Heterogeneity and purchase event feedback in choice models: An empirical analysis with implications for model building

Abstract

This paper evaluates empirically the most commonly used methods of incorporating heterogeneity and purchase event feedback in choice models. We evaluate each method with respect to model fit, forecast accuracy, differences in estimated marketing mix response, and stability of estimated market segments. Our general findings are that (a) feedback and preference heterogeneity contribute substantially to fit and forecast accuracy; response heterogeneity improves upon this but not substantially; (b) estimated response parameters vary significantly across methods, but the corresponding elasticities do not; and (c) large segments are fairly stable in their preference and elasticity structures but small segments are much less so. Our comparison of specific methods suggests that, if researchers wish to use feedback and heterogeneity as controls for forecasting purposes or for estimating marketing mix response, BLOY, a smoothed measure of previous purchases, may be used in conjunction with a simple preference heterogeneity measure. If researchers wish to study heterogeneity and feedback per se, they should use a relatively complex heterogeneity method such as a finite mixture model, but they must make a trade-off between BLOY and a simple last-purchase indicator variable (LAST). BLOY provides a better fit but yields less stable market segments. LAST yields poorer fit but the market segments are more stable.