Controlling firm fixed effects in the ordered logit regression model

[Mohan Jangid]

Hello Members,

I am running an ordered logit model where my dependent variable ranges from 1 to 8. I want to control for a firm fixed effect for this model as the Hausman test for choosing fixed or random effects recommended running a fixed effect model for this data. Since the Xtologit command in Stata runs a random effect model, How can I control the firm fixed effects in an ordered logit model?

Further, I have tried to use the clogit command with the group(Firm) option to control for firm fixed effects in Stata, but it has failed to run and shown an error that "outcome does not vary in any group". What are the other options I have at my disposal? Can I use the pooled ordered logit model for this data?

[Andrew Musau]

You can take a look at feologit from the Stata Journal, although I have not seen wide adoption of this technique in the literature.

search feologit

clogit is conditional binary logit, and can be applicable if you are able to combine some categories so that only two remain.

[Mohan Jangid]

Thanks for the reply. I"ll try to explore feologit if it is suitable.

[Jeff Wooldridge]

I’d recommend using the correlated random effects approach: add the time averages of the explanatory variables. This gives FE in the linear case but doesn’t suffer from the incidental parameters problem.