Modelling preference heterogeneity using a Bayesian finite mixture of Almost Ideal Demand Systems

Demand studies often use observable characteristics to proxy preference heterogeneity. It is likely, however, that some households with the same observable characteristics have quite different preferences. An alternative approach is to use a Gaussian mixture of Almost Ideal Demand Systems to capture the heterogeneity. We show how to estimate this with censored purchase data for 5 food categories using Bayesian inference. Using model outputs we infer four different preference classes; how distinct these classes are from one another and which food categories are driving the segmentation process.