home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!LE.AC.UK!DDS
- Via: UK.AC.LE; 5 SEP 92 13:30:38 BST
- Site: University of Leicester, Leicester, U.K.
- X-Name: David D Stretch
- Phone: +44 533 522522 ext 3251, or +44 533 523251 (direct line)
- Fax: +44 533 523293
- Organisation: Organisation??? Who?? Me??? Don't make me laugh!
- X-Address: Lecturer in Psychology, Dept. of Psychiatry,
- X-Address: Clinical Sciences Building, Leicester Royal Infirmary,
- X-Address: PO Box 65, Leicester, LE2 7LX, United Kingdom.
- X-Mailer: ELM [version 2.3 PL0 (LUT)]
- Message-ID: <28515.9209051225@irix.le.ac.uk>
- Newsgroups: bit.listserv.stat-l
- Date: Sat, 5 Sep 1992 13:25:26 BST
- Reply-To: dds@leicester.ac.uk
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: David Stretch <dds@LE.AC.UK>
- Subject: npar1way (fwd)
- Lines: 45
-
- Sorry, I am a bit busy at the moment, so my reply will be rather brief...
-
- >
- > 1. Why should these tests give wildly different P values such as
- > P =< .007 in one test and P=<.17 on another?
-
- Because they are testing different things (If you still have it, read my
- previous message about Likert scales, looking at the bit where I talk about
- the paper by MacRae (1988) --- I imagine that going back to the sources *may*
- be the only way of finding out what *exactly* the assumptions (and hence the
- null hypostheses etc) of these tests are, but try looking at Bradley's book
- on nonparametric statistics - although its now old, it may have the
- information you need).
-
- > 2. What statistical or theoretical criteria should I use to determine
- > which test to believe?
-
- If you haven't, you need to look very carefully at what you want help with
- in deciding what's going on in the data. If you can be more specific in
- *exactly* what you are prepared to assume and what the research question in
- this instance is, then this should be used to direct your choice, in preference
- to using purely statistical criteria that do not take account too much of the
- area of application.
-
- > 3. Is there any such measure in any of these non-parametric tests that
- > would be equivelent to a Tukey or Scheffe post-hoc?
-
- Hmmm... I think so - For example, with the Kruskall-Wallis Test, isn't it
- advisable to do post-hoc comparisons on pairwise comparisons using the
- Wilcoxen Rank Sum Test (Mann Whitney U test, to some), but adjusting the
- alpha level (for example, by making sure that the overall significance
- level is still as close to, say, 0.05 as possible - I think this *may* be
- something related to the Bonferroni correction, but others on here will
- probably know more about it than me.)
-
- I hope this helps. Sorry for being very brief - I may suffer being a bit
- incomprehensible in places because of it.
-
- David Stretch
-
- +-----------------------+---------------------------------------------+
- | Dr. David D Stretch | I was gratified to be able to answer |
- | Dept. of Psychiatry | promptly, and I did. I said I don't know. |
- | dds@leicester.ac.uk | Mark Twain |
- +-----------------------+---------------------------------------------+
-