An error components approach to segmentation and modelling brand choice dynamics

Abstract

We develop and illustrate a new methodology for estimation of marketing mix coefficients, in brand choice models of the attraction type. The method yields consistent and efficient estimates of choice coefficients and elasticities, for different consumer segments. Analysis is at the household level and we derive a generalized least squares (GLS) estimator of coefficients of a linear form of the multinomial logit model of choice. We determine consumer segments on the basis of similarities in households' random choice processes, and then estimate the impact of marketing instruments on each homogeneous group. Our segmentation and estimation method accounts for heterogeneity of preferences and the dynamics of choice behaviour. The end result of our analysis is a comprehensive profile of each segment of a market, in terms of psychological characteristics which can be used by marketers to develop target marketing strategy. The model is also found to predict choice well, compared to standard models. The methodology offers marketing managers who have access to scanner panel data, a way to segment their markets and obtain insights into the choice behaviour of their consumers.