Marginal effects in logit with fixed effects

[Alessandr De Palma]

Dear Statalists,

I am conducting an analysis with a fixed effect logit model. I am aware of the incidental parameter problem. As I see from Stata, xtlogit and clogit handle the problem estimating a conditional logistic function. The only problem is that the estimation of average marginal effect assumes fixed effects to be costant (=0). Thus, the AMEs are not precise (am I right?). Moreover, when I include linear time trends (c.year) in my regression and afterward I try to compute AMEs through the "margins, dydx" command, I get this warning: "prediction constant over observations". And all the AMEs are equal to 0 and st.errors are omitted. Unfortunately, I cannot show you my output since I am working from a disconnected room, given that I am working with sensitive data. So, my questions are:
- Do you know something about that warning?
- Given that marginal effects are not well computed in logit with id fixed effects, should I present log odds or odds ratio in my paper result tables?
- Or should I just use a Linear probability model?

I truly thank you in advance for your support!

Alessandro.

[Alessandr De Palma]

Dear All,
Can someone give me a hint about this? I would really appreciate it!!

[Joao Santos Silva]

See if this helps.

[Alessandr De Palma]

Thank you Joao Santos Silva! So, what you suggest is to present average semi-elasticities? Can you give me an example of how to interpret them? Moreover, what happens if I include interaction terms?

[Alessandr De Palma]

Dear Statalist,
Can anyone give me a definition of semi-elasticities and how to interpret them when I include interaction terms. For example, if I want to see the impact of temperatures on air conditioning adoption, and I include an interaction between temperature and whether you are the owner or not or the house. And suppose the coefficient of the semi-elasticity is positive = 0.005. How do I interpret it?
Thank you!!