A Corrected and More Efficient Suite of MCMC Samplers for the Multinomal Probit Model

The multinomial probit (MNP) model is a useful tool for describing discrete-choice data and there are a variety of methods for fitting the model. Among them, the algorithms provided by Imai and van Dyk (2005a), based on Marginal Data Augmentation, are widely used, because they are efficient in terms of convergence and allow the possibly improper prior distribution to be specified directly on identifiable parameters. Burgette and Nordheim (2012) modify a model and algorithm of Imai and van Dyk (2005a) to avoid an arbitrary choice that is often made to establish identifiability. There is an error in the algorithms of Imai and van Dyk (2005a), however, which affects both their algorithms and that of Burgette and Nordheim (2012). This error can alter the stationary distribution and the resulting fitted parameters as well as the efficiency of these algorithms. We propose a correction and use both a simulation study and a real-data analysis to illustrate the difference between the original and corrected algorithms, both in terms of their estimated posterior distributions and their convergence properties. In some cases, the effect on the stationary distribution can be substantial.
View on arXiv