home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.stat-l
- Path: sparky!uunet!uvaarpa!darwin.sura.net!newsserver.jvnc.net!gmd.de!Germany.EU.net!mcsun!sunic!ugle.unit.no!news
- From: kyrre.svarva@avh.unit.no (Kyrre Svarva)
- Subject: Re: Violated t-test assumptions: how to adjust?
- In-Reply-To: ph@physiology.oxford.ac.uk's message of 25 Jan 93 19:55:06 GMT
- Message-ID: <1993Jan26.202805.20915@ugle.unit.no>
- Sender: news@ugle.unit.no (NetNews Administrator)
- Organization: University of Trondheim, Norway
- References: <1993Jan25.195506.18216@motor.physiol.ox.ac.uk>
- Date: Tue, 26 Jan 93 20:28:05 GMT
- X-News-Reader: VMS NEWS 1.11b
- Lines: 27
-
- In <1993Jan25.195506.18216@motor.physiol.ox.ac.uk> ph@physiology.oxford.ac.uk writes:
-
- > In his classic 1960 paper, Boneau discussed the effects on the t test
- > and related tests of violating the parametric assumptions. From the
- > inferential point of view, I remember the most drastic problem is the
- > situation of unequal sample sizes, where the smaller sample has the
- > larger variance.
-
- As far as I know there is a special version of the t-test that does not
- depend on homogeneity of variance. In this test, the SE estimate (the
- denominator in the t formula) is computed in a different manner so as to
- adjust the value of t downwards in the situation you describe. The df's are
- also adjusted. This approach is described in Robert Johnson: Elementary
- Statistics, chapter 10, and is available in SPSS ("Separate variance estimate
- t-test") although there a different (better?) way of assessing the number
- of df's is used. There also is something called the "Welch t-test",
- which I believe is a different name for the same thing.
-
- As to your particular situation: Although the t-test takes sample size into
- account, I would hesitate to draw conclusions with only 5 observations in one
- sample.... with this sample size, one non-representative sample member could
- have a substantial effect on the mean as well as the variance of the sample.
-
- - - What do you REAL statisticians out there think of these matters??
-
- Kyrre Svarva
-
-