Hello,

I am performing analysis on an unbalanced panel dataset, of 160 firms which have data available for approximately 6 years. My dependent variable is the relative change in carbon intensity, which implies a gaussian distribution. However, I am very new to this type of analysis. I have already read a lot of forum questions on this topic because it seems people are struggling with the same things as I am. However, the information is really scattered and I am really struggling to find the right approach.

I have read that xtreg is very common in this type of analysis. However, it is not clear to my what characteristics make xtreg the better. I know the Hausman test can be a guide in determining wether I should use fixed effects or random effects. But then a new question arises, is there a difference between xtreg, re and gee?

Another point I am struggling with is if and how I should test for heteroscedasticity, multicollinearity and autocorrelation. I assume I can test for multicollinearity in the same way as I should in one wave (for example by looking at VIF). I have already read about xttest2 to test for heteroscedasticity and autocorrelation, in which I should use -robust- to take into account heteroscedasticity and autocorrelation. However, I am struggling a lot with autocorrelation. My intuition tells me regressing with xtreg and adding robust is not sufficient. Shouldn't I switch to another regression method to take this autocorrelation into account (for example xtgls, or xtregar, fe)

There are a lot of options with respect to panel data, and I cannot see the wood for the trees. I hope your insights in this matter will help.

Kind regards,
Timea