OGLM command

[Dario Flood]

I'm having trouble finding out how to include fixed effects when I run the OGLM command.

My data uses subjective well-being and income taken from BHPS so I am running an ordered probit regression. I use the command xtoprobit to account for individual random effects in my normal regression, but I also want to account for these effects in a model that accounts for heteroskedastic error terms.

Roberta Distante in her 2012 paper called 'Subjective Well-Being, Income and Relative Concerns in the UK' managed to account for fixed effects using the oglm command and I have no idea how.

Any help?

[Charlie Joyez]

Unfortunatly, I think that oglm (unlike gologit2 for instance) doesn't allow factor variables (i.var), that are usefull to include individual terms in the regression and act as fixed effects.
However, you could try with the xi prefix, it is said to work with oglm.
For more details see the help files but also Richard Williams oglm page, and his various article on it.

Charlie

[Richard Williams]

No, oglm allows for factor variables; in fact it was way ahead of gologit2 in doing so. But as to how to control for fixed effects I don't know. (Just include a lot of dummies?) I occasionally get requests for xtoglm and xtgologit2 but I don't know how to program them.

[Richard Williams]

Incidentally, whenever I get a request for something I don't know how to do, my usual response is that if you can figure out how to do it with ologit you can probably do something very similar for oglm or gologit2.

[Charlie Joyez]

Ok, thanks for the precision Richard, and sorry for the mistake on oglm, have have to catch up the updates!

So if factor variables are allowed, using them (adding a lot of individual dummies) will control for individual unobserved invariant characteristics, which was what I meant by saying that they act as fixed effects.
However, adding a lot of dummies raises new issues (freedom desgrees, autocorrelation, omitted variable, and so on.)

The pannel command with -xt- (xtreg, xtlogit,...) don't act the same and are based on individual mean-differencing, to remove indivual trend (I learn that in some Cameron & Trivedi book, microeconometrics using Stata, chapter 8).
It might be a lead for you Richard to write a xtoglm and xtgologit2! (But I guess you already knew that).

[Dario Flood]

Thanks, including individual dummies isn't an option for me sadly. Thanks for your help though.