John said
Having successfully harvested my crop of beta coefficients, I've turned
my attention to other parts of the regression report. I'd like to
accumulate the t-stat and p-value for the various levels of the anova.
I thought I might be able to recover them from the e(return) list,
where an F-stat is kept, but I don't see the relationship between the
t-stat in the regression output and the corresponding F-value ( t^2= F
with degrees of freedom of (1,N-2) ) that I was expecting. Is that not
Student's t in the regression table? Is this another quirk of anova
interactions that is spoiling the t/F relationship? Is it possible to
capture the t-stat in some other way...perhaps from the e(V) matrix
(although that didn't look straight-forward either).
I'm not that familiar with how -anova- kicks out its results, but you
should be able to work with the diagonal elements of the e(V) matrix
just as easily as the elements of e(b), right? Their square roots are
the estimated standard errors, so the ratio of a b/sqrt(V) is a
"t-stat". The relationship between t(.) and F(1,.) only applies for one
numerator df -- so if using auto you do
anova mpg foreign rep78
anova, regress
It is indeed true that the t on foreign in regression, squared, is the
F reported in ANOVA. But the F for rep78 cannot be constructed from the
4 t-stats, since it is a joint test.