st: clogit with cluster-level variables

["Lim, Raymond" <rl2240@columbia.edu>]

I am trying to run a firm fixed effect logit model and cluster the standard errors by industry-year, thus I'm using clogit, ...group(ID) cl(industry_year). However, one of my control variables is measured at the industry-year level and the standard errors on this variable explodes with the clustering. Strangely, this is not the case when I run the same specifications using OLS. 

For illustrative purposes, let's say the command is 

egen industry_year=group(industry year) 

clogit improve funding i.year, group(firmID) cl(industry_year) nonest 

xtreg improve funding i.year, fe i(firmID) cl(industry_year) nonest

Where improve is a dummy variable and funding is measured by industry and year. The xtreg comes out statistically significant, but the clogit's standard errors are 1000 times larger. 

Any ideas as to why the standard errors would explode under logit, but not ols?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/