Simulation of Multinomial Probit Probabilities and Imputation of Missing Data

We use simulation methods to impute missing data. First we suggest how one can iteratively estimate a large number of parameters associated with a joint normal distribution function fof latent variable associated with the data. We suggest a way to test the joint normality assumption next. Finally, we propose a method to use draws from the estimated distribution efficiently in a method of simulated moments or simulated maximum likelihood procedure. In the second half of the paper, we apply the proposed methods ot two data sets from Jamaica with significant missing data problems. We find that the procedure provides better parameter estimates in simple models than present popular methods