Non Normal Panel Data

[Sarah Jacob]

Hi, I am relatively new to econometrics. I want to know does panel data have to be Normal? My data of 96 observations is not normal. How do I transform it in Stata?

[Maarten Buis]

No real data is truly normal; the normal distribution is a potentially useful model, but models by definition not true. So it depends on what you mean with not normal. Do the residuals (remember, it is the residuals that need to be normal, not the dependent variable) slightly deviate from normality, then you could probably use the "standard" methods. If the deviation is larger, then there are many options depending on the exact deviation, the purpose of the model, the intended audience, etc. etc. etc. Also what is a "slight deviation" and what is a "larger" deviation depends. So without context, this is about all that can be said.

[Carlo Lazzaro]

Sarah:
as an aside to Maarten's helpful reply, I would be more concerned about having 96 observations only in a panel dataset.

[Sarah Jacob]

Thanks Maarten and Carlo. Carlo Lazzaro It is a 4 year study of 24 courts, thus 96 observations. More data is unavailable. Maarten Buis How do I know the normality of residuals. As I only know Shapiro Wilk test, which I apply to all variables.

[Carlo Lazzaro]

Sarah:
you can also visually inspect the residual distribution via -help qnorm-.
If residuals distribution remarkably deviates from normality, you can consider imposing -robust- or -vce(cluster panelid)- standard errors.

[Sarah Jacob]

Thanks Carlo