Multivariate Visual Diffusion for Social Groups

Abstract

Car purchases are among the largest and most expensive purchases consumers ever make. While functional and economic concerns are obviously important, we examine whether social influence also plays a role. Using a Bayesian probability model along with data on over 1.6 million cars sold over a nine-year period, we examine how social influence affects car purchase volume. In particular, we investigate three questions: (1) Are people more likely to buy a car if others around them have recently done so? (2) Do these effects vary based on the identity (e.g., age and gender) of the prior purchasers? (3) Are they moderated by visibility, the ease of seeing others' behavior? After subjecting our estimates to an extensive set of tests (a quasi-experimental design using distance, matching, and dosage; model- and covariate-based controls; permutation re-sampling; and out-of-sample tests) to assess whether they might be due to something other than social influence, we find that social effects are (1) present (one additional purchase is made for every seven to ten prior purchases made locally by a given subgroup), (2) stronger for prior purchases by men than by women, and (3) larger in areas where others behavior should be more visible (i.e., more people commute in car visible ways).