Fixed effects in a mlogit

[Frederik Thorbjoern]

Hello StataCorp,

I want to control for country fixed effects in a mlogit

I have tried the following code:

xtset country
xtmlogit depvar A B c.C##c.D

However, it returns the error code: infeasible number of permutations

I have also tried using femlogit:

femlogit depvar A B c.C##c.D

h

However, apparently femlogit does not work with factor variables

Is it okay to instead run the regression as below or is that bad practice in nonlinear models and will mess up the results?

mlogit depvar A B c.C##c.D i.country

I have 16 countries with observations ranging from 3 to 302 within each country.

Any suggestions or comments on using fixed effects for a mlogit?

Best, Frederik

[Stephen Jenkins]

I suspect you're misunderstanding the term "fixed effects". You have individual-level data -- individuals within countries. Yes? -xtmlogit- (built-in) and -femlogit- (Stata Journal, by Klaus Pforr -- as you should say: see FAQ) refer to 'fixed effects' at the individual level. You appear simply to want to pool your data and allow intercepts to vary by country. In this case, your final specification with i.country is the right way to go.

[Frederik Thorbjoern]

Thanks for replying Stephen. I have data for 900 parent firms OECD and their entry into one of the BRICS countries across years from 2013 - 2021. Parent firms are from one of the 16 OECD countries, and I wanted to check for unobserved heterogeneity not explained by my cultural and uncertainty variables presented on country level. I am not sure if I totally understand if i.country is the best way to go then?

[Andrew Musau]

See the discussion in https://www.statalist.org/forums/for...uated-at-means on the misuse of the term "fixed effects", which goes to Stephen's point.