Hello Statalisters,
I'm using Stata 15.1. My question relates to the "xtgee" commands (in particular, I'm now interested to the binary outcome case, by my question is more general) for panel-data models.
I know one can use the user-written "qic" command as a selection criterion of the "within" correlation structure, and that, with the post-estimation "estat wcorr" command, one can collect useful information to reason about how much a given within-group correlation pattern is plausible. Nevertheless, I wonder whether it is possible to perform tests of statistical significance on such correlation coefficients. For example, the "independence" structure forces the non-diagonal elements of the correlation matrix to 0. I'd like first to evaluate the plausibility of such assumption, by performing a test of such model against the "unstructured" one, that would translate in a test on whether n*(n-1) correlation coefficients are jointly null. Is there a way to do it?
Federico
0 Response to Correlation structure of xtgee: tests of statistical significance
Post a Comment