(Pseudo) R Squared for paneldata tobit/ordered probit

[David Johansen]

Dear Stalist,

I try to obtain Pseudo R_Squared for both an Xttobit and an XToprobit model. However, I can't figure out how to do this. Is there any specific command to obtain for instance McKelvey & Zavaiona's R2? For XTlogit I use -r2_mz-. Unfortunately, this command does not help for Tobit and ordered probit models.

Kind regards,

David

[Nick Cox]

That should be Zavoina. Also, r2_mz is from SSC as you are asked to explain. I am not a big fan of any of these measures, but you shouldn't care about that. But it's contentious territory. Loosely speaking, the main deal seems to be assessing models by metrics that are at best loosely related to how they were fitted. All that said. If you have a definition you like for each command, it should be a matter of getting predict to calculate some results and then writing your own script to calculate the measure you need.

[David Johansen]

Dear Nick,

Thank you for your prompt reply. Is there a reason why you are not a fan of Pseudo Rsquared measures? Do you have a preference for other goodness of fit measures?

[Nick Cox]

Well, all single-number measures can conceal more than they reveal. Beyond that, I prefer measures with a simple interpretation. The whole tribe of pseudo measures give away the problem in their name: if you like R-square, this is similar except insofar as it is different.

[David Johansen]

Many thanks again. I've decided for consistency to use McKelvey & Zavoina's R2 in all my models. However, I struggle to obtain Variance of y* and Variance of error. Do you have any suggestion on how to obtain them with Panel Data?

[Rich Goldstein]

I'm not sure exactly what you are asking for here but (1) for xttobit, the save matrix "r(table)" includes what I think you are asking for; (2) for xtoprobit, I don't see that there are two such variances so please clarify (the one variance usually reported is saved and can be examined and captured starting with -ereturn list-)

[David Johansen]

Hi Rich, I try to manually compute the McKelvey & Zavoina's R2. As the formula is R^2=Var(y*) - Var(error) /(Var(y*) i need to obtain Variance of y* and Variance of error. In cross-sectional datasets, they can be obtained by using the -fitstat- command. However, I use panel data and therefore struggle to derive them.