How to test between Fixed effects and Random effects after xtivreg, vce(cluster)

[William Kratochwill]

Hello,
I used stata 17.0 and xtivreg to instrument a variable and vce(cluster var). I want to use the Hausman test for FE and RE. However, it is impossible while vce(robust) and -xtoverid- after -xtivreg- do not provide it, as I need. Is there any command to help?
I appreciate any help you can provide.

[Maxence Morlet]

You could conduct a Mundlak (1978) test, but with instruments I am unsure of the operationalisation of such a test.

Just to know, are all of your explanatory variables of interest time-varying?

If yes, just run a fixed effects regression with at least unit and time fixed-effects, simply as a reflex. Random effects makes completely implausible assumptions.

Also, how many clusters (or units) do you have?

[William Kratochwill]

Dear Maxence,
Thank you for your response. Yes, all my variables are time-varying. I'll run exclusive regressions for different time-invariant variables. I must be missing something about Random-effects.

xtivreg lwageh age agesq i.race i.eatt i.posocup  b12.VD4010 `rind' (lu3 = l1lu3) if age < 71, fe vce(cluster regions) baselevels

Fixed-effects (within) IV regression            Number of obs     =  2,254,672
Group variable: IDP                             Number of groups  =  1,314,421

R-squared:                                      Obs per group:
     Within  = 0.0132                                         min =          1
     Between = 0.2684                                         avg =        1.7
     Overall = 0.2484                                         max =          2


                                                Wald chi2(36)     =   58637.50
corr(u_i, Xb) = 0.3159                          Prob > chi2       =     0.0000

                                                                    (Std. err. adjusted for 77 clusters in regions)

[Jeff Wooldridge]

Here's a paper where we show how to do this.

https://www.jstor.org/stable/pdf/10....in=&acceptTC=1

You include the time averages of all exogenous variables and test their joint significance using xtivreg with the re option (and vce cluster regions), of course.

[William Kratochwill]

Dear Professor Wooldridge,
I really appreciate your recommendation, and I'll carefully read your paper. I am sure there is so much in it to expand my knowledge that I cannot miss a word.