or Cascades?” PDMA Research Forum, Atlanta, October 21–22. ———. 2007. “A Critical Review of Marketing Research on Diffusion of New Products.” Review of Marketing Research, Chapter 2, 39–80. Cool, Karel O., Ingemar Dierickx, and Gabriel Szulanski. 1997. “Diffusion of Innovations Within Organizations: Electronic Switching in the Bell System, 1971–1982.” Organization Science 8 (5), 543–59. Davidson, R., and John G. MacKinnon. 1981. “Testing the Specification of Multivariate Models in the Presence of Alternative Hypotheses.” Journal of Econometrics 23, 301–13. Engel, James F., Roger D. Blackwell, and Robert Kegerreis. 1969. “How Information Is Used to Adopt an Innovation.” Journal of Marketing 33 (July), 15–19. Gatignon, Hubert, and Thomas S. Robertson. 1985. “A Propositional Inventory for New Diffusion Research.” Journal of Consumer Research 11, 849–67. Goldenberg, Jacob, Barak Libai, and Eitan Muller. 2002. “Riding the Saddle: How Cross-Market Communications Can Create a Major Slump in Sales.” Journal of Marketing 66, 21–16. Golder, Peter N., and Gerard T. Tellis. 1997. “Will It Ever Fly? Modeling the Takeoff of Really New Consumer Durables.” Marketing Science 16 (3), 256–70. ———. 2004. “Growing, Growing, Gone: Cascades, Diffusion, and Turning Points in the Product Life Cycle.” Marketing Science 23, 207–18. Grewal, Rajdeep, and Raj Mehta. 2000. “The Role of the Social-Identity Function of Attitudes in Consumer Innovativeness and Opinion.” Journal of Economic Psychology 21 (3), 233–52 Hauser, John, Gerard J. Tellis, and Abbie Griffin. 2006. “Research on Innovation: A Review and Agenda for Marketing Science.” Marketing Science 25 (6), 687–717. Jain, Dipak, and Ram C. Rao. 1990. “Effect of Price on the Demand for Durables: Modeling, Estimation, and Findings.” Journal of Economic and Business Statistics 8 (2), 163–70. Jiang, Zhengrui, Frank M. Bass, and Portia Isaacson Bass. 2006. “The Virtual Bass Model and the Left-Hand Truncation Bias in Diffusion of Innovation Studies.” International Journal of Research in Marketing 23 (1), 93–106. Karmeshu, K., and Debasree Goswami. 2001.”Stochastic Evolution of Innovation Diffusion in Heterogeneous Groups: Study of Life Cycle Patterns.” IMA Journal of Management Mathematics 12, 107–26. Lehmann, Donald, and Mercedes Esteban-Bravo. 2006. “When Giving Some Away Makes Sense to JumpStart the Diffusion Process.” Marketing Letters 17, 243–54. Linhart, H., and Walter Zucchini. 1986. Model Selection. New York: John Wiley & Sons. Mahajan, Vijay, Charlotte H. Mason, and V. Srinivasan. 1986. “An Evaluation of Estimation Procedures for New Product Diffusion Models.” In Innovation Diffusion Models of New Product Acceptance, ed. Vijay Mahajan and Yoram Wind. Cambridge, MA: Ballinger. Mahajan, Vijay, and Eitan Muller. 1998. “When Is It Worthwhile Targeting the Majority Instead of the Innovators in a New Product’s Launch?” Journal of Marketing Research 35, 488–95. Mahajan, Vijay, Eitan Muller, and Frank M. Bass. 1990. “New Product Diffusion Models in Marketing: A Review and Directions for Research.” Journal of Marketing 54, 1–26. ———. 1995. “Diffusion of New Products: Empirical Generalizations and Managerial Uses.” Marketing Science 14 (3), 79–88. Mahajan, Vijay, Eitan Muller, and Rajendra K. Srivastava. 1990. “Determination of Adopter Categories by Using Innovation Diffusion Models.” Journal of Marketing Research 27 (February), 37–50. Meade, Nigel, and Towhidul Islam. 1998. “Technological Forecasting: Model Selection, Model Stability and Combining Models.” Management Science 44, 1115–30. ———. 2006. “Modeling and Forecasting the Diffusion of Innovation: A 25-Year Review.” International Journal of Forecasting 22 (3), 519–45. Midgley, David F. 1977. Innovation and New Product Marketing. New York: John Wiley & Sons. Midgley, David F., and Grahame R. Dowling. 1978. “Innovativeness: The Concept and Its Measurement.” Journal of Consumer Research 4, 229–42.

INNOVATION DIFFUSION ACCORDING TO ROGERS, BASS, AND MOORE

57

Moore, Geoffrey A. 1991. Crossing the Chasm. New York: HarperBusiness. ———. 1995. Inside the Tornado. New York: HarperBusiness. Moore, Geoffrey A., Paul Johnson, and Tom Kippola. 1997. Gorilla Game. New York: HarperBusiness. Moore, William L., and Edgar A. Pessemier. 1993. Product Planning and Management: Designing and Delivering Value. New York: McGraw-Hill. Muller, Eitan, Renana Peres, and Vijay Mahajan. 2007. “Innovation Diffusion and New Product Growth: Beyond Interpersonal Communications.” Working paper. Muller, Eitan, and Guy Yogev. 2006. “When Does the Majority Become a Majority? Empirical Analysis of the Time at Which Main Market Adopters Purchase the Bulk of Our Sales.” Technological Forecasting and Social Change 73 (10), 1107–20. Perreault, William, and Jerome E. McCarthy. 1996. Basic Marketing: A Global Managerial Approach. Homewood, IL: Irwin. Rink, David R., and John E. Swan. 1979. “Product Life Cycle Research: A Literature Review.” Journal of Business Research 7 (2), 19–242. Rogers, Everett M. 1986. Communication Technology: The New Media in Society. New York: Free Press. ———. The Diffusion of Innovations. 4th ed. New York: Free Press. Schmittlein, David C., and Vijay Mahajan. 1982. “Maximum Likelihood Estimation for an Innovation Diffusion Model of New Product Acceptance.” Marketing Science 1, 57–78. Solomon, Michael R. 2006. Consumer Behavior. 7th ed. Englewood Cliffs, NJ: Prentice Hall. Srinivasan, V., and Charlotte H. Mason. 1986. “Nonlinear Least Square Estimation of New Product Diffusion Models.” Marketing Science 5, 169–78. Stone, M. 1974. “Cross-Validatory Choice and the Assessment of Statistical Predictions.” Journal of the Royal Statistical Society B 36, 11–147. Stremersch, Stefan, and Gerard J. Tellis. 2004. “Understanding and Managing International Growth of New Products.” International Journal of Research in Marketing 21 (4), 421–38. Stremersch, Stefan, and Christophe Van den Bulte. 2007. “Contrasting Early and Late New Product Diffusion: Speed Across Time, Products and Countries.” Working paper. Sultan, Fareena, John. H. Farley, and Donald R. Lehmann. 1990. “A Meta-Analysis of Applications of Diffusion Models.” Journal of Marketing Research 27 (1), 70–77. Summers, John. O. 1971. “Generalized Change Agents and Innovativeness.” Journal of Marketing Research 8, 313–16. Tanny, Stephen M., and Nicholas A. Derzko. 1988. “Innovators and Imitators in Innovation Diffusion Modeling.” Journal of Forecasting 7, 225–34. Tellis, Gerard J., Stefan Stremersch, and Eden Yin. 2003. “The International Takeoff of New Products: The Role of Economics, Culture, and Country Innovativeness.” Marketing Science 22 (2), 188–208. Utterback, James. 1994. Mastering the Dynamics of Innovation. Boston: Harvard Business School Press. Van den Bulte, Christophe, and Yogesh V. Joshi. 2007. “New Product Diffusion with Influentials and Imitators.” Marketing Science 26 (3), 400–424. Van den Bulte, Christophe, and Stefan Stremersch. 2004. “Social Contagion and Income Heterogeneity in New Product Diffusion: A Meta-analytic Test.” Marketing Science 23, 530–44. Watts, Duncan J., and Peter S. Dodds. 2007. “Influentials, Networks, and Public Opinion Formation.” Journal of Consumer Research 34(4), 441–58.

58

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

CHAPTER 3

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Abstract The purpose of this chapter is to explore two of Eric Raymond’s key open source product development principles. To do this, we empirically examine the relationships between project community size (“eyeballs”) and development activity, and between development activity and product adoption. We find strong evidence to support the premise that “developer eyeballs” are positively related to development activity. Based on a proportional hazard analysis of time to adoption takeoff, we also find that product development activity is significantly related to the speed of product adoption. We interpret these results as supporting some key principles of the open source bazaar: (1) attracting a large developer base is important to further product development, and (2) the early market evolution and acceptance of open source products is driven by product development activity. Contrary to the tenets of the bazaar model, however, we find that “user eyeballs” do not significantly contribute to increased development activity. In addition, we find evidence suggesting that product success does not in turn drive additional product development. Thus, our results also suggest that the bazaar community development model involving developers and users originally proposed by Raymond needs revision for the more typical open source development project. Introduction Open source products, services, and ideas are those in which the intellectual inputs and outputs are nonproprietary in nature and freely shared (free as in free speech, not as in free beer; see www. opensource.org and www.gnu.org/philosophy/free-sw.html). The term originated in computer programming, where the open source software movement is largely comprised of highly skilled professional developers with a commitment to the philosophy of “open source code” (David, Waterman, and Arora 2003). These developers collaborate over the Internet, generally voluntarily developing software during their own spare time. In the software industry, the open source development approach has produced reliable, high-quality products quickly and inexpensively (e.g., “Business” 2001). From its humble beginnings among a few dedicated programmers, the open source software movement has now grown to over 100,000 projects and 1 million registered users (a count as of May 2006 from sourceforge.net, the largest repository of open source projects on the Internet). For example, open source software is pervasive in the web server domain, with Apache leading the way in having the largest market share (almost 65 percent, according Netcraft’s Web Server survey as of May 2006), and the Linux operating system is emerging as a strong contender in the 58

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

59

desktop category. High-profile successes like these have piqued the imagination of the business world since the open source approach promises that high-quality products can be produced in a relatively short period of time, at very little cost, by some of the best developers in the industry (e.g., O’Reilly 1999; Goetz 2003; Krishnamurthy 2003). The term “open source” is now used more broadly to characterize situations that have similar philosophical underpinnings to open source software. For example, a group of students at the ITUniversity in Copenhagen have created Vores Øl (Our Beer) as an open source beer. According to their website (www.voresoel.dk), The recipe and the whole brand of Our Beer is published under a Creative Commons license, which basically means that anyone can use our recipe to brew the beer or to create a derivative of our recipe. You are free to earn money from Our Beer, but you have to publish the recipe under the same license (e.g. on your website or on our forum) and credit our work. You can use all our design and branding elements, and are free to change them at will provided you publish your changes under the same license (“Attribution & Share Alike”). Jeff Howe, a contributing editor of Wired magazine, coined the term “crowdsourcing” to describe situations in which a firm’s business model depends on work being done by a combination of amateurs and volunteers who use their spare time to offer new product ideas, create content, solve problems, or even do corporate R&D (Howe 2006). YouTube, the popular website for downloading video clips, is a prominent example. Not surprisingly, the open source approach is generating a lot of research attention from academic researchers (see the paper repository at opensource.mit.edu). For example, economists are intrigued by the underlying altruistic motivations behind the open source philosophy since it seems to be inconsistent with traditional (rational) utility models (e.g., see the review in Lerner and Tirole 2002). Legal scholars (e.g., Benkler 2002), organizational academics (e.g., see the review in von Hippel and von Krogh 2003), and technology management researchers (e.g., see the review in von Krogh and von Hippel 2003) are carefully studying the democratic “community” model of shared activities inherent to the open source movement. Even marketing researchers have recently become interested in the branding implications associated with the open source philosophy (Pitt et al. 2006). Embedded within the open source concept is a very different perspective on product development. In particular, traditional organizations manage the risks of new product development via rigid and bureaucratic processes, often employing structures involving centralized control and decision-making, hierarchical governance, and constrained information flow (Sharma, Sugumaran, and Rajagopalan 2002). As shown in Figure 3.1, open source development communities are very different since they are usually composed entirely of volunteers, they are nimble and flexible, have shared governance, and allow for a free flow of information (see again Sharma, Sugumaran and Rajagopalan 2002). Clearly, the open source development model lacks the command and control structure and enforcement of rules typically associated with traditional organizational environments. The open source development process has been characterized as a bazaar, a usage coined by Raymond (1998). Aside from the nuances outlined in Figure 3.1, the bazaar model of product development is based on two fundamental principles: (1) “given enough eyeballs, all bugs are shallow,” and (2) “release often and release early” (Raymond 1998). This approach is in direct contrast to the traditional software development process, which Raymond characterizes as a cathedral (i.e., the Stage-Gate process where products are developed using a centralized approach,

60

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Figure 3.1



The Open Source Development Model

Structure % % % % % %

Culture

Peer supervision Democratic decision making No organizational boundaries Informal networks Fluid political coalitions Reputation as the basis of authority

% % % % % % %

Shared risk and ownership Shared reward based on merit Motivated by altruism Democratic decision making Flexible work structure Shared trust Shared loyalty

Product Development % % % % %

Voluntary participation Self-governance Community members assign tasks Performance is visible to entire community Motivated by reputation building

Source: Adapted from Sharma, Sugumaran, and Rajagopalan (2002).

designs are carefully drafted, and there are no beta prototype releases until all the major bugs are resolved [Raymond 1998]). Despite all the attention it has received, paradoxically, we still have a relatively limited understanding of the evolution of open source systems and especially their underlying product development process (e.g., Kemerer and Slaughter 1999; Godfrey and Tu 2000; Krishnamurthy 2002; Healy and Schussman 2003; Scacchi 2003; Paulson, Succi and Eberlein 2004). Originally constructed based on a qualitative examination of Linus Torvalds’s development of the Linux operating system, Raymond’s community development model has been supported by case studies of large successful software systems including Linux (e.g., Godfrey and Tu 2000), Apache (e.g., Mockus, Fielding, and Herbsleb 2000; Lakhani and von Hippel 2003), Mozilla (e.g., Mockus, Fielding, and Herbsleb 2002), and GNOME (e.g., Koch and Schneider 2000). Nevertheless, research to date is equivocal on whether the bazaar development principles apply to more “typical” open source software projects that never achieve the kind of market success generally associated with these larger projects (e.g., Hunt and Johnson 2002; Krishnamurthy 2002; Crowston, Annabi, and Howison 2003; Healy and Schussman 2003; Koch 2004). For example, based on a study of thousands of projects available from the Sourceforge online database, Healy and Schussman (2003) find that the typical open source software project has one developer, has no discussion or bug reports, and is not downloaded by anyone. Krishnamurthy (2002) reports similar results, proposing that the development of open source software is more consistent with a lone developer (or cave) model. However, these conclusions are based on

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

61

simple descriptive statistics and the observation that development effort and project activity across open source software projects is highly skewed (i.e., many projects have few developers and little or no activity, while only a few have a lot of development interest and activity). We are unaware of any empirical studies providing multivariate statistical support for (or against) the key bazaar development principles. Thus, before its tenets are widely accepted and implemented, further understanding of the underlying product development process associated with open source software is still needed. The purpose of this chapter is to further explore the open source software bazaar. Consistent with Raymond’s key bazaar principles, our primary interest is in whether there is a statistical relationship between project community size (“eyeballs”) and development activity, and between development activity and product adoption. We empirically examine these relationships in a random sample of several hundred “typical” open source products hosted on Sourceforge. In agreement with the bazaar model, we find strong evidence that “developer eyeballs” are positively related to development activity (i.e., bug fixes, bug reports, support requests, commits, version releases). Based on a proportional hazard analysis, we also find that product development activity (e.g., rate of version releases and timing of first version release) is significantly related to the speed of product adoption. We interpret these results as supporting some key principles of the open source product development bazaar: (1) attracting a large developer base is important to further product development, and (2) the early market evolution and acceptance of open source products is driven by product development activity. Contrary to the tenets of the bazaar model, however, we find that “user eyeballs” do not significantly contribute to increased development activity. In addition, we find evidence suggesting that product success does not in turn drive additional product development. Thus, our results also suggest that the bazaar community development model involving developers and users originally proposed by Raymond (1998) needs revision for the more typical open source development project. In particular, it appears that users do not always play a critical role in development during the early evolution of a project, and development activities significantly drop off once downloads take off. The remainder of this chapter is organized as follows. In the next section, we describe the open source in more detail and explain the basis for our research model. We then describe the data available for our analyses, present the empirical analysis, and discuss our findings. We conclude with some directions for future research in the final section. Conceptual Framework The underlying conceptual model that guides our study is shown in Figure 3.2. In line with the product and software development literatures, we expect that: (1) project community size (indicating development efforts) is positively related to development activity, and (2) development activity (and its associated product improvements) is positively related to the speed of product adoption (e.g., Clark and Wheelwright 1994; Cusumano and Selby 1996; Krishnan and Ulrich 2001; MacCormack, Verganti, and Lansiti 2001; Garcia 2004; Wynn 2004). We note that this basic framework is consistent with results reported in the economics and technology literatures that new innovations typically begin in a primitive form and the early stages of new market growth focuses on continual product improvement (e.g., Klepper 1997; Agarwal and Bayus 2002; Wynn 2004). In addition, several studies demonstrate that product improvements, relative to process improvements, are emphasized in the early stages of a new market (e.g., Utterback 1994; Klepper 1997).

62

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Figure 3.2



Research Model

Project Community Size             

Development Activity

Product Adoption

In the context of open source software, this basic framework is consistent with Raymond’s (1998) bazaar model of product development. In particular, the ability to parallelize the debugging process is touted as a primary benefit of a large community. According to Linus’s law, given a sufficiently large group of developers and beta-testers (i.e., users), bugs will be found and fixed rapidly (Raymond 1998). In other words, developer and user “eyeballs” are expected to be positively associated with product development activity (e.g., bug reports, bug fixes, support requests, version releases). Raymond (1998) states that an important project activity is releasing new versions early and often. New version releases enhance the debugging process, and thus lead to improved products. Following the product development literature (e.g., Krishnan and Ulrich 2001; Agarwal and Bayus 2002), we expect that open source development activity is positively related to the speed of product adoption. Importantly, this simple conceptual model and its associated relationships have not been empirically tested in the open source software environment (Garcia 2004). Instead, researchers have focused their attention on the motivations of programmers to voluntarily contribute to open source projects (e.g., Raymond, 1998; Cohendet, Créplet, and Dupouët 2001; Hars and Ou 2001; Kasper 2001; Tuomi 2001; Lakhani and Wolf 2003; Rossi 2004), how the open source innovation process works (e.g., Mockus, Fielding, and Herbsleb 2000; Godfrey and Tu 2001; Lerner and Tirole, 2002; Jensen and Scacchi 2004a, 2004b; Paulson, Succi, and Eberlein 2004, Zhao and Deek 2004), and the competitive dynamics associated with open source software products (e.g., Bonaccrossi and Rossi 2003; Dedrick and West 2003; Overby, Bharadwaj, and Bharadwaj 2004). To test the relationships in our conceptual model, we statistically analyze the cross-sectional variation across a large sample of “typical” open source projects. Based on our discussion in this section, we have three primary hypotheses: (1) projects with many developer eyeballs have more development activity than projects with few developer eyeballs; (2) projects with many user eyeballs have more development activity than projects with few user eyeballs; and (3) projects with versions that are released early and often have quicker product adoption than projects with little development activity. We empirically examine each of these hypotheses in the next section. The Empirical Study Data Rather than emphasize only the larger, well-known open source projects, we explore the relationships between project community size and development activity, as well as development activity and product adoption, using a random sample of open source projects hosted by Sourceforge. In addition to being the largest host for open source projects, Sourceforge is a rich source of data on

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

63

Table 3.1

Number of Developers and Project Managers Number of Developers Number of Project Managers One

More than one

One

More than one

51.4%

22.2%

(n = 505)

(n = 218)

0.0%

26.4%

(n = 0)

(n = 260)

the open source development process (e.g., bugs, support requests, version release) and, hence, is an excellent data source for researchers (Garcia 2004).1 Not surprisingly, other researchers are also drawing on this data source (e.g., Chengalur-Smith and Sidorova 2003; Healy and Schussman 2003; Xu and Madey 2004). Data for our study were collected from sourceforge.net in December 2003. In general, we exercised caution in acquiring, cleaning, and analyzing data (see Howison and Crowston 2004). As suggested by Garcia (2004), we used a web crawler to generate a comprehensive list of projects from the six largest project categories listed on Sourceforge (this comprised a total of 17,035 projects). It is important to note that assignments to project categories are not mutually exclusive. For the purposes of this study, the first category listed was chosen. From these projects, we randomly selected 1,000 projects to be included in our study. With the help of a webbot, data on project characteristics and development activity were then collected for these projects. Since fifteen projects had to be dropped for lack of information or being duplicates of other projects in the set, the sample size for our study is 985 projects. Our sample includes projects that are in different stages of development: 21 percent are in the planning phase, 14 percent are in the pre-alpha phase, 16 percent are in the alpha phase, 26 percent are in the beta phase, 20 percent are in the production/stable category, and 1.5 percent are in the mature phase of development (only three projects in our sample were listed as inactive). In terms of project age (as of December 2003), more than three-quarters of the projects are less than two years old, and none are older than four years. The basic characteristics of our sample are similar to other studies using Sourceforge data (e.g., see Healy and Schussman 2003). As shown in Table 3.1, about half of our sample has only one developer and one project manager. Additionally, almost half of the open source projects in our sample have more than one developer (Table 3.1: 22.2% + 26.4%). As reported in Table 3.2, a majority of the projects in our sample had at least one version release and one commit. At the same time, more than half of our sampled projects have no bug reports, no bug fixes, and no support requests. We note that the highly skewed nature of these measures is consistent with the findings reported by other researchers (e.g., Hunt and Johnson 2002; Krishnamurthy 2002; Healy and Schussman 2003; Koch 2004).

64

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Table 3.2

Project Activity Project Activity Number of bug fixes Number of bug reports Number of support requests Number of commits Number of versions released

0

r1

61.8% (n = 446) 52.0% (n = 386) 82.9% (n = 561) 36.5% (n = 292) 17.8% (n = 156)

38.2% (n = 276) 48.0% (n = 357) 17.1% (n = 116) 63.5% (n = 509) 82.2% (n = 720)

Developer Eyeballs and Development Activity To empirically examine Linus’s Law, we now explore the relationship between developer eyeballs and development activity. Our measure of developer eyeballs is based on the number of developers participating in a project. For the projects in our sample, the number of developers and project managers working on each project as of December 2003 was collected. To check for stability over time, we revisited this measure six months later. Consistent with Krishnamurthy (2002), we find that the developer count is relatively stable over the life of a project. (Of the 985 projects in our study, fewer than 15 had any major count change.) Given the existence of a highly skewed distribution of development effort across open source projects, we define two categories to facilitate our analysis and interpretation: one developer and more than one developer.2 Software development activity is typically characterized as a function of both new development and the fixing of bugs (Garcia 2004). In line with this, product development activity is measured as the number of bugs fixed, bug reports, version releases, support requests, and commits to make future coding changes. Three of the measures—bugs fixed, version releases, and commits—are directly indicative of product development activities. Although the other two dimensions—support requests and bug reports—most likely originate from the user community, they are likely to spur development and hence serve as indirect indicators of development activity. Again, due to the highly skewed nature of these measures across open source projects, we define two categories for each measure: zero and at least one. An analysis of developer eyeballs and development activity is in Table 3.3. It is clear that projects with more than one developer are significantly more likely to have bug fixes, bug reports, support requests, commits, and version releases.3 For example, 29.1 percent of open source projects with one developer had at least one bug report, whereas 52.2 percent of projects with more than one developer had at least one bug report (this difference is significant at better than 0.01). We note that this does not really contradict prior findings that a small number of developers (especially in large project communities) contribute a large fraction of the code (e.g., Ghosh and Prakash [2000] find that 10 percent of the contributors in their study accounted for over 70 percent of the code). Indeed, CVS (concurrent versions system) data on lines of code for the projects in our sample (with

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

65

Table 3.3

Cross-Tabulations of Project Activity and “Developer Eyeballs”

At least 1 bug fix At least 1 bug report At least 1 support request At least 1 commit At least 1 version release

1 Developer

> 1 Developer

21.0%* (n = 328) 29.1% (n = 327) 10.1% (n = 328) 36.6% (n = 328) 69.7% (n = 505)

41.9% (n = 322) 52.2% (n = 322) 17.7% (n = 322) 71.7% (n = 322) 77.2% (n = 478)

*21.0% of projects with 1 developer had at least 1 bug fix, and 100 – 21.0 = 79% had no bug fixes.

at least six months of lines of code information; n = 349) indicate that more than 20 percent of the open source projects with one project manager and one developer were able to generate 100,000 or more lines of code. This is not statistically different than projects with more than developer. We interpret these results to strongly indicate that a larger developer base is associated with increased levels of product development activity, supporting the bazaar development principle that more developer eyeballs are associated with higher development activity. User Eyeballs and Development Activity Unfortunately, direct measures of the number of users for an open source product are generally unavailable (e.g., Garcia 2004). Consequently, we use the cumulative number of downloads as a proxy for users (see also Garcia 2004; Wynn 2004). We recognize that one user may download multiple copies and/or pass along a downloaded version to several other users, and, hence, measuring downloads has its limitations (e.g., see Howison and Crowston 2004). But we believe these problems are less severe for small or medium-size projects (the type of which are “typical” of open source projects on Sourceforge and our sample) where alternate distribution channels are not the mainstay. Examples of the cumulative download pattern for four open source projects in our sample are shown in Figure 3.3. These patterns exhibit the well-known “takeoff” phenomenon, i.e., downloads are very low (if not zero) for several months during the early stages of a project; for successful products, downloads eventually sharply increase (e.g., Golder and Tellis 1997; Agarwal and Bayus 2002). In most cases, the takeoff in downloads can be visually identified.4 For example, the takeoff in downloads for megaZeux occurs at 16 months, and RadeonTweaker has a takeoff at 6 months. All the open source products in our sample either exhibited a similar takeoff in downloads (649 projects) or had zero downloads for the entire observed period (336 projects). For the projects with a takeoff in downloads, the mean (median) time to takeoff is 13 (9) months. To explore the relationship between user eyeballs and development activity, we compare our development activity measures before (when there are few user eyeballs) and after5 (when there

66

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Figure 3.3

Examples of the Download Takeoff for Some Open Source Products 19 6 4

LibSIMD

300,000 2500

Total Downloads

Total Downloads

250,000 200,000 150,000 100,000

2000 1500 1000 500

50,000

Se

p0 No 1 v01 Ja n0 M 2 ar -0 2 M ay -0 2 Ju l-0 2 Se p02 No v02 Ja n0 M 3 ar -0 3 M ay -0 3 Ju l-0 3 Se p03 No v03

0

M

ay -0 Ju 1 l Se -01 pNo 0 1 v Ja -01 n M -01 ar M -01 ay -0 Ju 1 l Se -01 pNo 0 1 vJa 01 n M -01 ar M -01 ay -0 Ju 1 l Se -01 p No - 0 1 vJa 01 n M -01 ar M -01 ay -0 Ju 1 l-0 1

0

Time

Time

RadeonTweaker 600000

Total Downloads

400000 300000 200000

0

No

v0 Ja 0 n M -01 ar M 01 ay -0 Ju 1 l Se -01 pNo 01 vJa 01 nM 02 ar M 02 ay -0 Ju 2 lSe 02 pNo 02 vJa 02 nM 03 ar M -03 ay -0 Ju 3 lSe 03 pNo 03 v03

3

03 v-

No

3 -0

l-0

ar

Ju

M

2

02 v-

No

2 -0

l-0

ar M

Ju

1

01 v-

No

1 -0

l-0

ar

Ju

M

0

00 v-

No

0

l-0 Ju

-0

v-

ar M

No

500000

100000

99

Total Downloads

megaZeux 20000 18000 16000 14000 12000 10000 8000 6000 4000 2000 0

Time

Time

are many user eyeballs) the download takeoff. Thus, for this analysis we only consider open source projects that exhibited a takeoff in downloads. An analysis of development activity before and after takeoff is in Table 3.4. Based on a statistical test of two proportions,6 it is significantly more likely that there are bug reports, bug fixes, support requests, commits, and version releases before takeoff than after takeoff. In other words, product development activity is more prevalent before there is a large number of downloads, that is, when there are few user eyeballs. Interestingly, this result is consistent with the product development literature (e.g., Klepper 1997; Krishnan and Ulrich 2001; Agarwal and Bayus 2002; Wynn 2004), but is contrary to some of the bazaar development principles as more user eyeballs fail to translate into more bug reports, bug fixes, or support requests (e.g., Raymond 1998). To examine the robustness of our finding that “user eyeballs don’t matter,” we examine development activity for different levels of development effort before and after the download takeoff in Table 3.5. Generally speaking, open source projects with more than one developer are significantly more likely to have development activity before the download takeoff than after the takeoff (in Table 3.5, compare the before and after takeoff columns for projects with more than one developer). In addition, Table 3.5 confirms our results that developer eyeballs are significantly related to development activity before the takeoff in downloads (but not after the download takeoff). This can be seen by statistically comparing the proportions within each column (e.g., 20.7 percent of the projects with one developer and one project manager had at least one bug report before the takeoff, whereas 48.1 percent of the projects with one project manager and more than one developer had at least one bug report before the takeoff). In summary, we report three important and related findings: (1) strong empirical evidence that user eyeballs are not significantly associated with product development activity, (2) developer eyeballs are positively related to development activity only when there are few user eyeballs, and

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

67

Table 3.4

Cross-Tabulations of Project Activity and “User Eyeballs” (only projects that exhibited a takeoff in downloads)

At least 1 bug fix At least 1 bug report At least 1 support request At least 1 commit At least 1 version release

Before Download Takeoff

After Download Takeoff

27.1%* (n = 649) 35.6% (n = 648) 10.3% (n = 649) 52.5% (n = 649) 92.6% (n = 649)

12.5% (n = 431) 17.2% (n = 431) 6.3% (n = 431) 20.9% (n = 431) 29.4% (n = 431)

*27.1% of projects had at least 1 bug fix before the download takeoff, and 100 – 27.1 = 73.9% had no bug fixes before the download takeoff.

(3) development activities significantly drop off once downloads dramatically increase, and project development after the download takeoff does not really depend on development effort. One possible explanation for our first finding indicating a lack of impact of user eyeballs on development activity is that today the profile of a typical user is very different from that of early open source projects (e.g., Linux during the early 1990s). Users today can be described as being passive eyeballs. This is in contrast to the hacker community members who served not only as developers but also as primary users for the early projects, and who were technologically adept and could potentially play an active role as users. But with the adoption of open source products well beyond the hacker group, typical users are now less sophisticated in the workings of the technology (relative to the hacker group) and hence are more interested in using the software for their needs rather than participating as beta testers. Even commercial vendors that use beta-testers usually rely on the lead user group and not their typical user. One implication of these results is that studies simulating the open source development process should be cautious in their assumptions regarding the impact of a large user population (e.g., Dicker 2004) Our second finding of a positive relationship between developer eyeballs and development activity only in the case of low user eyeballs gives us a more refined understanding of the importance of developer eyeballs. Our third result provides an initial answer to the question raised by Garcia (2004) about whether product acceptance leads to further increases in project community size and product development. Our results do not support this idea. Instead, we find that development activity after a takeoff in downloads is substantially lower than the activity before takeoff. Development Activity and Product Adoption We now explore the relationship between development activity and product adoption. In line with Wynn (2004) and Garcia (2004), we use cumulative downloads as our measure of product adoption (recognizing the limitations as discussed in the previous section). It is important to point out that our data are truncated (i.e., right-censored) in that we stopped observing download activity in

68

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Table 3.5

Cross-Tabulations of Project Activity, “Developer Eyeballs,” and “User Eyeballs” (only projects that exhibited a takeoff in downloads)

At least 1 bug fix 1 project mgr. & 1 developer 1 project mgr. & >1 developer > 1 project mgr. & >1 developer At least 1 bug report 1 project mgr. & 1 developer 1 project mgr. & >1 developer > 1 project mgr. & >1 developer At least 1 support request 1 project mgr. & 1 developer 1 project mgr. & >1 developer > 1 project mg. & >1 developer At least 1 commit 1 project mgr. & 1 developer 1 project mgr. & >1 developer > 1 project mgr. & >1 developer At least 1 version release 1 project mgr. & 1 developer 1 project mgr. & >1 developer > 1 project mgr. & >1 developer

Before Download Takeoff

After Download Takeoff

12.9%* (n = 295) 38.3% (n = 162) 39.5% (n = 192)

14.7% (n = 191) 13.0% (n = 108) 9.1% (n = 132)

20.7% (n = 294) 48.1% (n = 162) 47.9% (n = 192)

18.8% (n = 191) 15.7% (n = 108) 15.9% (n = 132)

4.4% (n = 295) 13.0% (n = 162) 17.2% (n = 192)

7.3% (n = 191) 6.5% (n = 108) 4.5% (n = 132)

29.8% (n = 295) 70.4% (n = 162) 72.4% (n = 192)

18.3% (n = 191) 29.6% (n = 108) 17.4% (n = 132)

94.9% (n = 295) 87.7% (n = 162) 93.2% (n = 192)

27.2% (n = 191) 29.6% (n = 108) 32.6% (n = 132)

*Of the projects with 1 project manager and 1 developer, 12.9% had at least 1 bug fix before the download takeoff, and 100 – 12.9 = 77.1% had no bug fixes before the download takeoff.

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

69

December 2003, that is, for numerous products in our sample we do not observe when (or whether) a download takeoff occurs at some later time. Consequently, we need to employ an appropriate statistical methodology to account for this kind of potential sample selection bias.7 We follow the general analysis approach used by Golder and Tellis (1997) and Agarwal and Bayus (2002) to study the timing of a sales takeoff for product innovations. Cox’s (1972) proportional hazards regression model is used to study the timing of a takeoff in downloads (i.e., speed of product adoption). The proportional hazards model is appropriate since it allows for estimation of the determinants of the hazard rate, that is, the conditional probability of takeoff in month t given that the product has not taken off till month t – 1. For the ith open source product, the hazard rate function hi(t) is defined as log hi (t) = log h(t; xi) = a(t) + x'i B

(1)

where a(t) is an arbitrary and unspecified baseline hazard function, xi is a vector of measured explanatory variables for the ith product, and B is the vector of unknown coefficients to be estimated. The risk ratio, calculated as eB, gives the marginal effect of each explanatory variable on the hazard rate, that is, a risk ratio greater than 1 (B > 0) indicates the percentage increase in the hazard rate due to a unit increase of the corresponding explanatory variable, while a risk ratio less than 1 (B < 0) indicates the percentage decrease in the hazard rate, respectively. Note that the risk ratio is interpreted as a deviation from mean values, that is, the effect on the hazard rate as a change from the average hazard rate function. Following Golder and Tellis (1997) and Agarwal and Bayus (2002), we do not include a term for unobserved heterogeneity since we only analyze nonrepeated events. Parameter estimation is accomplished using the partial likelihood method as implemented in the SPSS package. Because we want to specifically test Raymond’s (1998) development bazaar principle of the importance of “release often and release early,” we consider the average rate of version releases before download takeoff (calculated as the number of version releases before takeoff divided by the number of months to takeoff) and the time to first version release before takeoff as explanatory variables in the hazard model. We also examine the role of bug reports, bug fixes, support requests, and commits before download takeoff in explaining the speed of open source product adoption. Estimation results for single variable models, as well as all the explanatory variables together, are reported in Table 3.6. With the exception of bug reports and bug fixes, all the other measures of development activity are significantly related to the conditional probability of download takeoff. In agreement with Raymond’s (1998) dictum, open source projects with a high rate of version releases before takeoff have quicker download takeoff times (the estimated coefficient for the rate of version releases in Table 3.6 is positive), and projects with earlier releases have faster takeoff times (the estimated coefficient for the time of first version release in Table 3.6 is negative). Interestingly, we also find that projects with support requests and outstanding commits have relatively slow download takeoffs. This suggests that downloads tend to be delayed for open source projects that are still being improved. Conclusions and Directions for Future Research In sum, our study finds support for key elements of the bazaar model of open source product development. The implications for open source project managers is that attracting a larger number of developers is important in achieving higher levels of product development activity. Also, project developers may note that product success in terms of large numbers of users (downloads)

aSignificant

at 0.01 level or better.

Time to first version release before takeoff (months) –2 LL Chi-square

Average rate of version releases before takeoff

Bug fixes before takeoff (binary: 0 or r1) Bug reports before takeoff (binary: 0 or r1) Support requests before takeoff (binary: 0 or r1) Commits before takeoff (binary: 0 or r1)

7597.2 0.43 (p = 0.51)



— 7608.0 3.27 (p = 0.07)





7603.3 7.47 (p < 0.01)





7607.3 4.04 (p = 0.05)



–0.16 (0.08) —



–0.36a (0.13) —







–0.06 (0.08) —



Model 4



Model 3



Model 2





–0.16 (0.09) —

Model 1

7147.6 44.86 (p < 0.01)



0.34a (0.05) —

–0.04a (0.01) 7059.1 26.79 (p < 0.01)









Model 6









Model 5

Cox Proportional Hazard Analysis of Product Adoption Timing (standard errors in parentheses)

Table 3.6

–0.32 (0.16) –0.11 (0.15) –0.42a (0.14) –0.24a (0.09) 0.43a (0.06) –0.02a (0.01) 6919.9 118.69 (p < 0.01)

Model 7

70 BALAJI RAJAGOPALAN AND BARRY L. BAYUS

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

71

can be achieved by higher levels of development activity. But product success is also associated with a drop-off in development activity, suggesting that further product development and improvement is not the norm. It seems that typical open source products reach a level of being “just good enough,” and developers then turn their attention to new projects that are more interesting and challenging. Contrary to the tenets of the bazaar model, we find that “user eyeballs” do not significantly contribute to increased development activity. Thus, our results also suggest that the bazaar community development model involving developers and users originally proposed by Raymond needs revision for the more typical open source software project. Although our study indicates that developer eyeballs are positively related to development activity and that project activity is related to product adoption, research exploring other measures of project community size (developer and user eyeballs), development activity, and product adoption should be conducted to confirm our findings. As the developer community size increases, so do the number of peripheral members (i.e., individuals that contribute on less than a regular basis to the project). In particular, future research might explore the idea that above a threshold the number of additional developer eyeballs might have diminishing returns in terms of development activity. Future work might also extend our work by examining various measures of project complexity and how it may influence these relationships (e.g., see Jorgensen 2001). The underlying reason that development activity after a takeoff in downloads is substantially lower than the activity before takeoff is an important question that can be explored. One way this can be approached is to examine if the motivation levels of the participants have changed after there is a sharp increase in download activity. Our results also indicate that many small (e.g., one-person) teams of open source projects do achieve product success. This raises important questions as to how and why some of these projects succeed with apparently minimal resources. There is some evidence pointing to knowledge “reuse” across projects that benefits the smaller teams as they take advantage of the work of larger communities and are able to leverage the code already generated (e.g., Brown and Booch 2002). So, while the apparent community size may seem to be small, the actual development participation (including indirect knowledge and experience gained through other projects) might really be higher. This aspect of small team success needs to be further explored. Another interesting aspect in this domain is that a myriad of open source products have been initiated as an outgrowth of other successful projects. For example, the popular open source database MySql has spurred facilitating products like PhpMyAdmin, which supports the administrative functions of the database engine. Shocker, Bayus, and Kim (2004) suggest that facilitating products can positively influence the success of existing products. Future research might study such intra-product relationships and whether open source products that have the support of facilitating products experience quicker download takeoff. Our study also suggests some directions for researchers interested in modeling the market evolution of open source products. A look at the cumulative download patterns in Figure 3.2 on page 62 suggests that some products may conform to the widely used S-shaped diffusion curve (MegaZeux), while others may not follow the standard diffusion model (e.g., RadeonTweaker and LibSIMD). Future research might explore the applicability of different new product diffusion models to open source innovations. Finally, our results seem to be consistent with the 1 percent rule of online communities: 1 percent of users participate a lot and account for most contributions (it can seem as if they do not have lives because they often respond just minutes after whatever event they are commenting on occurs), 9 percent of users contribute from time to time (but other priorities dominate their time), and 90 percent of users are lurkers, that is, they read or observe, but do not contribute (Nielsen

72

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

2006). Our observed phenomenon of products reaching a level of being “just good enough” is consistent with the idea that the best programmers contributing to open source projects want to be on the “latest and greatest” project. Further research needs to better document this phenomenon and to study its underlying antecedents and consequences. Notes We thank Katherine Stewart and Kevin Crowston as well as the participants at the 2005 Utah Winter Conference on Product and Service Innovation for their comments on an earlier draft of this paper. 1. Freshmeat.net is another large host of open source applications. However, data for the variables of interest to us are not available from this data source. 2. Due to the highly skewed nature of these data, regression (correlation analysis) is inappropriate (e.g., Garcia 2004). Since our study is exploratory in nature, we rely on simple statistical tests of two proportions rather than more complex methods involving count models (e.g., Poisson or Negative Binomial models). 3.Although not reported here, we also find that open source projects with more than one project manager are significantly more likely to have bug reports, bug fixes, support requests, and commits (but not version releases). 4. As discussed in Agarwal and Bayus (2002), we follow the procedure outlined in Gort and Klepper (1982) to identify the download takeoff point for each product in our sample. Basically, this procedure is a systematic search for the first occurrence of a very large increase in downloads after a product is made available. 5. Our measures of development activity are based on project information for twelve months after the download takeoff. Constructing our development activity measures using all available information (up to December 2003) gives the same conclusions as those reported here. 6. A paired-sample t-test gives the same results. 7. Using the hazard model framework, we estimate that the average (median) time to a download takeoff for our entire sample of open source projects is 16.7 (13) months.

References Agarwal, R., and B. Bayus. 2002. “The Market Evolution and Sales Take-off of Product Innovations.” Management Science 48 (8), 1024–41. Benkler, Y. 2002. “Coase’s Penguin, or, Linux and the Nature of the Firm.” Yale Law Journal 112 (3), 369–446. Bonaccrossi A., and C. Rossi. 2003. “Why Open Source Software Can Succeed.” Research Policy, 32 (7), 1243–58. Brown, A., and G. Booch. 2002. “Reusing Open-Source Software and Practices: The Impact of Open Source on Commercial Vendors.” Available at www.ics.uci.edu/~wscacchi/Software-Process/Readings/oss-reusepractices-Brown-Booch-ICSR7-Apri12002.pdf (accessed August 15, 2004). “Business: An Open and Shut Case.” 2001. Economist, 359 (8221), 67. Chengalur-Smith, S., and A. Sidorova. 2003. “Survival of Open Source Projects: A Population Ecology Perspective.” Proceedings of the International Conference on Information Systems, Seattle, 782–87. Clark, K.B., and S.C. Wheelwright. 1994. The Product Development Challenge: Competing Through Speed, Quality, and Creativity. Boston: Harvard Business School Press. Cohendet, P., F. Créplet, and O. Dupouët. 2001. “Organisational Innovation, Communities of Practice and Epistemic Communities: The Case of Linux.” In Economics with Heterogeneous Interacting Agents, ed. A. Kirman and J.B. Zimmermann. Berlin: Springer. Cox, D. 1972. “Regression Models and Life-Tables (with Discussion).” Journal of the Royal Statistical Society B34, 187–220. Crowston, K., H. Annabi, and J. Howison. 2003. “Defining Open Source Software Project Success.” 24th International Conference on Information Systems. Available at opensource.mit.edu/papers/crowstonannabihowison.pdf (accessed August 10, 2004). Cusumano, M., and R.W. Selby. 1996. “How Microsoft Competes.” Research Technology Management 39 (1), 26–31.

EXPLORING THE OPEN SOURCE PRODUCT DEVELOPMENT BAZAAR

73

David, P., A. Waterman, and S. Arora. 2003. “The Free/Libre/Open Source Software Survey for 2003.” SIEPR Working Paper. Available at www.stanford.edu/group/floss-us/ (accessed August 10, 2004). Dedrick, J., and J. West. 2003. “Adoption of Open Source Platforms: An Exploratory Study.” HBS-MIT Sloan Free/Open Source Software Conference. Available at opensource.mit.edu/papers/west.pdf (accessed August 15, 2004). Dicker, V.G. 2004. “A Dynamic Feedback Framework for Studying Growth Policies in Open Online Collaboration Communities.” 10th Americas Conference on Information Systems, New York. Garcia, J.M. 2004. “Quantitative Analysis of the Structure and Dynamics of the Sourceforge Project and Developer Populations: Prospective Research Themes and Methodologies.” Working paper, Stanford Institute for Economic Policy Research.” Available at siepr.stanford.edu/programs/OpenSoftware_David/ JuanM-G_FOSS-PopDyn_Report2+.pdf (accessed August 14, 2004). Ghosh, R.A., and V.V. Prakash. 2000. “The Orbiten Free Software Survey.” First Monday 5 (7). Available at firstmonday.org/issues/issue5_7/ghosh/index.html (accessed August 10, 2004). Godfrey, M., and Q. Tu. 2000. “Evolution in Open Source Software: A Case Study.” Proceedings of the International Conference on Software Maintenance, October, 131–42. Goetz, T. 2003. “Open Source Everywhere.” Wired (November 11). Available at www.wired.com/wired/ archive/11.11/opensource.html (accessed November 10, 2006). Golder, P., and G. Tellis. 1997. “Will It Ever Fly? Modeling the Takeoff of Really New Consumer Durables.” Marketing Science 16 (3), 256–70. Gort, M., and S. Klepper. 1982. “Time Paths in the Diffusion of Product Innovations.” Economic Journal 92 (3), 630–53. Hars, A., and S. Ou. 2001. “Working for Free? Motivations for Participating in Open Source Projects.” International Journal of Electronic Commerce 6 (2), 25–39. Healy, K., and A. Schussman. 2003. “The Ecology of Open-Source Software Development.” Working paper. Available at opensource.mit.edu/papers/healyschussman.pdf (accessed August 10, 2004). Howe, J. 2006. “The Rise of Crowdsourcing.” Wired (June 14). Available at www.wired.com/wired/archive/14.06/crowds.html (accessed November 10, 2006). Howison, J., and K. Crowston. 2004. “The Perils and Pitfalls of Mining Sourceforge.” Proceedings of the 1st International Workshop on Mining Repositories, Edinburgh, Scotland, 7–11. Hunt, F., and P. Johnson. 2002. “On the Pareto Distribution of Sourceforge Projects.” Proceedings of the Open Source Software Development Workshop. February, 122–29. Jensen, C., and W. Scacchi. 2004a. “Data Mining for Software Process Discovery in Open Source Software Development Communities.” Proceedings of the Workshop on Mining Software Repositories. Edinburgh, Scotland. ———. 2004b. “Discovering, Modeling, and Reenacting Open Source Software Development Processes.” Working paper. Institute for Software Research, University of California, Irvine. Jorgensen, N. 2001. “Putting It All in the Trunk: Incremental Software Development in the FreeBSD Open Source Project.” Information Systems Journal, 11, 321–36. Kasper E. 2001. “Epistemic Communities, Situated Learning and F/OSS Software Development.” Working paper. Available at: opensource.mit.edu/papers/kasperedwards-ec.pdf (accessed August 10, 2004). Kemerer, C., and S. Slaughter. 1999. “An Empirical Approach to Studying Software Evolution.” IEEE Transactions of Software Engineering 25 (4), 493–509. Klepper, S. 1997. “Industry Life Cycles.” Industrial Corporate Change 6 (1), 145–81. Koch, S. 2004. “Profiling an Open Source Project Ecology and Its Programmers.” Electronic Markets 14 (2), 77–88. Koch, S., and G. Schneider. 2000. “Results from Software Engineering Research into Open Source Development Projects Using Public Data.” Diskussionspapiere zum Tatigkeitsfeld Informationsverarbeitung und Informationswirtchaft. Available at wwwai.wu-wien.ac.at/~koch/forschung/sw-eng/wp22.pdf (accessed August 10, 2004). Krishnamurthy, S. 2002. “Cave or Community? An Empirical Examination of 100 Mature Open Source Projects.” First Monday 7 (6). Available at www.firstmonday.org/issues/issue7_6/krishnamurthy/index. html (accessed August 10, 2004). ———. 2003. “A Managerial Overview of Open Source Software.” Business Horizons (September–October), 47–56. Krishnan, V., and K. Ulrich. 2001. “Product Development Decisions: A Review of the Literature.” Management Science 47 (1), 1–21.

74

BALAJI RAJAGOPALAN AND BARRY L. BAYUS

Lakhani, K., and E. von Hippel. 2003. “How Open Source Software Works: ‘Free’ User-to-User Assistance.” Research Policy 32 (6), 923–43. Lakhani, K., and R.G. Wolf. 2003. “Why Hackers Do What They Do: Understanding Motivation Efforts in Free/F/OSS Projects.” Working paper 4425–03, MIT Sloan School of Management. Lerner, J., and J. Tirole. 2002. “Some Simple Economics of Open Source.” Journal of Industrial Economics 52, 197–234. MacCormack, R., A. Verganti, and M. Lansiti. 2001. “Developing Products on ‘Internet Time’: The Anatomy of a Flexible Development Process.” Management Science, 47 (1), 133–50. Mockus, A., R. Fielding, and J. Herbsleb. 2000. “A Case Study of Open Source Software Development: The Apache Server.” Proceedings of the 22nd International Conference on Software Engineering. June, 263–72. ———. (2002). “Two Case Studies of Open Source Software Development: Apache and Mozilla.” ACM Transactions on Software Engineering and Methodology 11 (3), 309–46. Nielsen, J. 2006. “Participation Inequality: Encouraging More Users to Contribute.” October. Available at www.useit.com/alertbox/participation_inequality.html (accessed November 29, 2006). O’Reilly, T. 1999. “Lessons from Open-Source Software Development.” Communications of the ACM 42 (4), 32–37. Overby, E.M., A.S. Bharadwaj, and S.G. Bharadwaj. 2004. “An Investigation of Firm Level Open-Source Software Adoption: Theoretical and Practical Implications.” Working paper, Emory University, Atlanta. Paulson, J., G. Succi, and A. Eberlein. 2004. “An Empirical Study of Open-Source and Closed-Source Software Products.” IEEE Transactions on Software Engineering 30 (4), 246–56. Pitt, L., R. Watson, P. Berthon, D. Wynn, and G. Zinkhan. 2006. “The Penguin’s Window: Corporate Brands from an Open-Source Perspective.” Journal of the Academy of Marketing Science 34 (2), 115–27. Raymond, E. 1998. “The Cathedral and the Bazaar.” First Monday 3 (March). Available at www.firstmonday. org/issues/issue3_3/raymond/ (accessed August 10, 2004). Rossi, M.R. 2004. “Decoding the Free/Open Source (F/OSS) Software Puzzle: A Survey of Theoretical and Empirical Contributions.” Working paper, April. Available at opensource.mit.edu/papers/rossi.pdf (accessed August 15, 2005). Scacchi, W. 2003. “Understanding Open Source Software Evolution: Applying, Breaking, and Rethinking the Laws of Software Evolution.” Working paper, April. Available at opensource.mit.edu/papers/scacchi3. pdf (accessed August 10, 2004). Sharma, S., V. Sugumaran, and B. Rajagopalan. 2002. “A Framework for Creating Hybrid-Open Source Software Communities.” Information Systems Journal 12, 7–25. Shocker, A., B. Bayus, and N. Kim. 2004. “Product Complements and Substitutes in the Real World: The Relevance of Other Products.” Journal of Marketing 68 (January), 28–40. Utterback, J.M. 1994. Mastering the Dynamics of Innovation. Boston: Harvard Business School Press. von Hippel, E., and G. von Krogh. 2003. “Open Source Software and the Private-Collective Innovation Model: Issues for Organization Science.” Organization Science 14 (2), 209–23. von Krogh, G., and E. von Hippel. 2003. “Special Issue on Open Source Software.” Research Policy 32 (7), 1149–57. Wynn, D. 2004. “Organizational Structure of Open Source Projects: A Life Cycle Approach.” Proceedings of the 7th Annual Conference of the Southern Association for Information Systems. Available at sais.aisnet. org/sais2004/Wynn1.pdf (accessed August 15, 2004). Xu, J., and G. Madey. 2004. “Exploration of the Open Source Software Community.” Proceedings of the NAACSOS Conference, Pittsburgh. Zhao, L., and F. Deek 2004. “User Collaboration in Open Source Software Development.” Electronic Markets 14 (2), 89–103.

A NEW SPATIAL CLASSIFICATION METHODOLOGY FOR STP ANALYSIS

75

CHAPTER 4

A NEW SPATIAL CLASSIFICATION METHODOLOGY FOR SIMULTANEOUS SEGMENTATION, TARGETING, AND POSITIONING (STP ANALYSIS) FOR MARKETING RESEARCH WAYNE S. DESARBO, SIMON J. BLANCHARD, AND A. SELIN ATALAY

Abstract The Segmentation-Targeting-Positioning (STP) process is the foundation of all marketing strategy. This chapter presents a new constrained clusterwise multidimensional unfolding procedure for performing STP that simultaneously identifies consumer segments, derives a joint space of brand coordinates and segment-level ideal points, and creates a link between specified product attributes and brand locations in the derived joint space. This latter feature permits a variety of policy simulations by brand(s), as well as subsequent positioning optimization and targeting. We first begin with a brief review of the STP framework and optimal product positioning literature. The technical details of the proposed procedure are then presented, as well as a description of the various types of simulations and subsequent optimization that can be performed. An application is provided concerning consumers’ intentions to buy various competitive brands of portable telephones. The results of the proposed methodology are then compared to a naïve sequential application of multidimensional unfolding, clustering, and correlation/regression analyses with this same communication devices data. Finally, directions for future research are given. Introduction Competitive market structure analysis is an essential ingredient of the strategic market planning process (cf. Day 1984; Myers 1996; Wind and Robertson 1983). DeSarbo, Manrai, and Manrai (1994) describe the primary task of competitive market structure analyses as deriving a spatial configuration of products/brands/services in a designated product/service class on the basis of some competitive relationships between these products/brands/services (see also Fraser and Bradford 1984; Lattin and McAlister 1985). As mentioned in Reutterer and Natter (2000), to provide managers with more meaningful decision-oriented information needed for the evaluation of actual brands’ positions with respect to competitors, competitive market structure analyses have been enhanced with positioning models that incorporate actual consumers’ purchase behavior/intentions, background characteristics, marketing mix, etc. (cf. Cooper 1988; DeSarbo and Rao 1986; Green and Krieger 1989; Moore and Winer 1987). Thus, contemporary approaches for the empirical modeling of competitive market structure now take into account both consumer heterogeneity (e.g., market segmentation) and the competitive relationship of products/brands/services (e.g., 75

76

WAYNE S. DESARBO, SIMON J. BLANCHARD, AND A. SELIN ATALAY

Figure 4.1

The STP Process

Market Segmentation 1. Identify basis variables and segment the market 2. Develop profiles of resulting segments

Market Targeting 3. Evaluate attractiveness for each market segment 4. Select target segment(s)

Market Positioning 5. Identify positioning concepts for each target segment 6. Select, develop, and communicate the selected positioning concept(s)

Source: Modified from Kotler (1997).

positioning). Such analyses are integrated into the more encompassing Segmentation-TargetingPositioning (STP) approach (Kotler 1997; Lilien and Rangaswamy 2004). Figure 4.1, modified from Kotler (1997), illustrates the various steps or stages of this STP concept concerning implementation. Here, each of the three stages is depicted as discrete steps performed in a sequential manner. In stage I, segmentation, basis variables defined for segmentation (e.g., customer needs, wants, benefits sought, preference, intention to buy, usage situations, etc.) are collected depending on the specific application. Specification of these basis variables is important to ensure that the derived market segments are distinct in some managerially important manner (e.g., with respect to behavior). These data are then input into some multivariate procedure (e.g., cluster analysis) to form market segments. Myers (1996) and Wedel and Kamakura (2000) provide reviews of the various methodologies available for use in market segmentation, as well as their pros and cons. These derived market segments are then typically identified using profile variables that aid the firm in understanding how to serve these customers (e.g., shopping patterns, geographic location, spending patterns, marketing mix sensitivity, etc.), as well as how to communicate to these customer segments (e.g., demographics, media usage, psychographics, etc.). It is important that the derived market segments exhibit the various traits for effective segmentation (cf. DeSarbo and DeSarbo 2002; Kotler, 1997; Wedel and Kamakura, 2000): differential behavior, membership identification, reachability, feasibility, substantiality, profitability, responsiveness, stability, actionability, and predictability. In stage II, targeting, one evaluates the financial attractiveness of each derived market segment in terms of demand, costs, and competitive advantage. Based on this financial evaluation, one or more target segments are selected as “targets” based on the profit potential of such segments and their fit with the firm’s corporate goals and strategy. The “optimal” level of resources is then determined to allocate to these targeted market segments. As a final phase of this intermediate targeting stage, customers and prospects are often identified in these targeted segments. Kotler and Keller (2006) list five different patterns of target market selection, including single segment concentration (specialize in one segment), selective specialization (concentrate in S segments), product specialization (produce product variant in one major product class across all segments), market specialization (serve the complete needs in a specified market), and full market coverage (full coverage of all market segments and products). Finally, in stage III, positioning, the marketer identifies a positioning concept for the firm’s products/services that attracts targeted customers. Kotler and Keller (2006) define positioning as

A NEW SPATIAL CLASSIFICATION METHODOLOGY FOR STP ANALYSIS

77

“the act of designing the company’s offering and image to occupy a distinctive place in the mind of the target market” (p. 310). In such a framework where a firm targets one or more groups or market segments with its offerings, positioning then becomes a segment-specific concept. As noted by Wind (1980), most firms produce multiple products/services, and the positioning decision of any given product cannot ignore the place the product occupies in the firm’s product line as perceived by the consumer. The product offering of a firm should lead to an optimal mix of product positioning by market segments, that is, the product positioning of any given product should not be designed and evaluated in isolation from the positioning of the firm’s other products and the market segments at which they are aimed (p. 70). In addition, the focus of a product line positioning should not be limited to the firm’s own products but rather should take explicitly into account the product lines of its competitors as well. Empirical modeling approaches in the STP area have taken typically two forms: (1) sequential use of multidimensional scaling and cluster analysis, and (2) parametric finite mixture or latent class multidimensional scaling models. The more traditional approaches have employed the sequential use of multidimensional scaling (MDS) and cluster analysis to first spatially portray the relationship between competitive brands and consumers, and then segment the resulting consumer locations to form market segments. As mentioned in DeSarbo, Grewal, and Scott (2008), a number of methodological problems are associated with such a naïve approach. One is that each type of analysis (multidimensional scaling and cluster analysis) typically optimizes different loss functions. (In fact, many forms of cluster analysis optimize nothing.) As such, different aspects of the data are often ignored by the disjointed application of such sequential procedures. Two, there are many types of multidimensional scaling and cluster analysis procedures, as documented in the vast psychometric and classification literature, where each type of procedure can render different results (cf. DeSarbo, Manrai, and Manrai 1994). In addition, there is little a priori theory to suggest which methodological selections within these alternatives are most appropriate for a given marketing application. And the various combinations of types of multidimensional scaling and cluster analyses typically render different results as well. Finally, as noted by Holman (1972), the Euclidean distance metric utilized in many forms of multidimensional scaling is not congruent with the ultrametric distance formulation metric utilized in many forms of (hierarchical) cluster analysis. Efforts to resolve such methodological problems associated with this naïve sequential administration of multidimensional scaling and cluster analyses have resulted in the evolution of parametric finite mixture or latent class multidimensional scaling models (cf. DeSarbo, Manrai, and Manrai 1994; Wedel and DeSarbo 1996). In particular, latent class multidimensional scaling models for the analysis of preference/dominance data have been proposed by a number of different authors over the past fifteen years employing either scalar products/vector (Slater 1960; Tucker 1960) or unfolding (Coombs 1964) representations of the structure in two-way preference/dominance data. In such latent class multidimensional scaling models, vectors or ideal points of derived segments are estimated instead of separate parameters for every individual consumer. Thus, the number of parameters is significantly reduced relative to individual-level models. Latent class multidimensional scaling models are traditionally estimated using the method of maximum likelihood (E-M algorithms [cf. Dempster, Laird, and Rubin 1977] are typically employed). For example, DeSarbo, Howard, and Jedidi (1991) developed a latent class multidimensional scaling vector model (MULTICLUS) for normally distributed data. DeSarbo, Jedidi, Cool, and Schendel (1990) extended this latent class multidimensional scaling model to a weighted ideal point model. De Soete and Heiser (1993) and De Soete and Winsberg (1993), respectively, extended the MULTICLUS latent class multidimensional scaling model by accommodating linear restrictions on the stimulus

78

WAYNE S. DESARBO, SIMON J. BLANCHARD, AND A. SELIN ATALAY

coordinates. Böckenholt and Böckenholt (1991) developed simple and weighted ideal-point latent class multidimensional scaling models for binary data. DeSarbo, Ramaswamy, and Lenk (1993) developed a vector latent class multidimensional scaling model for Dirichlet-distributed data, and Chintagunta (1994) developed a latent class multidimensional scaling vector model for multinomial data. Wedel and DeSarbo (1996) extended the entire family of exponential distributions to such latent class multidimensional scaling models for two-way preference/dominance data. As noted in DeSarbo, Grewal, and Scott (2008), such latent class multidimensional scaling models also have a number of limitations associated with them. One, they are parametric models that require the assumption of specific distributions. Oftentimes, continuous support distributions are utilized in the finite mixture and are applied to traditional discrete response scales (e.g., semantic differential, Likert, etc.), which cannot possibly obey such distributional assumptions. As such, violations of such distributional assumptions may invalidate the use of the procedure. Two, such latent class multidimensional scaling procedures require the underlying finite mixture to be identified (cf. McLachlan and Peel 2000), which is often problematic when the underlying distributions deviate from the exponential family. When multivariate normal distributions are employed, identification problems can arise in the estimation of separate full covariance matrices by derived market segment. Three, most of the latent class multidimensional scaling procedures are highly nonlinear in nature and require very intensive computation. Such procedures typically utilize an E-M approach (cf. McLachlan and Krishnan 1997) or gradientbased estimation procedures, which may take hours of computation time for a complete analysis to be performed. Four, at best, only locally optimum solutions are typically reached, and the analyses have to be repeated several times for each value of the dimensionality and number of groups. Five, the available heuristics employing various information criteria typically result in different solutions being selected. For example, the BIC and CAIC are considered as more conservative measures resulting in the selection of fewer dimensions and groups in contrast to AIC and MAIC, which are considered to be more liberal criterion (cf. Wedel and Kamakura 2000). As discussed in Wedel and Kamakura (2000), there are still other heuristics utilized for model selection for such finite mixture models (e.g., ICOMP, NEC, etc.), which are equally plausible but might also result in different solutions being selected. Finally, the underlying framework assumed by these latent class multidimensional scaling procedures involves a partitioning of the sample space, although the estimated posterior probabilities of membership often result in fuzzy probabilities of segment membership, which could be difficult to interpret or justify in applications requiring partitions. In fact, pronounced fuzziness of the resulting posterior probabilities of segment membership may be indicative of poor separation of the conditional support functions’ centroids. Our goal in this research is to contribute to current STP research by devising a new deterministic, clusterwise multidimensional unfolding procedure for STP that overcomes some of the abovementioned limitations of the current procedures. Our goal is to simultaneously derive a single joint space where derived “segments” are also determined and represented by ideal points, and brands via coordinate points, and their interrelationship in the space denotes some aspect of the structure in the input preference/dominance data. In addition, our approach (like the GENFOLD2 multidimensional unfolding procedure of DeSarbo and Rao [1986]) explicitly relates brand attributes to brand locations in the derived joint space. The proposed approach does not require distributional assumptions, such as latent class multidimensional scaling procedures, and it provides a concise spatial representation for the analysis of preference/dominance data, as will be illustrated in our application to buying intentions for portable telephone communication devices. In the proposed model, no finite mixture distribution identification is required, unlike its latent class multidimen-

A NEW SPATIAL CLASSIFICATION METHODOLOGY FOR STP ANALYSIS

79

sional scaling counterparts. We generalize the DeSarbo, Grewal, and Scott (2008) clusterwise vector model to the unfolding case (see also DeSarbo, Atalay, LeBaron, and Blanchard 2008) with reparametrization of the derived brand space. The estimation procedure developed is relatively fast and efficient, and it converges in a matter of minutes on a PC (although potential locally optimum solutions are possible here also). Finally, our proposed procedure accommodates both overlapping segments (cf. Arabie et al. 1981), as well as nonoverlapping segments. These two segmentation schemes account for consumer heterogeneity within the data. While nonoverlapping segment structures exemplify the separation in preference between segments, overlapping segments acknowledge that individuals may have multiple needs or goals and therefore be members of multiple segments (cf. DeSarbo, Atalay, LeBaron, and Blanchard 2008). In this chapter, we propose a method that simultaneously identifies segments of consumers and relates product characteristics to the derived product space. The segment-level ideal point representation allows for a parsimonious representation of preferences, and the relation between product characteristics and the attribute space eases the understanding of the brand space and facilitates the generation of actionable new products. In the following section, we review the major trends in optimal product positioning research in the area of STP analysis. Then, we present the clusterwise multidimensional unfolding model (with reparametrization of the attributes) in an optimal product positioning framework for STP. Following this, a comparison with a traditional sequential procedure is presented (MDS and cluster analysis) in terms of an application based on communication devices research data initially presented in DeSarbo and Rao (1986). Finally, we discuss several avenues for future research. Optimal Positioning in STP Most recent methods of optimal product positioning operate under the assumption that competition between brands in a designated product/service class can be adequately represented in a joint space map. The preferences of consumers are often represented in the derived joint space map as ideal points that denote consumers’ ideal brands: the brands that would maximize their utility and that are most likely to be chosen if they were to be offered in the market. Such maps also help illustrate the relationships between the brands and their attributes. As the most basic models assume that an individual’s preference is inversely related to the distance between the brands and his/her ideal point, the closer an ideal point is to a brand in the derived space, the greater is the individual’s preference for that brand. The two most common methods for obtaining such cognitive spaces are multidimensional unfolding (MDU) and conjoint analysis (CA). Baier and Gaul (1999) offer comprehensive lists of the studies in both MDU and CA in the context of optimal product positioning and design. These methods generally focus on either product positioning or product design, which have consistently been confused in the literature (see Kaul and Rao [1995] for a detailed review). Product positioning models use abstract perceptual dimensions (e.g., quality, prestige, beauty, etc.), while product design models use more specific product characteristics or features (e.g., color, price, size). Whereas product design models often miss the impact of the marketing mix variables, the products generated from positioning models are often difficult to execute in practice. The need of integration between derived dimensions and attributes has been stressed in several models (Baier and Gaul 1999; DeSarbo and Rao 1986; Hauser and Simmie 1981; Michalek, Feinberg, and Papalambros 2005; Shocker and Srinivasan 1974). This is critical because specific marketing strategies and actions must be determined to obtain desired or optimal positions (Eliashberg and Manrai 1992).

80

WAYNE S. DESARBO, SIMON J. BLANCHARD, AND A. SELIN ATALAY

A model for generating the choice behaviors of the consumers must also be specified. The first generation of optimal positioning models assumed a deterministic model of choice in which consumers would choose the brand that they most prefer with certainty (Albers 1979; Albers and Brockhoff 1977; Gavish, Horsky, and Srikanth 1983; Zufryden 1979). In the context of ideal point models, this suggests that consumers always choose the brand that is positioned closest to their ideal points. The implicit assumption here was that these methods were to be used for choice behaviors about infrequently chosen objects and durables for which consumers have been assumed to select the best option with certainty. These models also assume that the consumers have strongly held preferences and render no consideration to situational (e.g., different uses, environmental changes) or contextual (e.g., variety-seeking tendencies, mood) factors that also may impact choice. Other approaches followed a probabilistic choice rule where products have a nonzero probability of being chosen depending on their distance to the consumer ideal points (e.g., Bachem and Simon 1981; Gruca and Klemz 2003; Pessemier et al. 1971; Shocker and Srinivasan 1974). Generally, the inverse of the distance between a product and an ideal point is taken as a proxy for the probability of that product’s being chosen. These probabilistic models have been found, in simulations, to provide somewhat better solutions than the single choice deterministic models (Shocker and Srinivasan 1979; Sudharshan, May, and Shocker 1987). Note that this difference between deterministic and probabilistic models generating choice is not to be confused with that between nonstochastic and stochastic models. The latter implies that the model parameters such as the location of the ideal points and the brand positions are random variables drawn from pre-specified distributions (e.g., Baier and Gaul 1999). Both stochastic and nonstochastic methods are generally subject to Luce’s axiom (Luce 1959), which relies on often difficult assumptions to justify in the product positioning context (Batsell and Polking 1985; Bordley 2003; Meyer and Johnson 1995). Early optimization models maximized market share for the new brand based on predicted single choices (Albers 1979; Albers and Brockhoff 1977; Shocker and Srinivasan 1974). These models did not explicitly consider price/cost information (Hauser and Simmie 1981; Schmalensee and Thisse 1988). Several authors followed with profit formulations (Bachem and Simon 1981; Choi, DeSarbo, and Harker 1990; Green, Carroll, and Goldberg 1981; Green and Krieger 1985; DeSarbo and Hoffman 1987). Others improved realism by adding budget constraints (Thakur et al. 2000), by incorporating competitive reactions (Choi, DeSarbo, and Harker 1990), by adding variable costs (Bachem and Simon 1981; Gavish, Horsky, and Srikanth 1983), or by implementing a more complete cost structure that involves variable development costs (Bordley 2003). Even with these developments, all these models are highly dependent on the quality of the brand space and preference representations that they obtain in the first step, especially since degenerate solutions haunt most MDU procedures (cf. DeSarbo and Rao 1986). All such optimal product approaches are also limited by the difficulties associated with the collection of cost information (Green, Carroll, and Goldberg 1981). In addition to these modifications, these optimal positioning models also differ in a number of other ways. It was argued that because of self-interest, individuals are likely to be more familiar with products they more highly prefer. DeSarbo, Ramaswamy, and Chatterjee (1995) examine the effect of differential familiarity in MDS models. Models should also ideally consider that the chosen product will vary depending on the situation and on individual factors; it will be chosen from a reduced set of options that depends partly on the type of usage that is intended. To account for this, some authors used a reduced consideration set of the k-closest options to their ideal points (Gavish, Horsky, and Srikanth 1983; Gruca and Klemz 2003; Sudharshan, May, and Shocker 1987).

A NEW SPATIAL CLASSIFICATION METHODOLOGY FOR STP ANALYSIS

81

However, the assumption that some products can have a zero probability of being chosen seems excessively strong (Pessemier et al. 1971; Schmalensee and Thisse 1988). These optimal positioning models also differ in their computational complexity. The initial models of Albers et al. (1977, 1979) are limited by the number of ideal points. As such, many authors suggested that grouping consumers into market segments could lead to reduced computational complexity (McBride and Zufryden 1988; Schmalensee and Thisse 1988). Zufryden (1982) proposed grouping individuals based on their utilities and on their usage intentions, whereas Shugan and Balachandran (1977) suggested forming segments based on the preference orders a priori. The way that the segments are created is especially important, as it should be closely related to the choice behavior at hand. Two-step models, where segmentation is obtained after or before the ideal points and the brand space are obtained, are less efficient than methods that simultaneously cluster individuals based on their ideal points and their perceived brand structure (see DeSarbo, Manrai, and Manrai 1994). Additionally, most of the optimal positioning methods follow a two-step structure (the brand space and ideal points are first identified before any segmentation or the optimal product positioning stage). The latter is advantageous because the positioning step does not depend on the type of data that is initially used to obtain the ideal points (Shocker and Srinivasan 1979). This has resulted in the use of a variety of algorithms and input data, such as paired comparisons and rank-order preferences, among others. However, once the brand positions and the ideal points are obtained, data considerations about the attributes and characteristics become increasingly important. Some attributes are inherently discrete, such as the presence of the air-conditioning option in a vehicle. Others, such as price, miles per gallon, and quality, represent continuous information. The models for optimal positioning have been limited by the type of data that they can deal with, partly because of the difficulties in using various types in optimization routines, but also because of the difficulties in getting the costs associated with the various attributes and characteristics. To optimize profit functions, it is important that product characteristics and elements of the marketing mix be quantifiable in terms of costs. This can be difficult with some attributes, which lead some to the discretization of continuous variables at levels for which product costs are known. This also simplified the optimization, although it precluded traditional (continuous) gradient procedures from being used in subsequent optimization. Even when the attribute/characteristics issues are resolved, not all combinations of characteristics may be feasible, and there is a need for models that can incorporate constraints on feasibility (Shocker and Srinivasan 1979). As previously mentioned, it is essential that attribute-based propositions can be translated into a feasible product space that can be successfully produced by the firm (Gavish, Horsky, and Srikanth 1983). Finally, the variants of the optimization problems are often difficult to solve for. The gradient procedures (when the attributes are continuous), given the nonconvex nature of the problem, are subject to numerous local optima. The mixed-integer nonlinear problems are NP-hard. Optimization procedures that have been used to improve on performance and computational times include genetic algorithms (Gruca and Klemz 2003), branch and bound (Hansen et al. 1998), dynamic programming (Kohli and Krishnamurti 1987), and nested partition methods (Shi, Olafsson, and Chen 2001). Given the above variants regarding optimal product positioning in STP problems, we propose a clusterwise multidimensional unfolding procedure that simultaneously identifies segments of consumers that have similar ideal products, and that reparameterizes the brand space to be a linear function of actionable product characteristics. The segment ideal points, the obtained brand

82

WAYNE S. DESARBO, SIMON J. BLANCHARD, AND A. SELIN ATALAY

space, and reparametrization coefficients can then be used to generate and select the best new product to introduce in the market. The clusterwise multidimensional unfolding procedure that we propose is a direct generalization of GENFOLD2 (DeSarbo and Rao 1986) to the clusterwise case. It presents several advantages over previous methods. As mentioned, it is important to select segments of consumers who follow a meaningful behavioral/choice pattern. By integrating the segmentation stage into the model, the model optimizes the real underlying objective to find the segments that have the most similar preference structure. It is also independent of all parametric assumptions. Finally, it provides a simple segment-level ideal point structure that will be computationally easier at the product optimization stage as the computation is greatly influenced by the number of ideal points. The optimization step that we propose is an extension of Shocker and Srinivasan’s (1974) “distance as probability” model for discretrized product characteristics, with a profit function and with segment-level ideal points. The brand reparametrization feature permits the link between derived brand coordinates and brand attributes, as it is important to suggest optimal brands that will be realizable, because the dimensions (brand coordinates) are obtained as a linear function of the brand attributes. Using this reparametrization, one now has a way not only to better interpret the dimensions but also to examine the effect in the derived joint space of altering attribute values for repositioning existing brands or positioning new brands. Finally, the optimal product-positioning step identifies and selects, through a probabilistic nonstochastic model, the product to introduce that will maximize a company’s profit function. Because of the reduction of the number of ideal points in the first step, it is possible to proceed to the complete enumeration of the realizable products and to select the globally optimal new product to introduce in the market. The Proposed Clusterwise Unfolding Model Clusterwise models are common in the marketing and psychometric literatures. These methods simultaneously group individuals into segments to represent sample heterogeneity and derive segment-level parameters for the particular model being estimated. Several clusterwise formulations have been proposed for the regression problem (DeSarbo, Oliver, and Rangaswamy 1989; Spath 1979, 1981, 1982; Wedel and Kistemaker 1989). Here, we develop a clusterwise multidimensional model with brand reparametrization options as in GENFOLD2 (DeSarbo and Rao 1986). Let: i = 1, . . . , N consumers; j = 1, . . . , J brands; k = 1, . . . , K brand attributes (K < J); s = 1, . . . , S market segments (unknown); r = 1, . . . , R dimensions (unknown); ij = the dispreference for brand j given by consumer i. We model the observed data as:

6

5

V

U

 LM  D  3LV  ; MU