Conditional logit (clogit) for cross-sectional data

[Thomas Hoppe]

Dear all,

I'm working on a project in which we analyze firms' location decision, i.e. the decision in which country to locate. Unfortunately, our variable of interest is available for one year only and, hence, we have to work with cross-sectional data. Our data set looks like this:

 
Location
Country
Subsidiary
Parent firm
Control variables

1
US
ABC
X
...

0
UK
ABC
X
...

0
AT
ABC
X
...

At the moment, we employ a conditional logit model (clogit) which, in a simplified version, looks like this:

clogit Location VariableOfInterest $controlvariables, group(Subsidiary) nonest vce(cluster Industry)

Are there any problems associated with the use of a conditional logit model for cross-sectional data? As clogit is often referred to as a panel data model (conditional fixed effects logistic regression), we feel increasingly insecure about using it for cross-sectional data.

We further wonder whether there are any special issues with regard to the post-estimation commands, such as margins (from here we already got that it might not be possible to estimate reliable marginal effects).


Thank you very much in advance and best regards,
Thomas

[Phil Bromiley]

While you don't have years, you do appear to have multiple observations per panel if the panel is Subsidiary. So clogit could be quite appropriate. There is a problem that predicted values assume one positive per panel - this has been discussed several times on Stata List.

[vrinda rawal]

Hi Thomas Hoppe

I am facing similar issues, can you please suggest the appropriate solution that you end up using for this problem.

Thanks