We propose an EM alrithm to estimate ordered probit models with endogenous
regressors. The proposed alrithm has a number of computational advantages in comparison
to direct numerical maximization of the (limited information) log-likelihood function. Fir