I have the task to compare regression models. The models are not
nested. For the simplest case let's consider the two simple regressions
ols y 0 x1
ols y 0 x2
Is it possible to test if x2 is a significant better regressor of y than x1 is? Should I
use an F-test to test this hypothesis? Isn't something like this already bulit in in
gretl?
thanks for any comments!
Daniel
Hi Daniel,
Basically, the F-test can not be used for nonnested models. In your
case, I think that you have 2 possibilities :
(1) compare the adjusted R-squared of the 2 models hoping that the
difference might not be very small ;
(2) or more naturally, turn into a nested model so as you can make use
of the F-test. Consider estimating the following model < ols y 0 x1 x2 >
and then proceed with a F-test for each simple model against the
composite one. However, if the 2 simple models are not very different in
terms of goodness-of-fit, the F-test may result in rejecting (or
accepting) both models.
artur