home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!uwm.edu!psuvax1!psuvm!auvm!SAS.UPENN.EDU!ECOOCH
- Return-Path: <ecooch@sas.upenn.edu>
- Posted-Date: Thu, 17 Dec 92 10:14:26 EST
- X-Mailer: ELM [version 2.3 PL11-upenn1.8]
- Message-ID: <9212171514.AA05740@mail.sas.upenn.edu>
- Newsgroups: bit.listserv.stat-l
- Date: Thu, 17 Dec 1992 10:14:26 EST
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: Evan Cooch <ecooch@SAS.UPENN.EDU>
- Subject: diff of means
- Lines: 34
-
- Many thanks to those who quickly responded to my late-night (hence
- poorly worded) request for a reminder on estimating the error variance
- of the difference between two sample means.
-
- Obviously, the expectation of the differences between two sample means
- (mu1 and mu2) is simly (mu1 - mu2). The variance of this difference is
- given by
-
- Var(mu1 - mu2) = v(mu1) - 2cov(1,2) + v(mu2)
-
- If the samples are truly independent, then cov(1,2) = 0, and the
- preceding reduces to
-
- Var(mu1 - mu2) = v(mu1) + v(mu2)
-
- Of course, v(mu1) = sample v / n ( = SE)
-
- thus, Var(mu1 - mu2) = SE(1) + SE(2)
-
- This, of course, is related to the denominator of the t-test.
-
- (((n1 - 1)v(1) + (n2 - 1)v2)/(n1+n2-2))((n1+n2)/n1n2)
-
-
- Its amazing the little things you forget (a) when its late and you're
- tired, and (b) you've spent much more time worrying about the
- significance of differences, rather than details about the magnitude of
- the difference.
-
- Thanks again...
-
-
-
- Evan Cooch
-