xtpoisson and sktest

[Ward Bruurs]

Dear Stata-fan

I am currently doing an xtpoisson regression and asked myself some questions with the sktest. More specifically, to what extent skewness and kurtosis should be taken into account.
Because I work with panel data and Poisson regression, it is clear that skewness is common, due to the counting variable. Some independent variables that also cause 'problems' with the sktest include: % of population that attends university, number of chemical companies per 100,000 inhabitants and number of universities per 100,000 inhabitants.
In the literature I think that skewness and kurtosis are discussed quite simply or not at all. I also performed robustness tests and the use of vce(robust) is certainly necessary.
Therefore, the question is whether it is absolutely necessary to make adjustments to my variable to resolve this skew, even though this does not benefit my model.

Thank you in advance
Ward Bruurs

[Joao Santos Silva]

Dear Ward Bruurs,

Skewness and kurtosis do not play any significant role in this context, so please just ignore the results of sktest.

Best wishes,

Joao

[Ward Bruurs]

Dear Joao Santos Silva


Thank you for the reply. Now I am also testing for autocorrelation (xtserial), which shows that first order autocorrelation is present.
Now I read on this forum that adding FE and vce(robust) helps to solve both heteroscedasticity (which is also present) and autocorrelation however this was the case with a xtreg regression.
Now I was wondering if this is indeed the case or if I should use other methods to solve this autocorrelation.

Yours sincerely
Ward Bruurs

[Joao Santos Silva]

Dear Ward Bruurs

Just explicitly cluster your standard errors.

Best wishes,

Joao