Heterogeneity in nested logit models: An estimation approach and empirical results

Abstract

A methodology is provided to account for heterogeneity in intrinsic brand preferences and intrinsic purchase propensities across households in nested logit models of purchase incidence and brand choice. The semiparametric approach used in the estimation is based on the random effects specification to accounting for heterogeneity in conditional logit brand choice models. The procedure involved approximates the unobserved, underlying joint multivariate distribution of preferences and propensities by a discrete distribution, and estimates the support points and probability masses associated with this distribution. The empirical application to scanner panel data on the yogurt category reveals that not accounting for heterogeneity in intrinsic preferences and purchase propensities results in (1) poor model fit to the data and (2) biased estimates for the effects of marketing variables on the category purchase and brand choice decisions. The semiparametric approach also provides insights into segmentation issues for the brands competing in the marketplace.