Variance explained by company FEs vs variance explained by independent variable in Conditional Logit Model

[Vera Mikhailovna]

Is there a way of comparing the variance explained by company fixed effects vs variance explained by certain independent variable? So for example, in the following dataset, how I can compare
the variance explained by read score to variance explained by pid fixed effects?

use https://stats.idre.ucla.edu/stat/stata/webbooks/logistic/hsbcl, clear
logit honcomp read math
clogit honcomp read math, group(pid)

[Maxence Morlet]

ssc install rego

[Vera Mikhailovna]

Thank you but I guess it only works with linear regression

[Jeff Wooldridge]

Vera: The problem is that clogit is not a "fixed effects" estimator in the sense that it doesn't include a dummy variable for each pid. It uses a conditioning argument similar to the within transformation in the linear case. In the linear case, the within transformation produces the same estimates as putting in pid dummies. But this is not true for clogit. In fact, putting in unit-level dummies is a poor idea unless you have a lot of observations per pid. You haven't shown us enough to say more.

Even if you have a lot of observations per pid, this is not an easy question. The logit model is nonlinear and the unit effects do not separate out from the covariates. I could imagine doing the following. Use logit with i.pid if you have enough observations per pid. Then, if the estimated effects are a^(i), you can study the variance of a^(i) compared with xij*b^. This will tell you the relative variance on the log odds. I think this is the best you can do.

[Vera Mikhailovna]

Dear Professor Wooldridge, thank you very much for reply. I did exactly you said and I found that a^(i) > xij*b^