[19191] in s-news-athena
Re: [S] Errors in Predictors
daemon@ATHENA.MIT.EDU (Viswanath Devanarayan)
Sat Sep 11 16:19:28 1999
Date: Sat, 11 Sep 1999 15:14:44 -0500
From: Viswanath Devanarayan <DEVANARAYAN_VISWANATH@lilly.com>
To: Tom Burr <tburr@nis.lanl.gov>
Cc: Viswanath Devanarayan <DEVANARAYAN_VISWANATH@lilly.com>,
S-news@wubios.wustl.edu
Message-Id: <052567E9.006F344A.00@aammta1.d51.lilly.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
For the linear errors-in-variables model you have described below,
it is very easy to implement in Splus in a few lines the closed form
expression for an estimate of the regression parameters that is given
in Fuller's book on Measurement Error Modes (chapter 1). The formula
is just a minor modification of the well know formula for computing
the OLS estimate in the absence of measurement error.
There are different ways to correct for the measurement error bias
depending on the type of additional information you have to adjust for
this bias. See chapter 1 of Fuller's book for more details.
For generalized linear and nonlinear models, the book by
Carroll, Ruppert and Stefanski is an excellent reference. Algorithms for
implementing the Regression-Calibration and SIMEX methods are
very clearly outlined in chapters 3 and 4 of their book. I have written
programs in Gauss to implement these methods, but it can be done
quite easily in Splus as well. Ray Carroll and his colleagues have
written a suite of Splus functions called QVF (I think it stands for
quasi-likelihood and variance-function) and have made it available
for public in their Texas A&M web-site.
Also, I heard that either the most recent version or an upcoming version
of STATA might contain some built-in functions for implementing the
SIMEX methods.
Tom Burr <tburr@nis.lanl.gov> on 09/11/99 02:25:01 PM
To: S-news@wubios.wustl.edu
cc:
Subject: [S] Errors in Predictors
A followup to my recent post:
I have not located any S+ functions to handle "errors in variables."
Regards,
Tom Burr
+++++++++++++++++++++++++++++++
Detail:
A few weeks ago, I inquired whether anyone knew of any
S+ functions to handle "errors in variables" (example: Y = Xb + e,
with nonnegligible error in X, and interest is on estimate of
b, rendering ordinary least squares or minor variations thereof
suboptimal, and thereby leading us to Carroll, Fuller, and other's work).
There were about 10 replies direct to me asking for
info if I got any, so there is modest potential customer
base.
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
