Estimating Disaggregate Model Using Aggregate Data via Augmentation of Individual Choice

The past two decades have witnessed a variety of applications of consumer demand model in the marketing literature employing aggregate sales scanner data and household panel data. Aggregate data are readily available and easy to collect, but are difficult to use to make inferences about household-level behavior unless the estimated models are linear. In this paper we describe a method of augmenting aggregate data with individual choices that facilitates the development of non-linear models of behavior. Our Bayesian procedure introduces latent choice data that are consistent with observed sales. Once the augmented choice data are available, analysis can proceed using a variety of models of consumer choice. We illustrate our method using both simulated and real purchase data. Results from the two experiments show strong support for the accuracy and validity of the proposed method.