Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Nick Cox <njcoxstata@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: test of significant between coefficients |
Date | Tue, 27 Sep 2011 16:35:47 +0100 |
Richard's answer overlaps with mine, which is fine. I want to underline the idea that often coefficients should be thought as being bundled together. For example, if a cosine term is included in a model a sine term should be too. Leaving out one or the other can omit some useful information about phase even if one coefficient is not significant. A more widely familiar example is a set of indicators. Degrading them so that all are significant just coarsens a model. Come to think of it, we've have had this discussion before. Just search for "Richard Williams" in the Statalist archives. Nick On Tue, Sep 27, 2011 at 3:48 PM, Richard Williams <richardwilliams.ndu@gmail.com> wrote: > At 10:35 AM 9/27/2011, Andrea Rispoli wrote: >> >> Dear Statalisters, >> I am running a test of significance between two coefficients of the >> same OLS regression. >> My question is : if the two coefficients are not significant, does it >> still make sense to conduct the test? I am asking because sometimes >> while the individual coefficients are not significant the difference >> between them is significant, so I was trying to understand the meaning >> of this result. >> Thank you! >> AR > > It can happen. The individual tests are testing whether the coefficients > equal zero. The equality test might be testing whether, say, -.5 > significantly differs from .5. In any event, there is nothing that says all > your tests have to be logically consistent with each other. The overall F or > chi-square statistic might be significant for a model, while none of the > individual coefficients are. > > A more common situation might be where a coefficient is significant in one > group but not in another. I always warn my students to be careful about > saying X is important for one group but not the other. If, say, you are > comparing whites and black, your white sample size might be much larger, > which can help the effect to achieve significance for whites but not blacks. > The actual estimated coefficients, however, may be quite similar. > * * 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/