Assumptions for a mediated model with multiple fixed effects regressions

[Siham Hari]

Deal all,

I am trying to study a mediating model with multiple fixed effects regressions and before running the xtreg, fe regressions
I am not sure which assumptions shall I test in order for my model to be valid.

Some commands on Stata don't work with fixed effects regressions (for example the vif command testing for multicollinearity), therefore I am wondering if I have to test these assumptions on a simple regression and if the result can be applied also if I have panel data.

Multicollinearity, normality of residuals and heteroscedasticity: are these assumptions enough or shall I run others as well?

Any help is much appreciated!

Regards,
Siham

[Carlo Lazzaro]

Siham:
if you plan to run a fixed effect regression, you can test those postestimation commands after -regress- with -i.panelid- among predictors.
That said, if you go -xtreg,fe- there's actually no Stata built-in postestimation suite of commands and you have to rely on graphical inspection of residual distribution to check normality and heteroskedasticity. Eventually, as wisely reminded by Clyde Schechter on this forum, quasi-extreme multicollinearity can be suspected by weird (usually too wide) confidence intervals.