A potential solution to problems in ordered choice models involving endogenous ordinal variables for self-reported questions

Most of the surveys in social sciences generally consist of ordinal variables. Sometimes researchers need to model behaviour of ordinal variables in simultaneous equation system involving many endogenous ordinal variables. This situation leads to a very complex likelihood function which is extremely hard to solve. The solutions suggested in the literature are even harder to understand by applied researchers. The present study suggests a simulation method to avoid this problem altogether by converting ordinal variables into continuous variables and use standard simultaneous regression models. The proposed method involves generating random numbers from continuous probability distributions (uniform and truncated normal distributions) within a discrete probability distribution. This method can be fruitfully be used in ordered logit and probit models. The limitations of this method are also discussed.