FEM vs REM which one to use?

[Jan Gleiman]

Hello,

My questions are as follows:
Should I use a Random Effects Model (REM) or a Fixed Effects Model (FEM)?
Am I interpreting the results of the tests correctly?
Am I running these tests correctly?
Should I consider a mixed effects approach?

I am using STATA IC 15 and I am trying to determine whether a fixed effects model (FEM) or Random Effects Model (REM) is appropriate for my research. I am using unbalanced panel data. I am using Hausman test, Breusch-Pagan, Bera Sosa-yoon, and other tests that are embedded in the commands xttest0, xttest1, xttest2, and hausman. For the models that I plan to run, my dependent variables are binary and categorical. Some of the independent variables are continuous, while others are factor variables.

I am getting contradictory results. Hausman shows Prob chi2=0.007 (FEM is more appropriate) and Breusch-Pagan shows a similar result (REM is more appropriate). xttest2 fails to run because I have too many panels and STATA IC only allows a matsize of 800.

Here is an abbreviated sample of the commands I am using

xtset dyad year
xtreg Y1 X1 X2 i.X3..., re
estimates store re
xttest0

xtreg Y1 X1 X2 X3..., re
* I have to remove the factor variable indicator for this test to run*
estimates store re1
xttest1

xtreg Y1 X1 X2 i.X3..., fe
estimates store fe
hausman fe re, sigmamore

xtreg Y1 X1 X2 i.X3..., fe
estimates store fe
xttest2


Thanks
JKG

[Sagnik Bagchi]

I guess we use xttest0 to select between random effects and the pooled model. While for the FE model, you consider the Hausman test. Also, while looking at the results of the FE model, look into corr within values and the F value of all u_i = 0. They also give you a fair indication of the FE model.

You may consider going to the following link: https://www.princeton.edu/~otorres/Panel101.pdf

[Ruchika Agarwal]

First of all,shouldn't you be using the xtlogit command because your dependent variable is binary and not xtreg(which is for when dependent variables are continuous variables)?