st: re: confidence intervals on r-squared

[Kit Baum <baum@bc.edu>]

Marcello says

Unless, of course, you use an r^2 that accounts for the number of
covariates you include. That is why I asked about which r^2 you are
interested in. To say that "R-squared is not a statistical concept." is
silly.

Silly, perhaps. But I also think talking about 'which r^2' is a bit  
silly. r^2 is a distinct concept: the squared Pearson correlation  
between observed and predicted. Marcello is certainly correct in  
suggesting that we _could_ speak of a confidence interval around a  
correlation coefficient. Strangely enough a significance level cannot  
be produced by -correlate-, although it can by -pwcorr-. There is, as  
Nick Cox points out, an entire bestiary of R^2-type measures, pseudo- 
r^2s, r^2 from a model lacking a constant term, etc. But when we  
speak of r^2 arising from linear regression with a constant term  
included I don't think there is room for disagreement about "which one".

Of course, what you may observe in this dialogue is the difference  
between a real statistician and a nominal statistician. Marcello has  
a much stronger claim on the subject than do I. I just use the stuff.

Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


*
*   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/