Fixed effects for ID in a linear regression?

[Marcus Eklund]

Hi all,

Just a rather simple question that I can't get my head around. I have a dataset in a long format with around 1000 IDs. About 200 of them occur two times (the collection of the data was over two years), while the rest are only included once. I want to make a linear regression (not a panel regression), but how should I control for the recurring IDs? Should I include fixed effect for ID, or is it better to include fixed effect for year?

Best,
Marcus

[Carlo Lazzaro]

Markus:
the issue here rests on your dataset, not on the methods.
That said, if you want to go -regress- under a presumed -fe- specification, you should add both -panelvar- and -timevar- categorical predictors and, with 200 out of 1000 IDs with 2-wave data, this approach is totally inefficient.
Therefopre, i'd go:

 xtset panelid timevar
xtreg <depvar> <indepvars> <controls> i.timevar, fe vce(cluster panelid)