Degree of Freedom in panel data

[Pedro Antonio]

Hi,

I've used xtreg to estimate a FE model with N=48, T=7 and nine additional explanatory variables (plus constant). When use testparm to test X6=X6=X8, Stata14 output is:


testparm X6 X7 X8

( 1) X6 = 0
( 2) X7 = 0
( 3) X8 = 0

F( 3, 47) = 1.58
Prob > F = 0.2060

but I can't understand the denominator degrees of freedom. With N·T = 336 they should be 279. How is wrong with this calculation?

[George Ford]

I think when you cluster or fe, Stata uses a conservative approach by using the number of groups (G - 1) (the group observations are not independent). You can force a different df, but you'd need to study the implications.

[Pedro Antonio]

I see your point. Thanks.