home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!CORNELLC.BITNET!TCD
- Message-ID: <SAS-L%92072211352883@UGA.CC.UGA.EDU>
- Newsgroups: bit.listserv.sas-l
- Date: Wed, 22 Jul 1992 11:13:04 EDT
- Reply-To: Tim Dorcey <TCD@CORNELLC.BITNET>
- Sender: "SAS(r) Discussion" <SAS-L@UGA.BITNET>
- From: Tim Dorcey <TCD@CORNELLC.BITNET>
- Subject: Re: question
- In-Reply-To: Message of Sat,
- 18 Jul 1992 12:29:08 -0500 from <rottgeri@REUSEL.LEIDENUNIV.NL>
- Lines: 35
-
- On Sat, 18 Jul 1992 12:29:08 -0500 Huub Rottgering said:
- >Dear Sas Community,
- >
- >
- >Does any one have a simple straight forward solution for
- >using SAS for the following problem:
- >
- >I have a data set with N observation X_i and Y_i
- >and variances for Y: sigma_y_i. I would like to fit
- >a least square model to this data set that takes into
- >account the known variances sigma_y_i.
-
- The WEIGHT option in PROC REG gives the correct estimates, standard
- errors and hypothesis tests under the assumption that the variance of
- errors is known up to a scalar multiple. That is COV(y) = D*sigma
- where D is a known diagonal matrix and sigma is an unknown variance
- parameter. If the variances themselves are actually known, then using
- the reciprocal of sigma_y_i as the WEIGHT var will yield the correct
- estimates, but the standard errors and hypothesis tests would need to
- be adjusted in the following way:
- 1) divide the output standard error values by the square root of the
- output MSE residual term.
- 2) multiply the output t-values by the square root of the output MSE
- residual term and then look them up in a Normal (rather than t) table.
- 3) ignore the output F-test and instead look up the Model SS in a
- Chi-square table with Model DF degrees of freedom.
-
- I can't imagine many (any?) situations where you would actually want to
- do this.
-
- Tim Dorcey BITNET: TCD@CORNELLC
- Statistical Software Consultant Internet: TCD@CORNELLC.CIT.CORNELL.EDU
- Cornell Information Technologies Phone: (607) 255-5715
- Cornell University
- Ithaca, NY 14853
-