home *** CD-ROM | disk | FTP | other *** search
/ NetNews Usenet Archive 1992 #30 / NN_1992_30.iso / spool / sci / math / stat / 2639 < prev    next >
Encoding:
Internet Message Format  |  1992-12-20  |  940 b 
  1. Path: sparky!uunet!olivea!spool.mu.edu!agate!stanford.edu!rock!concert!duke!news.duke.edu!duke.edu!feh
  2. From: feh@duke.edu (Frank Harrell)
  3. Newsgroups: sci.math.stat
  4. Subject: Re: stepwise regression
  5. Summary: amen
  6. Message-ID: <8120@news.duke.edu>
  7. Date: 18 Dec 92 02:15:30 GMT
  8. References: <8091@news.duke.edu> <thompson.724621440@daphne.socsci.umn.edu>
  9. Sender: news@news.duke.edu
  10. Lines: 12
  11. Nntp-Posting-Host: biostat.mc.duke.edu
  12.  
  13.  
  14. I very much agree with Scott Thompson's thoughful post. More often
  15. than not, the best prediction arises from a full model fit
  16. (i.e. collinearities aren't really all that dangerous) followed
  17. by shrinkage. Spiegelhalter (Stat in Med 5:421-33, 1986) says
  18. this better than I could.
  19.  
  20. -- 
  21. ----------------------------------------------------------------------------
  22. Frank E Harrell Jr            feh@biostat.mc.duke.edu
  23. Associate Professor of Biostatistics
  24. Division of Biometry                    Duke University Medical Center
  25.