Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Joerg Luedicke <joerg.luedicke@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Why F-test with regression output |
Date | Wed, 4 May 2011 17:54:31 -0400 |
On Wed, May 4, 2011 at 5:19 PM, Steven Samuels <sjsamuels@gmail.com> wrote: > > Nick, I've seen examples where every regression coefficient was non-significant (p>0.05), but the F-test rejected the hypothesis that all were zero. > This can happen even when the predictors are uncorrelated. So I don't consider the test superfluous. This is not surprising since "p>0.05" does not mean that the contribution of a predictor is zero. I cannot see an argument here why this F-test is not superfluous? I personally think that these kinds of omnibus significance tests are useless since they carry little information and thus little meaning. Say we have a model with 5 predictors, I want to see what each contributes in terms of effect sizes. If I see that all effects are essentially zero I can interpret that accordingly. What would it help if I looked at the F-test which does not even carry information about things like effect sizes etc.? I guess it is always printed out because it became standard at some point in time, and I believe especially in Psychology and experimental research. In Psychology, a common modeling strategy is to do an omnibus test first and if the null-hypothesis is rejected a more closer look is warranted. If not, the model gets discarded right away. J. * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/