Modelling contingent valuation iterated elicitation data with an MCMC approach

Abstract

The valuation of non-market goods involves iterated elicitation questions which obtain more information from the sample respondents and lead to more efficient welfare estimates. In this paper we consider the improvements which could be obtained by utilising a Bayesian MCMC approach to model this type of data. A fully informative prior resulting from previous stages is compared with a flat non-informative prior utilising both simulated and empirical data. These priors are combined with data in each stage to form the posteriors which are simulated with Gibbs sampling algorithms. The models are applied to an elicitation tree involving two successive dichotomous choice questions followed by an open-ended question. Monte Carlo simulations show that taking into account the information process implicit in successive elicitation improves the performance of the results at each stage and increases efficiency. Thus, the model allows the researcher to consider the evolving process along the elicitation tree, while increasing useful information obtained from the individual.