home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!ogicse!plains!news.u.washington.edu!martin
- From: martin@biostat.washington.edu (Don Martin)
- Newsgroups: sci.math.stat
- Subject: Re: Standard Deviation.
- Summary: Fun with n+1,n,n-1,n-2
- Keywords: (n) versus (n-1)
- Message-ID: <1992Aug17.212702.21327@u.washington.edu>
- Date: 17 Aug 92 21:27:02 GMT
- Article-I.D.: u.1992Aug17.212702.21327
- References: <1992Aug14.172833.11844@cbfsb.cb.att.com> <c48nbgtf@csv.warwick.ac.uk> <dmurdoch.43.714062116@mast.queensu.ca>
- Sender: news@u.washington.edu (USENET News System)
- Organization: Biostatistics, School of Public Health, Univ. of Wash.
- Lines: 21
-
- This is off the top of my head but I think that it is
- reasonably accurate.
-
- Personal opinion: There are a lot of bad stat books out there.
-
- Main topic: Why divide by n-1 instead of n etc. Here are
- a few proposals:
-
- n+1 This has a smaller mean square error for s.
- n ML estimator
- n-1 unbiased variance but not standard deviation
- n-2 2 works better for 95% confidence intervals.
-
-
- In fact, the varince/standard deviation estimates tend to
- be pretty poor, depending on the distribution, for the
- range where the divisor makes a difference.
-
- Does anyone have an argument for n+2 or n-3 ?
-
- Don Martin
-