Hi,

I encountered the following problem when running IVPROBIT: the regressors include a few dummy variables (whether the respondent belongs to a particular country or not). The IVPROBIT (as well as PROBIT) gave very large variances to the coefficient estimates of these dummy regressors. Multicollinearity does not seem to be the reason: when I run the same model as 2SLS (assuming the dependent zero-one variable as regular quantitative variable), the variances of these regressors are fine. From this I conclude that the problem is in the way PROBIT calculates the variances. Looking forward to your advice on this matter.

Thanks in advance