home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!SARNOFF.COM!MCZ
- Message-ID: <9211232038.AA00648@grumpy>
- Newsgroups: bit.listserv.stat-l
- Date: Mon, 23 Nov 1992 15:38:55 EST
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: Martin Czigler <mcz@SARNOFF.COM>
- Subject: Proportion of total variance with correlated variates
- Lines: 22
-
- I'm interested in finding the proportion of total variance due to
- specific random factors. For example, suppose X1 and X2 are normally
- distributed with 0 mean, and Y=X1+X2. If X1 and X2 are uncorrelated,
- then the proportion of Y's variance due to X1 is given by
- sigma1^2/(sigma1^2+sigma2^2). However, when the factors are
- correlated, the appropriate answer is less clear. Suppose X1 and X2
- have the same variance, and are correlated with a coefficient of -1.
- Then Y will be constant, i.e. var[Y] = 0, while both X1 and have a
- positive variance. Is it meaningful to ask what proportion of Y's
- variability is due to X1 and X2 individually?
-
- Thanks,
-
- --Martin
-
-
- Martin Czigler
- David Sarnoff Research Center
- CN 5300
- Princeton, NJ 08543
- mcz@windo.sarnoff.com
- Martin_Czigler@maca.sarnoff.com
-