Verbek (2000) argues that it is necessary to test normality in the context of probit estimation to ensure consistence of Betas. Wooldrige (2002) affirms that it is true that in presence of heteroskedasticity and non normality the probit estimates are inconsistent but that focusing on the inconsistency of the Beta parameters "largely misses the point". Indeed the inconsistence of Betas is pratically irrelevant and the estimates of the partial effects are very good. The vast majority of econometric excercises in literature does not report test for normality after probit. Is this because they are not implemented in statistical software (as some authors sustain) or because of Wooldridge's argument?

Similar questions and discussions