CONFORMING TO PREEERENCES MORE THAN ACTIONS 201 wo puc and either 5,892( tafter viewing its inform tion)"They all read and information nts either had to make 10 p based or information on how much other people liked each product/how was to g ate a rating number (between I and 10)so that they tio abou to buy. they entered a raffle to receive five of the products they had chosen the You moved to the choice phase.The suvey pre tedpaticpantswith pnts were rar one of thei one of th would buy. r vs. ver than targ W zed the In t Results and Discussion had nducted a multivariate (a)the e We th standardized)and (b)the differ he lepen ine tandardized)both the diffe rating (i.c 6 ables Wald(1)=12.61.p<001,had a positive ()(c)participant (dumm on choice.In upport the the impa oth the .26SE 19. betw the coefficients:85.SE=34).Wald(1) 001)and the diffe 001 d that the diffe eb=1.6 mpact he npact e in the number of lik mber of views did not (=18.SE=22).WalddD)=65.p we had 101 nts (we used the ene ices we dumn ice (1 choo Follow-Up Study:Experimenter-Generated Choice Set on (a)the dif ence in num To sen be he matching-choosi digm.in a follo up study (n difference in number of views between the target and the compet As an ted four choice sets that used the besCadRoom afte 70 cd p and were bought by either(higher)or64%(lower)of thoseParticipants then moved to the Amazon context and read about two puddings that were offered online, differing in terms of “rating (range: 1 to 10)” and “sales (i.e., percentage of consumers who bought this product after viewing its information).” They all read that Pudding A (the “target”) was rated 8/10, and 60% of those who viewed it bought it. In the “missing preference information” condition, the sales information for pudding B (the “competitor”) was 70% and the rating information was missing. Participants’ task was to generate a rating number (between 1 and 10) so that they were “indifferent about which pudding to buy.” In the “missing action information” condition, pudding B (“competitor”) was rated 7/10, but the sales information was missing (see Appendix B). After creating choice sets using this matching procedure for both the YouTube clips and the Amazon puddings, participants moved to the choice phase. The survey presented participants with each of the choice sets he or she generated, and asked them to choose which video they would watch and which pudding they would buy. Results and Discussion Beginning with the pudding choice, we conducted a multivariate logistic regression with choice (1 choose the target option; 0 choose the competitor) as the dependent variable on (a) the dif￾ference in ratings between the target and the competitor (i.e., preference information; standardized) and (b) the difference in sales between the target and the competitor (i.e., consumption information; standardized). Both the difference in rating (i.e., preference information, b 1.52, SE .26), Wald(1) 34.17, p  .001, and the difference in consumption (b .67, SE .19), Wald(1) 12.61, p  .001, had a positive impact on choice. In support of our hypothesis, the impact of rating information was larger than the impact of consumption information (for the differ￾ence between the coefficients: b .85, SE .34), Wald(1) 6.39, p .011. A similar analysis of the video choice revealed that the differ￾ence in the number of likes positively predicted choice (b 1.63, SE .29), Wald(1) 31.02, p  .001, and the difference in the number of views did not (b .18, SE .22), Wald(1) .65, p .42. In support of the hypothesis, the impact of number of likes was stronger than the impact of number of views (for the difference between the coefficients: b 1.45, SE .46), Wald(1) 9.84, p .002.1 Follow-Up Study: Experimenter-Generated Choice Set To generalize our effect beyond the matching-choosing para￾digm, in a follow-up study (n 266 Mturk workers, 86 females, Mage 32), we test our predictions in a paradigm that offers higher ecological validity. Specifically, we had participants choose between 10 product pairs (in an Amazon condition) or 10 clip pairs (in a YouTube condition). For each trial (product or clip pair), we created four choice sets that used the same target option and manipulated information on the competitor (see Appendix C, Ta￾bles C1 and C2). For example, in the chocolate bars trial (Ama￾zon), the target options were rated 3.8/5 by those who decided to rate it, and 72% of the people bought it after viewing it. The competitors were rated as either 4.2/5 (higher) or 3.4/5 (lower), and were bought by either 80% (higher) or 64% (lower) of those who viewed these products. In the pair of pet videos (YouTube), 4,827 people liked the target option and 194,335 people viewed it, and either 5,892 (higher) or 3,762 (lower) people liked the com￾petitor, and either 232,193 (higher) or 156,477 (lower) people viewed it. Participants either had to make 10 purchase decisions based on information on how much other people liked each product/how many people bought each product (Amazon), or had to make viewing decisions based on information on how many people liked each video/how many people viewed each video (YouTube). Par￾ticipants’ choices were consequential. In the Amazon condition, they entered a raffle to receive five of the products they had chosen in the study. In the YouTube condition, after making their selec￾tions, participants were randomly assigned to watch one of their selections. For each trial, we randomly assigned participants to one of the four conditions: 2 (preference: competitor is higher vs. lower than target)  2 (action: competitor is higher vs. lower than target). We analyzed the data at the individual choice level. In the Amazon condition, we had 165 participants. Each participant made 10 different product choices; however, one choice from one partici￾pant was missing, which resulted in 1,649 binary choices. Because choices made by the same participant for the same product were dependent, we dummy-coded participants and products. We then conducted a multivariate logistic regression with choice (1 choose the target option; 0 choose the competitor) as the dependent variable on (a) the difference in customer ratings be￾tween the target and the competitor (i.e., preference information), (b) the difference in sales between the target and the competitor (i.e., consumption information), (c) participant variables (dummy coded), and (d) product variables (dummy coded). We find that both the difference in rating (i.e., preference information; b 3.26, SE .19, Wald(1) 300.96, p  .001) and the difference in consumption (b 2.32, SE .17, Wald(1) 190.71, p  .001) had a positive impact on choice, and in support of our hypothesis, the impact of rating information was larger than the impact of consumption information, t(1473) 14.79, p  .001. In the YouTube condition, we had 101 participants (we used a smaller sample for this condition, determined by the effect size gathered after we conducted the Amazon study). Each participant made 10 choices, resulting in 1,010 binary choices. We dummy￾coded participants and videos (to control for dependency in these variables), and conducted a multivariate logistic regression with choice (1 choose the target option; 0 choose the competitor) as the dependent variable on (a) the difference in number of likes between the target and the competitor (i.e., preference), (b) the difference in number of views between the target and the compet- 1 As an exploratory analysis, we also looked at the participants’ choice as a function of which type of information they were asked to generate (preference or action). In the Amazon condition, we find that participants in the missing-rating condition chose their self-generated option (72%; 78/108) more often than those in the missing-sales condition (16%; 18/ 112), 2 (1) 70.48, p  .001. In other words, those who generated preference information (ratings) were more likely to choose their self￾generated option. Similarly, in the YouTube condition, we find that par￾ticipants in the missing-likes condition chose their self-generated option (70%; 78/112) more often than those in the missing-views condition (16%; 17/108), 2 (1) 65.11, p  .001. Again, those who generated preference information (likes) preferred their self-generated option more. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. CONFORMING TO PREFERENCES MORE THAN ACTIONS 201