Baysian Flexible Mixture Distribution Modelling of Dichotomous Choice Contingent Valuation with Heterogeneity

This paper considers the performance of a model of mixture normal distributions for dichotomous choice contingent valuation data, which allows the researcher to consider unobserved heterogeneity across the sample. The model is flexible and approaches a semi-parametric model, since any empirical distribution can be represented by augmenting the number of mixture distributions. Bayesian inference allows for simple estimation of the model and is particularly appropriate for conducting inference with finite data sets. The proposed model is compared with other semi-parametric and parametric approaches using Monte Carlo simulation, under alternative assumptions regarding heteroscedasticity and heterogeneity in sample observations. It is found that the mixture normal model reduces bias and improves performance with respect to an alternative semi-parametric model, particularly when the sample is characterized by heterogeneous preferences.