XTLOGIT, RE vs XTOLOGIT

[George Akomas]

For my D.V I have 3 target markets coded as 0.1 and 2. My I.V include the categorical variables size, age and type, financial crisis; the continuous variables of GDP per capital growth, & of women, mobile subscription, portfolio at risk, ROE, institutional quality, inflation and 6 Time invariant variables for culture. Because of the time inv variables, I believe a random effects model is preferable but get different results when I use the xtlogit with random effects and the xtologit. Why is this and which is preferable? And if I use the xtologit model what can I use to test the validity of the model?

[Richard Williams]

Welcome to Statalist.

It would help if you showed your commands and output, using code tags. See pt. 12 of the FAQ.

My guess is that xtlogit is treating the 1 and 2 categories both as 1, whereas xtologit treats the two categories separately.

Is the dependent variable really ordinal? From your brief description, I would guess it to be unordered.

[George Akomas]

Please find the commands and the results attached.
the D.V is ordinal with 0,1 and 2 representing micro, small and medium markets respectively.

Thank you for your help

[Weiwen Ng]

Please find the commands and the results attached.
the D.V is ordinal with 0,1 and 2 representing micro, small and medium markets respectively.

Thank you for your help

George,

You are comparing an ordered logit model (ordered categorical data) to a logit model (binary data). In the logit model, Stata treated the dependent variable as 0 versus everything else (i.e. collapsed 1 and 2 like Richard said). They are not the same model. Many of your findings actually appear to be in the same direction and similar in magnitude. But what were you expecting?

I wasn't familiar with the xtologit command, but looking at the manual, -xtologit- will only fit random effects models. So, I'm not clear why you are discussing random effects versus fixed effects. You compared one random effects model to another random effects model (by the way, the default for -xtlogit- is random effects). Are you trying to compare a fixed effect version of the -xtologit- regression to the default random effects specification? If so, I have no idea how to accomplish that, aside from -ologit y x i.panel-, but maybe someone else has an idea.

[George Akomas]

Thank you so much for the explanation. I am not actually trying to compare. My data has time invariant variables and I was looking for the appropriate model to use. As -xtologit- fits only random effects effects models, how do I carry out a Hausman test as I cant seem to find a command for ordered logit model with fixed effects.

Once again, thank you for the support

[Richard Williams]

Check out this thread. Stata does not offer fixed effects ordered logit, but others claim that it can be done. I can't vouch for the techniques.

https://www.statalist.org/forums/for...git-regression

[Richard Williams]

Also see

http://ftp.iza.org/dp5443.pdf