When I took a regression course a while back, I learned the general F
test for testing whether the addition of a set of predictors explains
more of the response. The general F test is:
(RSS(reduced) - RSS(full))/p2
F = ------------------------------
RSS(full)/(n-p)
Anyway, Stata provides Wald tests via the command -test- and likelihood
ratio tests via -lrtest-. I seem to have read here a few times and in
the manuals that -lrtest- is preferred when possible. I don't remember
any discussion of this general F test that I'm more familiar with.
Is the general F test for regression (continuous Y) the same as both
the test and lrtest commands? I gather they must be more general.
Has anyone implemented the general F test for regression, or do I need
to gather the bits and pieces together from running the two models?