Dear Pierre Boutros ,

With Stata, you can try this (I am using Stata 16):

* Install the command firthlogit if you have not done it (or update it)

net install firthlogit, replace

* Use, for instance, the hiv1.dta. It contains three variables (hiv, cd4 and cd8). For illustrating the functioning of the command, assume that cd4 is categorical and cd8 is continuous. Indicate that cd4 is categorical in the regression so that Stata knows it and treat it as categorical when computing the marginal effects.

webuse hiv1, clear

firthlogit hiv i.cd4 cd8

* Marginal effects (average of the marginal effects across the sample)

margins, dydx(cd4 cd8)

I hope it helps. Mark the question as solved if that is the case.

Best,

José-Ignacio

Dear Dr José-Ignacio Antón,

Thank you! I have one more question.

Do you think that I should interpret the marginal effects of the firthlogit model or I should stick to the odds ratio as other papers are doing so?

In my opinion, this is mainly a matter of taste. I strongly prefer marginal effects. Sociologists and medical doctors usually report ORs. The conclusions you draw from them must be similar. If I remember well, I think that in order to get the ORs you have just to specify the option or in the filthprobit command.