the use of xtgls

[Nermine Medhioub]

Hello everyone!
Can I use the xtgls command when there are no problems of heteroscedasticity and autocorrelation? (The Hausman test gives me random effects however, the results are not significant with xtreg,re)

[Daniel Schaefer]

Do not do this. Estimating models until you get significant results is p-hacking. It sounds like you are shopping around for a model that gives you a type 1 error. Instead, you should ask yourself why the results are not not significant.

Is the null result contrary to theory or previous findings? In that case, a null result may be interesting. If not, is it possible that your N is not large enough to support your model? If none of the above, you might seriously consider that your predictor variables don't actually predict your outcome.

[Nermine Medhioub]

Ok thank you !