The authors investigate whether household scanner data provide representative inferences about brand choice behavior. They show the (analytical) equivalence of the homogeneous brand choice logit and the multinomial store sales models. Then, they determine how results for the homogeneous logit model estimated with actual household data deviate from results for the multinomial model estimated with actual store data from the same community. They also perform this comparison using model specifications that provide average estimated parameters while accommodating unobserved household heterogeneity. The authors find statistical support for the hypothesis that panelist households are not representative, whether household heterogeneity is or is not accommodated. Substantively, however, the average estimated price elasticities are close for the household and store data analyzed, if the household data are selected on the basis of purchase selection. An alternative selection procedure, called household selection, which is used for the analysis of complete household purchase records, provides results that strongly differ from the purchase selection results.
Analyzing the demand system for beer in the US: an application of the random coefficient logit model using scanner data
