SIMULATED MAXIMUM LIKELIHOOD FOR DOUBLE-BOUNDED REFERENDUM MODELS

Although joint estimation of referendum-type contingent value (CV) survey responses using maximum-likelihood models is preferred to single-equation estimation, it has been largely disregarded because estimation involves evaluating multivariate normal probabilities. New developments in the construction of probability simulators have addressed this problem, and simulated maximum likelihood (SML) for multiple-good models is now possible. This analysis applies SML for a three-good model under a double-bounded questioning format. Results indicate joint estimation substantially improves the variances of the parameters and willingness-to-pay estimates.