Fixed Effects and Clustered Standard Errors for Non-Linear Models

[Kevin Petersen]

Hi all,

I am running a logistic regression in which I have a nested data structure (two levels) and wish to control for both average differences between grouping units and serial correlation within grouping units. In a post from 2014 (https://www.statalist.org/forums/for...the-regression), Professor Schechter indicated that:

Notice that clustered standard errors and country-level effects are not mutually-exclusive. In fact, they tend to go together.
I emphasize that this advice applies only to linear regressions. If you plan to estimate non-linear models, things are different.

I'm wondering if anyone can elaborate on this second point, that things would be different in a non-linear model. I understand that heteroscedasticity is different in non-linear models, but why wouldn't fixed effects and clustered SEs for serial autocorrelation otherwise be appropriate in these models?

Thanks for any help you can provide!

[Andrew Musau]

If you have clustered data, you have clustered data. End of story. Otherwise, refer to this discussion in the following Stata FAQ: https://www.stata.com/support/faqs/s...nce-estimator/.