bivariate poisson count models

[Steve Sifo]

I have managed to run a simple bivariate model using bivcnto command in the code below


asdoc bivcnto edu economicstatus i.gender age householdsize i.electricity1, ///
copula(normal) irr nolog ///
title(Model 1: Copula Normal Results) append ///
dec(2) tzok
eststo model1


but now am failing to extract residuals to run model diagnostics, anyone with ideas please. Am really stuck here or suggest what diagnostics i can run for this model

[Daniel Schaefer]

bivcnto is a user submitted command described in the Stata Journal so you shouldn't expect it to support all of the features you might expect from a built-in command. As far as I know, regression diagnostics for logit and other generalized regressions are still an active area of research, and probably doubly so for a specialized model like this. If you really need residuals, you might be able to calculate the Pearson residuals yourself. This stats stack exchange thread gives some guidance, but in general you want to create predicted values based on the model, then find the difference between the predicted and actual values, then calculate the Pearson residuals from there. Usually you would do that with the -predict- command, but I doubt -predict- and -bivcnto- know how to talk to one another, so you may end up having to write down the full regression equation yourself to calculate predictions instead.

Suffice to say, I don't think the kinds of regression diagnostics you want are common practice with non-OLS generalizations of regression. You might be able to carry the diagnostics out anyway, but you won't be able to rely as heavily on built-in features to do it.