Logistic regression with industry and year fixed effects

[Nik Honig]

Dear Statalist Forum!

I want to perform a logistic regression in Stata. I am analyzing companies over a period of time, so I have a panel dataset in which gvkey represents the company and year the time. Moreover, I have a variable called industry representing the industry.

I defined my dataset as a panel dataset using the following command: xtset gvkey fyear

I want to regress my dependent variable A on the independent variables B C D and E. I want to include fixed effects for the year and the industry. Hausman test showed that I should go for fixed effects.

Now, I am not 100 % sure which command is the correct one. I am currently considering three options:

1) xtlogit A B C D E i.year i.industry
2) xtlogit A B C D E i.year i.industry, fe
3) clogit A B C D E i.year i.industry, group (gvkey).

Could you help me out and give me an advice for the correct one? Moreover, is one of them correct at all?

Best!

[George Ford]

TWFE with logit has problems (inconsistent parameter estimates).

Some discussion here:

https://www.statalist.org/forums/forum/general-stata-discussion/general/1589493-fixed-effects-for-a-logit-model

https://www.econstor.eu/handle/10419/145837

[George Ford]

look at clogit too

[Nik Honig]

Hi George Ford,

thanks for your response! So if I do understand you correctly, you advice me using my option number 3) ?

[George Ford]

Sure, but really think about what all those FE are doing and whether it's legit. FE and Logit don't play nice, so spend some time on it.