Bayesian Covariance and Variable Selection for Explaining Consumer Behaviour

摘要

We estimate the random coefficient model by means of Markov Chain Monte Carlo methods (MCMC) and simultaneously carry out variable selection and covariance selection during our modeling procedure. Following the statistical principle of parsimony this method yields a model, which includes only the significant variables and covariance elements and therefore allows a more efficient estimation. It offers a reasonable basis for making decisions in real applications. We will demonstrate this for marketing data which come from conjoint analysis. In this application the heterogeneous behaviour of consumers has to be explained from high-dimensional data.