Fixed effects std. error and p-values
smauricioj opened this issue · comments
It maybe my fault, but I'm not finding these statistics on the fixed effects models.
From the PanelOLS().fit() results, can see the estimated_effects dataframe.
But what about these values errors and p-values? Is this something still to be included, or am I totally missing it?
There is no inference on estimated effects. There is no clear sensible way to get reasonable asymptotic in the usual case of large N, fixed T for fixed effects. If you have a FE that you want inference on, and it has a small dimension relative to sample size, you can include it as a regressor and get the usual standard errors (e.g., in a large N, small T panel, you can include the time dummies as regressors).
I see. Actually, my dataset is small in all dimensions (N=4, T=16), but eventually it will grow to the usual case.
I'm finding autocorrelation on the residuals (dw test = 0.91521) even when using time effects, something I guesses was impossible. I imagined that, if the FE inference showed lack of validity, maybe I could say it's because of this.
Any idea why I'm finding this? You clearly know the subject more than I 😆