home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Subject: (no subject given)
- Path: sparky!uunet!paladin.american.edu!auvm!LAKE.SCAR.UTORONTO.CA!LTHOMAS
- Message-ID: <9211240310.AA11742@lake.scar.utoronto.ca>
- Newsgroups: bit.listserv.stat-l
- Date: Mon, 23 Nov 1992 22:10:14 -0500
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: Len Thomas <lthomas@LAKE.SCAR.UTORONTO.CA>
- Lines: 21
-
- subject: Bootstrap estimates of power
-
- I have been asked to compute a power/effect size curve for an experiment
- which used the Wilcoxon's signed-ranks test. Since the underlying distribution
- from which the samples were drawn is unknown I propose to bootstrap the
- samples to estimate power. Different effect sizes can be simulated by
- adding 1 to the value of each sample in one group with a probability set by
- an effect size variable, which will range from 0 to 1.
-
- Beeing rather new at this game, I'd appreciate any comments you folks may have.
- Are there any published references to bootstrap estimation of power?
- I feel fine simulating different effect sizes in this manner when there are
- only two groups, but how could I do it if there were 3 or more (i.e. for a
- Kruskal-Wallis test)?
-
- Please direct any thoughts to my internet and I'll summarise for the list.
-
- Len Thomas, lthomas@lake.scar.utoronto.ca
- Division of Life Sciences,
- Univerisity of Toronto at Scarborough.
- Canada.
-