vce(cluster clustvar) & fe with logit

[Sunita Rao]

1. Stata documentation includes provides examples of vce(cluster clustvar) with the regress command. Does it make sense to use vce(cluster clustvar) with the logit command? That is, will I be able to interpret the direction and magnitude of the coefficients if I use vce(cluster clustvar) with the logit command?

2. Does it make sense to use the fe option with the logit command? Are the logit command and xtlogit command the same?

[Carlo Lazzaro]

Sunita:
- a cautious reply would say: it depends. under -logit-, -robust()- and -cluster()-options take heteroskedasticity and autocorrelation into account, respectively. Under -xtlogit- both options do the same job. However, they do not affect magnitude and/or direction of the coefficient point estimate;
- -logit- and -xtlogit- are not the same beast. -xtlogit- is devised for analyzing panel data with categorical dependent variable. Panel data implies >=2 waves of data, whereas -logit- imply one wave of data only;
- due to the so called incidental parameter problem (see, if interested, http://www.econ.brown.edu/Faculty/To...meters1948.pdf) -fe-in -xtlogit- are actually conditional -fe- (a very different beast from the -fe- specification that we use under -xtreg-). As far as the requested advice about using or not -fe- in -xtlogit- is concerned, the best reply that springs to my mind is, again: it depends (on your research goal and dataset, at least)..

[Sunita Rao]

Dear Dr. Lazzaro,

Thank you so much for your response. I would be grateful for your thoughts on the following:

1. My dependent variable is categorical/ordinal with ranks. It is marked 0, 1, 2, 3. These four categories are ordered No international frameworks (marked 0), using AA1000 (1marked ), using ISAE3000 ( marked 2) , and, using both AA1000 & ISAE3000 (marked 3).

2. My observations are companies - each company has data points that have various predictors for 2-8 years. Does that mean >=2 waves of data? Should I be using xtlogit? or ologit? I am attaching my do file to this message.

3. Does outreg work with xtlogit and/or ologit?

Warm Regards,
Sunita

[Carlo Lazzaro]

Sunita:
-if your regressand is actually ordered (such as good; better; the best) and, as it seems from your description, you have a panel dataset, you should go -xtologit-.
Please note that, for sound theoretical reason, -xtologit- allows -re- specification only.
Unfortunately, I cannot say about the user-written programme -outreg-.
The best way to share excerpt/example of your dataset is via -dataex-, not attachments.
PS: please call me Carlo, as all on (and many more off) this list do. Thanks.

[Sunita Rao]

Hi Carlo,

Thank you so much for your response once again. It is very helpful.

I will certainly remember about -dataex-

Warm Regards,
Sunita