Re: [Gretl-users] F-test: Fixed Effects vs Pooled OLS
Am 20.07.2016 um 15:28 schrieb Bruno Gobeil:
Hello!
I would like to better understand the rational for using the following
formula for determining the denominator degrees of freedom for the
F-test: n(t-1)-(k-1); where n is the number of regions, t the number of
years, k the number of parameters, including the constant. I thought
that the use of the de-meaned data approach compared to LSDV approach
for the fixed effects model would lead to a higher number of denominator
degrees of freedom, as suggested in the Gretl Guide, with the following
formula: nt-k.
If I understand your question correctly, then here is my answer: LSDV
and de-meaned data fixed effects ("within" estimation) are equivalent
for testing purposes. To put it differently, de-meaning is also a
statistical estimation procedure, because you do not know the expected
population values a priori and hence you have to estimate them via the
arithmetic means. Hence you cannot gain d.o.f. by the within/de-meaned
estimator.
This is not gretl-specific, of course.
hth,
sven