Test for time effects in a random effects (RE) model?

[amira elshal]

Dear Sirs,

To see if time fixed effects are needed when running a FE model, I use the command testparm. Do I have to test for time fixed effects when running a RE model?

And when I run the Hausman test to test for RE versus FE, do I include time dummies in my regression (given that time fixed effects are needed)?

Thank you.

Kind regards,
Amira

[Carlo Lazzaro]

Amira:
I assume that you refer to -i.year-. You can simply see if their inclusion makes sense giving it a try, as in the following toy-example:

use "http://www.stata-press.com/data/r14/nlswork.dta", clear
xtreg ln_wage race grade i.year, fe
xtreg ln_wage race grade i.year, re

As far as -hausman- is concerned, beware of not reversing the order of estimates storing (that is, -fe- specification should come before -re- specification):

use "http://www.stata-press.com/data/r14/nlswork.dta", clear
quietly xtreg ln_wage race grade i.year, fe
estimates store fe
quietly xtreg ln_wage race grade i.year, re
estimates store re
hausman fe re

Evenyually, robust SEs are not supported by -hausman-.

[Sebastian Kripfganz]

Please have a look at the adivce by Jeff Wooldridge in the following Statalist topic:
Hausman test with year dummies, V_b-V-B is not positive definite

[amira elshal]

Hi Carlo,

Thank you for your reply.

I am asking this question because I have noticed that researchers test for time fixed effects while running FE models only. So, I thought that maybe it is irrelevant to do this test while running a RE model.

One more question: if I found that I should include time dummies in my FE model only, shall I also include time dummies in the FE model when doing the Hausman test (as in Option 1 below)? Or shall I omit time dummies from both models when doing the Hausman test (Option 2)?

*Option 1 xtreg ln_wage race grade i.year, fe estimates store fe xtreg ln_wage race grade, re estimates store re hausman fe re
*Option 2 xtreg ln_wage race grade, fe estimates store fe xtreg ln_wage race grade, re estimates store re hausman fe re
Kind regards,
Amira

[amira elshal]

Hi Sebastian,

Thank you very much for referring me to Wooldridge's post. I read it and concluded that I shall test for time effects in both FE and RE models. However, I cannot include time dummies in the Hausman test. I can instead use Mundlak's approach to test for RE versus FE.

I will follow Wooldridge's instructions and come back to you if I need further help.

Kind regards,
Amira

[amira elshal]

Hi Sebastian,

If I find that time fixed effects are needed when running the RE model, shall I include time dummies while running the Mundlak test for RE versus FE?

Suppose I have two time-varying covariates (x2 and x3), one time-invariant covariate (x1) and a set of year dummies that I have to include, are the commands below right (for the Mundlak test)?

bysort id: egen mean_x2 = mean(x2)
bysort id: egen mean_x3 = mean(x3)
xtreg y x1 x2 x3 mean_x2 mean_x3 i.t, vce(cluster id)
estimates store mundlak
test mean_x2 mean_x3
Thank you.

Kind regards,
Amira

[Sebastian Kripfganz]

Yes, you should include the time dummies. Your sequence of commands looks good to me.