MODELING HOUSEHOLD PURCHASE BEHAVIOR WITH LOGISTIC NORMAL REGRESSION

The successful development of marketing strategies requires the accura te measurement of household preferences and their reaction to variable s such as price and advertising. Manufacturers, for example, often off er products at a reduced price for a limited period. One reason for th is practice is that it induces households to try the promoted product with the hope of retaining them as permanent customers. The successful implementation of this strategy requires knowledge of the extent of p rice sensitivity in the population, effective methods of advertising, and the existence of a carry-over effect in the household's evaluation of the product. Logistic regression models are often used to relate h ousehold demographics, prices, and advertising variables to household purchase decisions. In this article we extend the standard model to in clude cross-sectional and serial correlation in household preferences and provide algorithms for estimating the model with random effects. T he model is applied to scanner panel data for ketchup purchases, and s ubstantive insights into household preference, brand switching, and au tocorrelated purchase behavior are obtained.