home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.stat-l
- Path: sparky!uunet!comp.vuw.ac.nz!cc-server4.massey.ac.nz!H.Edwards
- From: H.Edwards@massey.ac.nz (H. Edwards)
- Subject: Re: ANOVA & Kruskal-Wallis details
- Message-ID: <1992Nov6.012208.17255@massey.ac.nz>
- Keywords: ordered alternatives, multiple comparisons, testing vs. estimation
- Organization: Massey University, Palmerston North, New Zealand
- References: <9211031450.AA11068@hsr>
- X-Reader: NETNEWS/PC Version 2c
- Date: Fri, 6 Nov 92 01:22:08 GMT
- Lines: 26
-
- Re: ordered alternatives mu1 <= mu2 <= .. <= muk.
-
- You may also want to look at: Hayter, AJ (1990) "A one-sided studentised
- range test for testing against a simple ordered alternative", J.Amer.Stat.
- Assoc. 85, 778-785, which shows how the test can be inverted to give
- simultaneous one-sided confidence intervals for pairwise differences
- mu(i) - mu(j), i < j. This allows one to make useful inferences about
- the sizes of ordered differences.
-
- Another useful reference if you are only interested in estimating differences
- (as opposed to testing for them) is Bofinger, E (1985), "Multiple comparisons
- and Type III errors", J.Amer.Stat. Assoc. 80, 433-437, replacing Bofinger's
- approximate o statistic with Hayter's more accurate h statistic. A Type III
- error occurs when a test concludes that mu(i) > mu(j) (for some i,j different)
- when in fact mu(i) < mu(j). Personally I don't see much point in protecting
- P(Type I error|H0) if you know that the equality hypothesis is untrue anyway,
- but that's another chapter in the long-running multiple comparisons saga .
-
- --
-
-
- Howard Edwards Ph: (64)(6)3569099 ext 8331
- Dept of Statistics Fax: (64)(6)3505611
- Massey University email: H.Edwards@massey.ac.nz
- Palmerston North
- NEW ZEALAND
-