Am Donnerstag, den 27.09.2007, 05:05 -0700 schrieb Lloyd Dumont:
> Hello. I am running an OLS model in which
> observations fall into one of three mutually-exclusive
> and collectively-exhaustive categories. For clarity
> in reporting, I thought it would be a good idea to
> suppress the constant and report slope estimates for
> all three dummies.
>
> If I run the model both ways (either with two dummies
> and the constant vs. with all three dummies and no
> constant), the estimates and the standard errors are
> what they should be, i.e., are the same in relative
> terms to one another in both models, same t-stats,
> etc. But, without the constant, the R2 shoots up from
> something like .11 to something like .68.
>
> I sort of understand conceptually how this could
> happen--fit is now relative to zero than to the mean.
> But...
>
> 1. Is my understanding correct?
> 2. How can I explain this succinctly?
> 3. Am I being deceptive to report the .68?
>
> Thank you. Lloyd Dumont
Use option "hascons" togehter with "nocons".
On that occasion: There is an article in a German
econometrical/statistical journal, which heavily critizes Statistical
packages (including Stata) for using the term "r-square" if the
regression is forced through the origin. The authors emphasize that
r-square cannot be interpreted in the usual way under this setting and
should be therefore renamed or omitted from the output. Only "Minitab"
does this.
Ring/Ryll/Gaus (2006): Das Bestimmtheitsmaß R² bei linearen
Regressionsmodellen mit und ohne Intercept -- die Tücken der
Statistikprogramme Wirtschaft und Statistik, 2006, 11, 607-612.
(The third author is Wilhem Gaus, not Carl Friedrich, btw)
Many regards
Uli
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
