Marginal effects from clogit model estimation

[Shayan Khan]

Hi,

I want to know that is there any way to get the marginal effects after running a clogit model/ fixed effects logit model? If not, then are odd ratios the best way to interpret the coefficients?


Thanks

[Joao Santos Silva]

Dear Shayan,

This presentation may help.

Best wishes,

Joao

[Clyde Schechter]

If you are using current Stata, or even one or two versions back, the -margins- command with the -dydx()- option will do this. See -help margins- for details.

As for whether marginal effects or odds ratios are better, that really depends on your audience, their expectations, and their level of understanding.

[Joao Santos Silva]

Sorry, Clyde, but using -margins- after estimating a logit with FE is generally inappropriate; that is the point made in the presentation I suggested.

Best wishes,

Joao

[Clyde Schechter]

Joao,

Your post in #2 did not show up when I wrote mine. That's an interesting presentation and the illustration of the problem is quite convincing; I was not aware of that. I see in your penultimate slide you propose that -margins- should be disabled after -xtlogit, fe- and -xtpoisson, fe-. I wonder why Stata has not taken that advice: they're usually pretty good about things like that.

Be that as it may, thanks for pointing that out.

Clyde

[Joao Santos Silva]

Thanks, Clyde. Hopefully Stata will look into that soon.

Best wishes,

Joao

[Shayan Khan]

Thanks Joao. The presentation has confirmed my doubts about using marginal effects for models employing fixed effects using logit model.

[Mina Sami]

Dear Shayan,
I think that STATA is providing the marginal effect directly after your command clogit. If you type the command " margins, dydx(*) predict(xb)" after your clogit, you will get the same coefficients with dy/dx.

Mina