Date: Tue, 19 May 1998 14:43:35 -0400
Reply-To: "Zuckier, Gerald" <Zuckier@CHIME.ORG>
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From: "Zuckier, Gerald" <Zuckier@CHIME.ORG>
Subject: Re: geometric mean
Content-Type: text/plain
Hmm. The light dawns. Geometric mean is equivalent to log transform of
data, then taking mean (and transforming back). Works on that typical
Poisson looking data, I guess. Median would seem to be more general in
application.
> ----------
> From: Bassett Consulting Services,
> Inc.[SMTP:BASSETT.CONSULTING/0002395748@MCIMAIL.COM]
> Sent: Tuesday, May 19, 1998 1:31 PM
> To: SAS-L@UGA.CC.UGA.EDU
> Subject: Re: geometric mean
>
> ----------------------------------------------------------------------
> -
> CONTENT: re: geometric mean
> NAME: Michael L. Davis
> INTERNET: Bassett.Consulting@worldnet.att.net
> AFFILIATION: Bassett Consulting Services, Inc.
> P-ADDR: 10 Pleasant Drive, North Haven, CT 06473
> PHONE: (203) 562-0640
> FAX: (203) 498-1414
> ----------------------------------------------------------------------
> -
>
> I found Amy Savage's question and Lary Jones's reply about the
> geometric mean interesting because I had been told that the
> geometric mean was synonymous with the median. A quick call
> to SI Tech Support and examination of the SI sample code to
> compute the geometric mean quickly indicated the error of my
> understanding. I am glad that I read SAS-L regularly to help
> identify where some of my previous learning may be in error.
>
> Neverless, if the goal of using the geometric mean is to
> minimize the skew effect when analyzing small, asymetrical
> samples, then it appears that one might consider using the
> median instead of the geometric mean as a tool to estimate
> the center of the population distribution. Both computations
> minimize the effects of extreme values
>
> One advantage of using the median is that it is more easily
> grasped by mathematically challenged (such as myself). Also
> the median computation can be easily obtained from a base SAS
> procedure such as PROC UNIVARIATE or PROC CORR.
>
> Now all we need is a spirited discussion of which PCTLDEF=
> option to use <grin>.
>
| 
