Computing standard errors with xtlogit

[Juan Martin]

Dear Stata Masters,

I have a very large panel data set with data at three levels -firm, workplace and workers- over 7 years. The outcome variable is binary, so I'm using a xtlogit model. Independent variables are continuous (i.e. wage, firm size), factors (i.e. industry, location, occupation) and dummies (i.e. gender, year).

I am running a DD logit model: if I run it with random-effects (can't do fixed-effects because the dependent variable shows no variation for many subjects), it works fine. Tricky part is if I want to go a step further and cluster the standard errors at firm level (where "treatment" is actually taking place). Then I get the following message:

"panels are not nested within clusters"

Likewise, when I try the "robust" option, the "calculation of robust standard errors failed".

Does anyone know why this might be the case and/or what I can do to fix it?

Thanks in advanced,
Juan

[Maarten Buis]

If you want to use the three levels of nesting you can use melogit instead of xtlogit.

[Zoe Williams]

Hello,
I am having a similar problem. I have a panel data (country-year) and when I try to do a random effects xtlogit, vce(robust) I get the message that the "calculation of the robust standard errors failed". I am trying to include lagged independent variables in my model and when I remove them it is able to calculate the s.e.s Is there a way around this?

The lagged variables are theoretically necessary.

Thanks very much,
Zoe

[Zoe Williams]

Hello,
I am having a similar problem. I have a panel data (country-year) and when I try to do a random effects xtlogit, vce(robust) I get the message that the "calculation of the robust standard errors failed". I am trying to include lagged independent variables in my model and when I remove them it is able to calculate the s.e.s Is there a way around this?

The lagged variables are theoretically necessary.

Thanks very much,
Zoe