[Louisa Krekel]

Thanks a lot for the comments and help.
I will try to use the Mundlak method, thanks for the advice Williams!

[ALBA MARIA PRIEGO DE LA CRUZ]

Dear all,

I just read your posts about endogeneity.
In my case, I am analyzing a matched panel data through conditional logistic models (clogit distress_ prof fe re, group(match_) and I think that there may be a two-way relationship between the dependent and one independent variable.
Could anyone tell me how we could do an endogeneity test in our model?
Thank you in advance.

Alba

[Mads Moring]

Hi guys

Is it correctly understood, that xtoverid can be used after running a RE regression, such as

 xtreg y x1 x1...xn, re

?
With the hausman command I would need to estimate store after having run both and FE and RE model. But it seems that I can test whether I should use RE or FE without any regards to the FE model when using xtoverid as a kind of post estimation command?

[Carlo Lazzaro]

Mads:
you're correct,
-xtoverid- re-estimates a fixed effect regression itself: hence, providing robustified/clustered -xtreg, re- equation is enough for making the comparison running.

[Mads Moring]

Mads:
you're correct,
-xtoverid- re-estimates a fixed effect regression itself: hence, providing robustified/clustered -xtreg, re- equation is enough for making the comparison running.

Hey Carlo
Thanks for your answer! I have an additional question. (The following is mainly based on Wooldridge 2016)
This debate/thread concerns itself with the choice between a Fixed Effect model and a Random Effect model. I thought that the choice was Fixed effects or First difference against Random Effects or Pooled OLS. So if i based on the xtoverid command decide against Random Effects, how should I choose between Fixed Effects and First difference?

[Carlo Lazzaro]

Mads:
why bothering yourself with first difference when -xtreg,fe- is available?

[Mads Moring]

Because I was under the impression that Fixed Effects in only more efficient than First Difference when U_it is serially uncorrelated. I am not sure how to test whether that's the case?

[Carlo Lazzaro]

Mads:
what if you run -xtreg, fe- with roubustify/cluster your standard errors on -panelid- and compare your coefficients with those obtained with first difference?

[Peter Irabor]

Following the trend on choice of FE or RE, Can one conclude that the use of generalized Hausman test is less appropriate in comparison with xtoverid

[Carlo Lazzaro]

Peter:
as per -xtoverid- helpfile, it does not seem necessarily so.
However, there are situations in which -xtoverid- clearly outpeforms -hausman- (e.g., comparing -fe- vs -re- specification with robustified/clustered stndard errors).