Testing assumptions xtprobit,re

[Eva Straka]

Dear Statalisters,

I am running a random effects probit model and was wondering about the assumptions that need to hold (homoskedasticity, no multicollinearity, normal distribution of the residuals..) for this model to be consistent.

However, I am unable to find the proper tests that fit the model to do so; hence i have the following questions:

Do you know how to

1) calculate the residuals which have to be normally distributed for the xtprobit
2) test for homoskedasticity
3) test for multicollinearity ?

your help is very much appreciated!

[Phil Bromiley]

You need to read more about the theory. I don't think residuals will be normally distributed in a 0/1 dv model.

For multicolinearity, you may find that you can do equivalent tests using regress (colinearity is strictly a function of rhs variables so it should be similar). Homoskedasticity - do you get radically different results with clustered standard errors?

[Eva Straka]

Thank you for your answer!
The results that I have with and without the clustering of standard errors are not very different so I assume heteroskedasticity not to be an issue?

But do you know how to get to the residuals of such a model ? I am trying the following code:

    xtprobit later_larger i.treatment i.question_nr, re vce(cluster subject_ID)
    predict myresiduals, residuals

but this does not seem to work.

Thank you!