THE IMPACT OF ONLINE SEARCHES ON CONSIDERATION SET FORMATION AND CONSUMER CHOICE Zhao Hongyu (Master in Economics, Fudan University) A THESIS SUBMITTED FOR THE DEGREE OF PHD OF MARKETING DEPARTMENT OF MARKETING NATIONAL UNIVERSITY OF SINGAPORE 2008 Acknowledgements I am indebted to a great number of people who generously offered advice, encouragement, inspiration and friendship through my time at National University of Singapore. First of all, I would dedicate my deepest gratitude to my supervisors, Professor Jeongwen Chiang who introduced me into a challenging but interesting research area and provided the initial guidance for my dissertation work, and Professor Surendra Rajiv who provide vital input and invaluable insights for my research. I can not imagine the completion of this work without their encouragement and support. Thank you to the other faculty members who have helped me along the way. Professor Juinkuan Chong who is my mentor for coursework and also a good advisor for research work. Professor Junhong Chu who has been a friend and generously shared her personal experience of doing research. Professor Trichy Krishnan who always encouraged me by showing all-time enthusiasm towards academic research. Thank you to all the friends and colleagues who have made my PhD student life a whole lot more enjoyable. Cheng and Zhixing for spending time discussing topics varying from research to human nature and providing regular doses of encouragement. Shangfei, Suman and Sun Li for their availability and patience in i listening to my research ideas which very often turned out to be just mindless talks. Finally, thanks to my family who provided me with more love and support than I would otherwise think possible. To my mother who knows nothing about academic research but never questioned my decision, determination and capability of pursuing high education. To my father who has been influencing and supporting my academic pursuits as long as I can remember, and who has tremendously influence on who I am. Though he is not around to see my complete my education, but he would have been proud. ii Table of Contents TABLE OF CONTENTS .III SUMMARY V LIST OF TABLES VII LIST OF FIGURES .VIII LIST OF SYMBOLS IX CHAPTER 1: INTRODUCTION . CHAPTER 2: LITERATURE REVIEW . 2.1 2.2 2.3 CONSUMER CHOICE . CONSIDERATION SET . INTERNET USE AND ITS IMPACT ON CONSUMER CONSIDERATION AND CHOICE 16 CHAPTER 3: MODELS AND ESTIMATION 24 3.1 DECISION PROCESS AND MODEL ASSUMPTIONS 25 3.1.1 Conceptual description of the decision process . 25 3.1.2 Model assumptions 27 3.2 MODEL FORMULATION . 29 3.2.1 Information on Search Attributes and Category Consideration 29 3.2.2 Information on Experience Attributes and Model Consideration 34 3.2.3 Choice Decision . 40 3.2.4 Overall Decision 41 3.3 MODEL IDENTIFICATION . 42 3.3.1 Identification of Multivariate Probit Model 42 3.3.2 Identification of the Multinomial Probit Model . 44 3.3.3 Identification of the Multi-Stage Multivariate and Multinomial Probit Models . 45 3.4 DRAWING ALGORITHM 48 CHAPTER 4: DATA 53 4.1 J.D. POWER 2001 VEHICLE SHOPPING SURVEY DATA . 54 4.1.1 The Luxury Segment 55 4.1.2 Consumer Demographics 60 4.1.3 Consideration Set and Internet Use . 61 4.2 VEHICLE ATTRIBUTE DATA 66 CHAPTER 5: EMPIRICAL FINDINGS 69 5.1 VARIABLES DESCRIPTION 70 iii 5.2 MODEL COMPARISON 76 5.2.1 Decision Structure Comparison . 76 5.2.2 Internet Effect Comparison 79 5.3 MODEL ESTIMATES . 81 5.3.1 Internet Effects . 82 5.3.2 State Dependence . 84 5.3.3 Consumer’s Attribute Preference . 86 5.3.4 Other Variables Affecting Consideration Cost 87 5.3.5 Unobservable Factors . 87 5.4 MANAGERIAL IMPLICATIONS . 89 5.4.1 Consideration Set Entropy . 89 5.4.2 Consideration Set Composition . 92 CHAPTER 6: CONCLUSION 99 6.1 6.2 6.3 6.4 SUMMARY . 99 CONTRIBUTIONS . 100 RESEARCH DIRECTIONS 102 CONCLUSION 106 BIBLIOGRAPHY 108 APPENDIX I: FULL CONDITIONAL POSTERIOR DISTRIBUTIONS . 120 APPENDIX II: FIGURES AND TABLES . 124 iv Summary The Internet has been widely adopted for product information search. We have substantial understanding that the availability of the low-cost price information on the Internet can increase the consumer’s price sensitivity and drive down the market price. However, we lack the general view of how the use of the Internet, especially the search for non-price information, would affect consumer choice. Therefore, we extend the extant literature to investigate the impact of the online information search on consumer choice decision, especially the consideration set formation decision. We also examine the heterogeneity in the websites in terms of the types of information they deliver and how the difference would have distinct effects on the individual consideration and choice decision. We apply the multivariate Probit model to model consumer’s consideration decision. The empirical evidences we find from the JD Power New Vehicle Shopping Survey data show that a) the use of the Internet to search for vehicle information leads to more diversified consideration set; c) the diversification of the consideration set due to the Internet use is because of the increase in considering unfamiliar vehicle categories and models; c) the Internet is not homogeneous in terms of its influence on the consideration decision. The literature has a lot of discussion on how the low-cost online price information can influence consumer choice. We further the research in the area by v examining the influence of both online price and non-price information on consumer’s consideration and choice decision. We also differentiate the effects of different types of websites. The proposed 3-stage choice decision model also contributes to the choice model literature by explicitly modeling a consumer’s decision to search for search attributes and experience attributes. vi List of Tables Table 2.1 Marketing Literature on Internet Impact…………………………… 22 Table 4.1 Segment of Replaced, Considered and Purchased Vehicles …………56 Table 4.2 Vehicle Models in the Luxury Car Segment………………………….57 Table 4.3 Internet Use for Automotive Information Search – Total Replacement Sample ………………………………………………………………………….58 Table 4.4 Internet Use for Automotive Information Search – Luxury Car Only Sample………………………………………………………………………… 58 Table 4.5 Statistics on Consumer Demographics – Total Replacement Sample 59 Table 4.6 Statistics on Consumer Demographics – Luxury Car Sample……… 59 Table 4.7 Consideration Set Size……………………………………………… 62 Table 4.8 C-Set Composition Comparison I – Auto Internet User vs. Non-User (Luxury Car Only Sample) ……………………………………………… ……63 Table 4.9 C-Set Composition Comparison II – Independent vs. Manufacturer Site (Luxury Car Only Sample)………………………………………………… .…65 Table 4.10 Vehicle Attribute Summary Statistics……………………………….67 Table 5.1 Rotated Factor Loadings for Vehicle Shopping Characteristics…… .74 Table 5.2 Competing Models – Decision Structure…………… ………………77 Table 5.3 Consideration Set Component Hit Rate – Decision Structure Models.78 Table 5.4 Choice Hit Rate – Decision Structure Models …………….……… .78 Table 5.5 Competing Models – Internet vs. No Internet……………………… .80 Table 5.6 Consideration Set Component Hit Rate – Internet vs. No Internet Models………………………………………………………………………… .80 Table 5.7 Choice Hit Rate – Internet vs. No Internet Models………………… .80 vii List of Figures Figure 2.1 The Internet-Related Choice Decision Structure . 21 Figure 3.1: Mechanism for Drawing Category and Model Consideration Latent Utilities 52 Figure 5.1: Average C-Set Entropy Values for Different Internet Usage Scenarios . 91 Figure 5.2: Number of Categories Considered for Different Internet Usage Scenarios . 92 Figure 5.3: Consideration Probability of Vehicle Categories for Different Internet Usage Scenarios 93 Figure 5.4: Consideration Probability of Vehicle Categories among American Car Replacements 95 Figure 5.5: Consideration Probability of Vehicle Categories among European Car Replacements 95 Figure 5.6: Consideration Probability of Vehicle Categories among Asian Car Replacements 95 Figure 5.7: Average C-Set Entropy Values for Different Independent Site Visit Scenarios . 96 Figure 5.8: Share of Vehicle Models from Different Categories for Manufacturer Site Visit Scenarios . 97 viii List of Symbols i denotes an individual consumer m denotes a vehicle category, m = 1,… , M j denotes a vehicle alternative, j = 1,… , J Category Consideration Ci1 is consumer i ’s category consideration set cim is an indicator of consumer i ’s consideration of category m (equals to if category m is considered, and otherwise) 1* cim is the perceived net utility of considering category m by consumer i 1* is the perceived utility of considering category m by consumer i Utilityim 1* Costim is the perceived cost of considering category m by consumer i Im is a vector of category-specific dummies X m1 is a vector of attribute values of category m Costim is a vector of explanatory variables which affect perceived consideration cost γ is a vector of category-specific constant term parameters γi is a vector of category-specific constant term parameters for consumer i β is a vector of attribute preference parameters βi is a vector of attribute preference parameters for consumer i ρ is a vector of cost factor parameters Di consumer i ’s demographics Zi consumer i ’s shopping characteristics ε im multivariate-normal distributed variables at category consideration stage ix Marketing, Vol. 12, No. 1, 3-7. Roberts, John H. and Glen L. Urban (1988), “Modeling Multiattribute Utility, Risk, and Belief Dynamics for New Consumer Durable Brand Choice,” Management Science, Vol. 34, No. 2, 167-85. Roger O. Crockett (1999), “Heard Any Good Computer Files Lately? If the Record Business is to Thrive, it Must Embrace the Digital-Music Format,” Business Week (Sept. 27), EB16. Rosecky, Richard B and Algin B King (1996), “Perceptual Differences among Owners of Luxury Cars: Strategic Marketing Implications,” The Mid - Atlantic Journal of Business, Vol. 32, No. 3, 221-40. Rossi, Peter, Greg Allenby and Rob McCulloch (2005), Bayesian Statistics and Marketing, Hoboken, NJ: Wiley. Scott Morton, Fiona, Florian Zettelmeyer and Jorge Silva-Risso (2001), “Internet Car Retailing”, Journal of Industrial Economics, Vol. 49, No. 4, 501-20. Shocker, Allan D., Moshe Ben-Akiva, Bruno Boccara and Prakash Nedungadi (1991), “Consideration Set Influences on Consumer Decision-Making and Choice: Issues, Models, and Suggestions,” Marketing Letters, Vol. 2, No. 3, 181-97. Shugan, Steven M. (1980), “The Cost of Thinking,” Journal of Consumer Research, Vol. 7, No. 2, 99-111. Sismeiro, Catarina and Randolph E. Bucklin (2004), “Modeling Purchase 117 Behavior at an E-Commerce Web Site: A Task-Completion Approach,” Journal of Marketing Research, Vol. 41, No. 3, 306-23. Stigiler, Geogre J. (1961), “The Economics of Information”, Journal of Political Economy, Vol. 69, No. 3, 213-25. Tanner, Martin and Wing Hung Wong (1987), “The Calculation of Posterior Distributions by Data Augmentation,” Journal of the American Statistical Association, Vol. 82, No. 398, 528-40. Telang, Rahul, Peter Boatwright and Tridas Mukhopadhyay (2004), “A Mixture Model for Internet Search-Engine Visits,” Journal of Marketing Research, Vol. 41, No. 2, 206-14. Urban, Glen L., John R. Hauser and John H. Roberts (1990), “Prelaunch Forecasting of New Automobiles,” Management Science, Vol. 36, No. 4, 401-21. Wedel, Michel and Jie Zhang (2004), “Analyzing Brand Competition Across Subcategories,” Journal of Marketing Research, Vol. 41, No. 4, 448-56. Wu Jianan and Arvind Rangaswamy (2003), “A Fuzzy Set Model of Search and Consideration with an Application to an Online Market,” Marketing Science, Vol. 22, No. 3, 411-34. Yang, Sha, Greg M. Allenby and Geraldine Fennell (2002), “Modeling Variation in Brand Preference: The Roles of Objective Environment and Motivating Conditions,” Marketing Science, Vol. 21, No. 1, 14-31. 118 Zettelmeyer, Florian (2000), “Expanding to the Internet: Pricing and Communications Strategies When Firms Compete on Multiple Channels,” Journal of Marketing Research, Vol. 37, No. 3, 292-308. Zettelmeyer, Florian, Fiona Scott Morton and Jorge Silva-Risso (2002), “Cowboys or Cowards: Why are Internet Car Prices Lower?” Working Paper, University of California at Berkeley, Berkeley, CA. Zettelmeyer, Florian, Fiona Scott Morton and Jorge Silva-Risso (2006), “How the Internet Lowers Prices: Evidence from Matched Survey and Automobile Transaction Data,” Journal of Marketing Research, Vol. 43, No. 2, 168-181. 119 Appendix I: Full Conditional Posterior Distributions 1) α | Σ1(t −1) , Σ 2(t −1) , Σ3( t −1) (α ) | Σ ~ α trace( S Σ −1 ) χ v2*( M + J ) where α Σ1  is a positive constant, Σ =  Σ 0  0   , S is the prior scale of Σ3  Σ , v is the prior degree of freedom of Σ , and M and J are the dimensions of vehicle category and model consideration respectively. 2) ci1*(t ) | Ci2 , ci2*(t −1) , Θ(t −1) , Σ1(t −1) To draw the unidentified category consideration utilities ci1* , we first draw the identified utility vector ci1* and then rescale it with the working parameter (for details, see Imai and van Dyk 2005). Instead of directly sampling from the truncated multivariate normal distribution ci1* ~ TruncatedMVN (Θi X i1 , Σ1 ) , we reiteratively draw from the truncated 1* univariate normal distributions cim | ci1*,− m ~ TN ( µim ,τ im2 ) for all categories. Here µim and τ im2 are the normal mean and variance for category m conditional on all the other categories (for details of how to derive the mean and variance of univariate normal distribution from multivariate normal distribution, see McCulloch and Rossi 1994). Start with m = , 1* 1* If Σ Cij2 > , draw cim | ci1*,− m ~ TN ( µim ,τ im2 ){I (cim ≥ 0)} from j∈m upper-truncated normal. That is, if there are vehicle models from category 120 m being considered by consumer i , then draw the category utility from the area larger than zero. If Σ Cij2 = , we have to differentiate two conditions. If j∈m 1* cij2*(t −1) < for ∀j ∈ m , draw cim | ci1*,− m ~ N ( µim ,τ im2 ) from non-truncated 1* 1* normal; otherwise draw cim | ci1*,− m ~ TN ( µim ,τ im2 ){I (cim < 0)} from lower-truncated normal. Increment m and return to top. 1* Once draw cim for all categories, set ci1* = α 0ci1* . 3) ci2*(t ) | Ci2 , ci1*(t ) , Θ( t −1) , Σ 2( t −1) Same as drawing for the category utilities, we draw the alternative utilities reiteratively from truncated univariate normal distributions ( ) cij2* | ci2*,− j ~ TN µij ,τ ij2 . Since the alternatives from different categories are uncorrelated, the normal means and variances µij and τ ij2 are only conditional on the means and variances of alternatives from the same category. The truncation points of the alternative utility distributions also depend on the draws of the category utilities in step 2. For ∀j from m , 1*( t ) If cim ( j ) ≥ , that is the category m which alternative j belongs to is considered by consumer i , we draw cij2* from truncated normal distribution with truncation point determined by the consideration outcome. If Cij2 = , draw cij2* | ci2*,− j ~ TN ( µij ,τ ij2 ){ I (cij2* ≥ 0)} from upper-truncated normal. However, if Cij2 = , draw ( cij2* | ci2*,− j ~ TN µij ,τ ij2 ){I (c 2* ij < 0)} from lower-truncated normal. 121 1*( t ) If cim ( j ) < , that is the category m which alternative j belongs to is not considered by consumer i , we then draw cij2* from full distribution ( cij2* | ci2*,− j ~ N µij ,τ ij2 ) since the utility bound has been taken care at the category consideration. Increment j and return to top. Once draw cij2* for all alternatives, set ci2* = α 0ci2* . 4) ui(t ) | Ci2 , Ci3 , Θ(t −1) , Σ3( t −1) The choice utility follows truncated normal distribution. If yij = 1, draw uij from upper-truncated normal distribution { } uij | ui ,− j ~ TN ( µiju ,τ ij( u )2 ) I ( uij > max(ui ,− j , ∀(− j ) which Ci2( − j ) = 1) ) . If yij = 0, draw { } uij | ui ,− j ~ TN ( µiju ,τ ij(u )2 ) I ( uij < uik where (yik = & Cik2 = 1) ) . Once draw uij for all alternatives, then set ui = α 0ui . 5) Θ(t ) , α ( t ) | ci1*( t ) , ci2*(t ) , ui( t ) , Σ1( t −1) , Σ 2( t −1) , Σ3(t −1) First draw (α )(t ) from (α )(t ) {∑ ~ n i =1 (W *( t ) i ˆ (t ) − X iΘ ' −1 ) ( Σ ) (W ( t −1) i *( t ) } ) ˆ (t ) + Θ ˆ (t )' AΘ ˆ (t ) + trace α S (Σ (t −1) )−1  − X iΘ   χ (2n + v )( M + J ) then draw Θ(t ) from  ˆ (t ) Θ( t ) ~ N  Θ , (α )(t )  ( ∑ i =1 X i' Σ(t −1) n ( ) −1 Xi + A ) −1    122 where Wi *( t )  ci1*(t )  Σ1( t −1)    =  ci2*(t )  , Σ ( t −1) =   ui(t )      ˆ ( t ) = ∑ n X ' Σ(t −1) Θ  i =1 i ( ) −1 −1 Σ 2( t −1) n X i + A ∑ i =1 X i' Σ(t −1)   (   , Σ3( t −1)  ) −1 Wi *(t )  ,  6) Set Θ(t ) = Θ( t ) / α (t ) 7) Σ1( t ) , Σ 2(t ) , Σ3(t ) | ci1*( t ) , ci2*( t ) , ui(t ) , Θ(t ) , Θ(t ) Draw Σ(t ) from inverse Wishart distribution ' n Σ ( t ) ~ IW  n + v, α 02 S + ∑ i =1 Wi *( t ) − X i Θ( t ) Wi *( t ) − X i Θ(t )    ( 8) )( ) Rescale Σ1(t ) , Σ 2( t ) , Σ3( t ) , ci1*(t ) , ci2*(t ) , ui( t ) with σ 112(t ) 9) Set t = t + and go to step (1) 123 Appendix II: Figures and Tables ③ External Search Retailer Competition Retailer Choice ② Offline Channels ⑤ ① Brand Choice ④ ⑧ ⑦ Internet ⑥ ⑨ Consideration Set Manufacturer Competition ⑩ Figure 2.1 The Internet-Related Choice Decision Structure Table 2.1: Marketing Literature on Internet Impact Information Search Channel Consumer Choice ① Klein & Ford (2003); Ratchford, Lee & Talukdar (2003) Choice Only ④ Degeratu, Rangaswamy & Wu (2000) Consideration ⑥ ⑦ Wu & Rangaswamy (2003); and Choice ⑥ ⑨ and ⑥ ⑦ ⑧ Proposed Model Non-Price Competition Price Compeition Intensify ② Retailer ③ Reduce ④ ⑤ ③ Chen, Iyer & Lal & Savary (1999); Padmanabhan (2002); Lynch & Ariely Iyer & Pazgal (2003); (2000) Scott Morton et al. (2001); Zettelmeyer et al. (2002) Manufacturer ⑩ Brown & Goolsbee ④ (2002); (2004) ⑧ Kuksov ⑥ ⑨ and ⑥ ⑦ ⑧ Proposed Model Note: The numbers cited in the table correspond to the indicating numbers in Figure 2.1. 124 Table 4.5: Statistics on Consumer Demographics – Total Replacement Sample Total % Gender (20203) Non Automotive Automotive Internet Internet User User Sites Only Sites Only Both Sites % % % % % (7923) (12280) (2664) (1240) (7451) Independent Manufacturer Male 59.64 58.13 60.62 59.42 57.74 62.86 Female 40.36 41.87 39.38 40.58 42.26 37.14 (18232) (7106) (11126) (2367) (1110) (6820) Married 71.33 69.98 72.19 75.33 71.62 71.76 Single 14.42 11.74 16.13 11.41 16.76 18.06 Widowed 4.01 7.08 2.06 3.13 1.44 1.32 Divorced 10.23 11.20 9.62 10.14 10.18 8.86 (19826) (7797) (12029) (2610) (1224) (7299) White 90.28 91.34 89.59 88.93 91.26 89.70 Black 3.19 3.60 2.92 3.26 2.12 2.63 Asian 2.56 1.39 3.33 3.75 1.88 3.51 Hispanic 2.84 2.62 2.99 3.03 3.02 3.03 Other Race 1.12 1.05 1.17 1.03 1.72 1.14 (19938) (7824) (12114) (2623) (1221) (7368) [...]... influence the consumer consideration decision in terms of consideration set size and similarity among the components in the consideration set? An examination of the Internet’s impact on the consideration decision is in response to calls for research on the “shape” of consideration set, that is, whether similar or dissimilar components tend to appear together in a consideration set and under what conditions... focuses on how online product information affects the consumer consideration and choice decision in automobile purchases This chapter reviews the literature on the choice model and research on Internet use and its influence on the consumer choice decision and market structure 2.1 Consumer Choice Consumers make brand choices according to the rule of utility maximization: that is, they select the brand with... consumer consideration decision, as these two often act together and move the results in the same direction Without a direct measure of the consumer s expected search and evaluation costs, it is difficult to identify the influence of each in the consideration decision Therefore, most of the two-stage choice models construct a total consideration cost, rather than 13 independent information search and. .. information before they make their choice decisions However, neither the complete prior knowledge nor the unlimited processing capability assumption is realistic Two-stage choice models with consideration set formation as the pre -choice stage provide a solution to the above-mentioned limitations of one-stage choice models 2.2 Consideration Set A consideration set is the set of alternatives that are considered... influence of Internet use on a consumer s consideration formation The only exception is Wu and Rangaswamy (2003) They examine the way in which the use of two online search functions, sorting and forming personal lists, affect the perceived uncertainty in consideration utility However, the limited number of online search functions and grocery shopping samples restrict the generalizability of their estimation... not only a channel for price information, but that it also provides a lot of non-price information The focus here is on the Internet’s influence on consumer consideration and choice, which covers both price and non-price competition Insert Figure 2.1 here 21 Insert Table 2.1 here The only study in the literature that looks at the effect of an Internet information search on consumer consideration and choice. .. affect the consideration decision In our model, the consideration set is defined as the set of product subcategories or alternatives that a consumer decides to search and process prior to a choice Therefore, both the information search and the information processing costs will take effect at the consideration stage, with the information search cost taking effect at an earlier stage than the information... facilitates the investigation of the general impact of online information searches on consumer brand choice and industry competition Another contribution of this study is that it categorizes automotive Web sites in terms of whether or not a site provides information on competing brands Different types of Web sites can have different effects on consumer consideration and choice In the automobile industry, independent... Wu and Rangaswamy (2003) They examine how the use of two online search functions, sorting and forming a personal list, can affect the consideration of liquid detergent by changing the perceived uncertainty of the consideration utility They find that the use of these two functions induces different effects on the number of alternatives that are considered by the consumer Nevertheless, given that the. .. comparison of the search cost and the marginal returns from the search Metha, Rajiv and Srinivasan (2003) examine the relationship between consumer information search activities and the consideration decision Their model assumes that the consumer learns about product quality through the consumption experience but searches for price information at retail stores on each purchase occasion The set of alternatives . THE IMPACT OF ONLINE SEARCHES ON CONSIDERATION SET FORMATION AND CONSUMER CHOICE Zhao Hongyu (Master in Economics, Fudan University) A THESIS SUBMITTED FOR THE DEGREE OF PHD OF. consideration set size and similarity among the components in the consideration set? An examination of the Internet’s impact on the consideration decision is in response to calls for research on the. study consumer Internet use and search behavior on the “shape” of consideration set. Very little work has been done on the influence of Internet use on a consumer s consideration formation. The