home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!pagesat!spssig.spss.com!nichols
- From: nichols@spss.com (David Nichols)
- Newsgroups: bit.listserv.stat-l
- Subject: Re: Wanted: Rank test
- Message-ID: <1992Nov24.011503.15751@spss.com>
- Date: 24 Nov 92 01:15:03 GMT
- References: <STAT-L%92112311112401@VM1.MCGILL.CA>
- Sender: news@spss.com (Net News Admin)
- Organization: SPSS Inc.
- Lines: 65
-
- In article <STAT-L%92112311112401@VM1.MCGILL.CA> Walt Pirie <WLTPIRIE@VTVM1.BITNET> writes:
- >For the problem below, two ideas come to mind. One is Kendall's Test of
- >Concordance. Some Nonparametric books include it, or look in Kendall's
- >Rank Correlation Book. The other is the General Linear Models by Ranks
- >procedure by Hettmansperger and McKean, but I don't know if there is
- >a general software package for that. It's much too complex to do by hand.
-
- I don't think that Kendall's coefficient of concordance addresses the
- question asked here by Gene Glass. He wants to know (on my interpretation)
- if the men and women rank things differently. The Kendall coefficient of
- concordance will test whether the rankings of all raters, male or female,
- are unrelated. I just looked through Kendall's book, and I don't see any
- test in there to address this question. He makes reference near the end
- to several people who have developed tests for this situation in which
- multiple ranks are generated by raters forming groups, but does not
- discuss any such test in the book (unless I missed it).
-
- > I can't believe it! I'm too old for this. I had a student gather a
- > bit of data thinking that when the time came I would just pick up
- > Siegel or some other hoary text and look up the proper significance
- > test. Now I can't find any test that addresses the problem.
- >
- > A group of 5 men ranks four things; and a group of five women do also:
- > Thing1 Thing2 Thing3 Thing4
- > Man 1 : 4 2 1 3
- > Man 2 : 4 3 2 1
- > .
- > .
- > .
- > Woman 1 : 3 2 4 1
- > Woman 2 : 3 4 1 2
- > .
- > .
- > etc.
- >
- > Now, I'm not obsessive about being fooled by chance, but I would
- > like to know whether the disparity in the rankings between the men and women
- > is of suffucient size that it is unlikely to have arisen solely from the
- > chance selection of these two samples from populations in which men and
- > women rank the four things similarly (or identically or something).
-
- If there were no structure on the raters (no gender/sex designation),
- Kendall's coefficient of concordance could be used to test the null
- hypothesis that the rankings are unrelated. Friedman's test could also
- be used to test the null hypothesis that Thing1 to Thing4 all have the
- same ranking in a population of interest. But as far as I can tell,
- neither of these is really what is wanted here.
-
- SPSS basically has Siegel's set of nonparametrics in our package. We don't
- have anything that would address this question in the manner posed (at
- least as I read it). Unless I'm misinterpreting the question, this is
- somewhat analogous to a repeated measures ANOVA with two groups, except
- that instead of having means of four correlated variables to compare,
- we have four ranks. The question of interest here would seem to be the
- status of what would commonly be called the interaction term; that is,
- is there a difference in the ranking orders between men and women (the
- standard group term here wouldn't come into play, because the ranks
- sum to 10 for each rater). I've not seen a test for this, though I
- would bet that one must exist.
-
- --
- David Nichols Senior Statistical Support Specialist SPSS, Inc.
- Phone: (312) 329-3684 Internet: nichols@spss.com Fax: (312) 329-3657
- *******************************************************************************
- Any correlation between my views and those of SPSS is strictly due to chance.
-