Random effect model with moderating variables

[Maitena Iriondo]

Dear all,
I have run the Hausman test and I got the result that implies a Random Effect model is appropriate for my model. I am analyzing between firm effects in a 3 year panel dataset. To run the RE model though, I am having some doubts due to the nature of my conceptual model:

I have 1 dependent variable: WB

I have 3 independent variables: TERT, WEEK, INVES

I have 2 moderating variables: FAM, LAR

I have 7 controlling variables (3 for time and 4 for firm level variables): TIME1, TIME2, TIME3, CHAIR, MULT, IND, ROE

Which would be the command I should use? The examples I have found so far don´t mention moderation or any control variables so I am not sure about what to type after xtreg.
Also, would using MLE or GSL be more preferred?

Thank you in advance!

[Carlo Lazzaro]

Maitena:
welcome to this forum.
If you have three variables for time, how did you -xtset- your dataset?
2) about moderating variables, see -fvvarlist- and https://www.stata.com/why-use-stata/...ith/linear.pdf.
3) I would go -xtreg, re- (GLS).

[Maitena Iriondo]

Dear Carlo,

Stata gave me this when I entered the panel data. It seem OK to me.

Panel variable: compid (strongly balanced)
Time variable: Year, 2018 to 2020
Delta: 1 year

When I say I have 3 time variables I mean 3 dummies to control for time fixed effects: I created 3 columns, one for each year and filled it with 0 or 1 depending on the year I am controlling.

I have seen that moderators work with ##. Thank you for that.

Why wouldn´t my command be:
xtreg womenonBOD_n c.Womenintertiaryeducation##Familyownership1FOB0No nF c.Womenintertiaryeducation##Firmsize1Large0Nonlarg e c.Totalweeksofleavereservedfo##Familyownership1FOB 0NonF c.Totalweeksofleavereservedfo##Firmsize1Large0Nonl arge c.Investmentinchildcarechild##Familyownership1FOB0 NonF c.Investmentinchildcarechild##Firmsize1Large0Nonla rge Firmcontrolvariable1chairgend Firmcontrolvariable2multipled Firmcontrolvariable3industry Firmcontrolvariable4ROE, re

I have doubts wether I have to use re, mle or gsl...
I already used re for Hausman test, so would I have to run it again?

[Carlo Lazzaro]

Maitena:
1) -xtset- outcome is Ok;
2) I fail to get why you can't have one categorical variable only (0, 1, 2 according to the year you're controlling. BTW: why you need another -timevar- when you already have -Year-? Hene you laready checked for perfect collineraity issues?):
3) your code includes tons of predictirss and/or interactions. Please consider a more parsimonious specification of the right-hand side of your regression equation;
4) I would go -xtreg,re-

[Maitena Iriondo]

To control for fixed effects, I have been suggested to do so because the time could be a confound variable I am not controlling for with RE model...
As I have 3 IV and 2 moderators, it is hard to make it more simple ahahaha I know it´s very long...
I want to measure between firm effects comparing 2 countries, but I should include country as a control variables too, right? Because they are very different in terms of the IVs.

[Maitena Iriondo]

Also, as I have said, for the Hausman test I need to run FE and then RE, and then decide based on the Prob > chi2 = 0.9456. As I am already including the xtreg, re command here, and the result says I should do RE, do I need to run it again? Because the command and the results are exactly the same. Maybe in the Hausman I shouldn´t include the interactions and do it once the test tells me to do RE?

[Carlo Lazzaro]

Maitena:
go as you were suggested to if you think it is an effective an efficient way. Probably an example/excerpt of your dataset shared via -dataex- would be enlightening.
As far as -hausman- outcome is concerned, its outcome is identical because it focuses on the coefficients that are common for the -fe and the -re- specifications. Hence, if you add a time-invariant predictor, you obtain a coefficient from -re- only. Hence, being not common between the two specifications, -hausman- skips it.