home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!stanford.edu!bcm!convex!darwin.sura.net!paladin.american.edu!auvm!HILBERT.MATHS.UTAS.EDU.AU!MCPHERS
- X-Mailer: ELM [version 2.2 PL11]
- Message-ID: <9211162144.AA08550@hilbert.maths.utas.edu.au>
- Newsgroups: bit.listserv.stat-l
- Date: Tue, 17 Nov 1992 08:44:58 EST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: "D. Glen McPherson" <mcphers@HILBERT.MATHS.UTAS.EDU.AU>
- Subject: Re: guidelines for checking assumptions
- Lines: 34
-
- The thought that statisticians could come to a consensus about the choice and
- application of a set of formal test criteria to check assumptions in any area
- of statistical application is idealism in the extreme. However, this should
- not preclude us from offering quite specific guidelines to nonstatisticians
- who have little experience in how to proceed. The fact that two statisticians
- or two statistical reference books recommend different tests or vary in the
- manner of interpreting findings is lesser of an evil than a complete misuse
- of statistics by someone who fails to check a critical assumption.
-
- If we are to offer guidelines, there are two rules which I suggest must be
- obeyed. The first is to make the guidelines exhaustive. Nothing will turn a
- user off faster than to reach a point where the guidelines stop giving
- guidance. For example, if the guidelines suggest a test for equality of variance
- and there is evidence of heterogeneity of variance, there must be advice on
- what to do next. This may be specific, may be a reference to another
- publicationor may be the advice 'At this stage seek advice from a consultant
- statistician'.
-
- The second rule is to advise users to follow the usual convention in their own
- discipline, if one exists, unless there is clear evidence that such a convention
- is unreasonable. Publication of research findings is difficult enough without
- introducing methods that are unfamiliar to referees. Where a new method for
- checking assumptions is being introduced, it is generally wise for statistical
- consultants to play a part in arguing the case for the new method. Perhaps they
- may even offer to write the relevant section in the paper.
-
- Regards, Glen.
- --
- ---------------------------------------------------------------------------
- | Glen McPherson | | _--_|\ |
- | Department of Mathematics |-----------------------------------| / \ |
- | University of Tasmania | E-Mail: | \_.--._/ |
- | Australia. | mcphers@hilbert.maths.utas.edu.au | * |
- ---------------------------------------------------------------------------
-