home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!stanford.edu!bcm!convex!darwin.sura.net!paladin.american.edu!auvm!NIOSHE2.EM.CDC.GOV!DKW4
- Return-Receipt-To: <dkw4@NIOSHE2.EM.CDC.GOV>
- Encoding: 47 TEXT
- X-Mailer: Microsoft Mail V3.0 (beta-2)
- Message-ID: <2B00F525@router.em.cdc.gov>
- Newsgroups: bit.listserv.stat-l
- Date: Wed, 11 Nov 1992 15:55:00 EST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: dkw4@NIOSHE2.EM.CDC.GOV
- Subject: Friedman's Test w/Many Ties
- Lines: 39
-
- STAT-Lers,
-
- Thanks to everyone (Walter Pirie, Mike Fuller, Tony Dusoir, Matt Zack, and
- Tim Green, and anyone else?) who responded to my question of a few weeks
- ago. I asked if anyone knew what happened in Friedman's Test if there were a
- LARGE number of ties (measuring pain in body parts, in this case) in each
- block.
-
- The consensus seemed to be that a large number of ties reduces the power of
- the test. Also, I received differing advice on how to handle the problem,
- but I certainly appreciated all the thoughts.
-
- Lastly, a colleague, who is an Ohio State University graduate, asked Doug
- Wolfe his opinion about this situation.
-
- Doug's response follows....
-
- "The Friedman test does tend to be conservative in the presence of lots of
- ties.
- I would suggest running it anyway--if it rejects you are relatively safe in
- declaring significance. If it does not reject, then I would suggest
- dichotomizing the data (as suggested in the memo) and see if the test on this
- reduced data set leads to rejection.
-
- If one had the patience, one could generate the permutation distribution for
- Friedman's statistic, given the ties configuration. The inference would then
- be exact.
-
- Again, good to hear from you.
-
- Doug Wolfe"
-
-
- Thanks again -
-
- Deanna (Wild)
- National Institute for Occupational Safety and Health
- Cincinnati, Ohio USA
- Internet: DKW4@NIOSHE2.EM.CDC.GOV
-