Bayesian estimation of individual-behavior models using aggregate data.

摘要

There are many instances in marketing in which researchers do not observe the individual behavior of consumers (e.g., individual choices, coupon availability and coupon usage). However, the researcher may have aggregate information such as market shares and number of redeemed coupons (in total). In this context, the estimation challenge is to generate inferences about the distribution of the unobserved individual behavior of consumers from aggregate or limited data. Accordingly, the development of new Bayesian methods to address this problem constitutes the main research objective of this dissertation.; The estimation strategy presented in this dissertation involves three general steps: (i) formulate probabilistic assumptions about the unobserved individual behavior of consumers (e.g., choices follow a logit model, each consumer has the same probability of having received a coupon), (ii) use the aggregate data to specify constraints on the unobserved individual data (e.g., the total number of consumers redeeming a coupon in a given period has to be exactly consistent with the observed data) and (iii) simulate the unobserved individual data taking into account both the probabilistic assumptions and the constraints specified using the aggregate data. This general methodology is applied in each of the essays of this dissertation.