home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.mac.apps
- Path: sparky!uunet!psgrain!charnel!sifon!CC.UMontreal.CA!casgrain
- From: casgrain@ERE.UMontreal.CA (Casgrain Philippe)
- Subject: Follow-up to Statview Non-parametrics problems
- Message-ID: <1992Dec23.183544.7807@cc.umontreal.ca>
- Keywords: Mann-withney U, incorrect p-value
- Sender: news@cc.umontreal.ca (Administration de Cnews)
- Organization: Universite de Montreal
- Date: Wed, 23 Dec 1992 18:35:44 GMT
- Lines: 16
-
- Regarding my own post about some Statview possible bugs:
-
- The errors in StatView are not new to release 4.0, _but_, in the Mann-Whitney
- U test, the probability given (p-value) is that of the 'z' value, not of the
- 'U'. That can lead to problems if the sample size is small (e.g. < 30),
- because when it is 10 < sample size < 30, Legendre and Legendre (1983)
- suggest a correction to compensate for the assumption of continuity, which
- is not computed by Statview.
-
- Also, when the sample size is < 10, one should look in a "U-table" to determine
- the actual p-value, and _never_ take the p-value (from the z value) given
- by Statview. Sample size is just _too_ small to suppose continuity, and
- the probability should be computed with permutations of all possible cases.
-
- Philippe
-
-