On Tue, 27 Sep 2011, Giuseppe Vittucci wrote:
I was wondering if the F-test computed via "restrict" in
gretl is
heteroskedasticity (and autocorrelation) robust when it follows an
estimation with the robust option.
Yes. This is quite easily verified:
<hansl>
open data4-10
set echo off
set messages off
ols ENROLL 0 CATHOL PUPIL INCOME --robust
# do a Wald test manually
matrix R = zeros(2,4)
R[1,3] = 1
R[2,4] = 1
matrix q = zeros(2,1)
matrix b = $coeff
scalar Fw = (R*b-q)'*inv(R*$vcv*R')*(R*b-q) / 2
scalar pv = pvalue(F, 2, $df, Fw)
printf "*** manual Wald method: F = %g [%g]\n\n", Fw, pv
# now use "restrict"
restrict
b3 = 0
b4 = 0
end restrict --quiet
</hansl>
Allin Cottrell