Fixed effect in xtlogit

[Sungida Rashid]

Dear Scholars,
I am trying run a xtlogit model with fixed effect on industry, location and year. I used the following commands

xtset panelid year
xtlogit CREDIT d3c lnage proptnownL lctn_dhk lctn_cht d3b domestic_firm state_firm exp_manager sex_manager i.industry i.year

I am trying to find industry and year fixed effect. My question is do I need to write fe in my command? or I am good to go?

Thanks.

[Carlo Lazzaro]

Sungida:
your code seems to miss a -, fe- at ist end.
Then you can go that way.

xtset panelid year
xtlogit CREDIT d3c lnage proptnownL lctn_dhk lctn_cht d3b domestic_firm state_firm exp_manager sex_manager i.industry i.year, fe

As an aside, please be advised that under -xtlogit, fe- you will get coditional fixed effects, which are different from the ones you can get under -xtreg, fe-.

[Ivan Privalko]

You could also use the clogit command here. More information attached. The coded is a bit simpler, faster, and no need to xtset the data!


PS fixed effects will not be available for variables which do not vary by clusters. If your data is arranged by time, for example, measures like industry might be affected. This can be a bit of a headache, FWIW.

[Joao Santos Silva]

Dear Sungida:

Just to add to the excellent advice already provided, notice that the panel needs to be "large" in the dimensions of the fixed effects that you include in the model as dummies; it does not have to be large in the dimension of the fixed effect you condition out.

Best wishes,

Joao

[Sungida Rashid]

Thanks for your valuable suggestions. My dependent variable Credit is a dummy. Total sample size is 3298. I tried both xtlogit,fe and clogit, group(). But both of them takes aroung 5 to 6 hours to compute and finally tells me "congergence is not achived". To be mentioned, there are huge sample drop (around 2300) in both codes. Can I solve this issue in any other way?

Thanks in advance.

[Ivan Privalko]

Hi Sungida,

The reason why you have a huge drop in the sample is because fixed effects will only compute estimates for clusters which move between values of 0 and 1 for Credit. So if a cluster holds only values of "0" for each observation, or only values of "1" for each observation, it cannot contribute to the model. It sounds like your Dependent Variable tends to stick to one value. My 2c.

[Ivan Privalko]

Alternatively, it could be one of your independent variables, I would try building the model one variable at a time. Maybe one of you controls varies little over "year"

[Sungida Rashid]

Thanks a lot for your suggestions Ivan Privalko. I will recheck all of my variables again.