-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On
Behalf Of wuzzy
Sent: Saturday, March 30, 2002 9:47 AM
To: [EMAIL PROTECTED]
Subject: prediction: underspecifiation in observational studies


Maybe someone will point me to other newsgroups or mail groups on
biological or clinical statistics as I know that sci.stat.edu is about
the education of statistics not really about stats itself..

My question (frustration, rather) is: how do  you deal with the fact
that signs on coefficients of multivariable models change direction
and size when you remove a predictor of the dependant variable(s).

is there a test for this?
-----------------------------------------------------------------
Multivariate regression handles only one error effect, and that is assigned
to the right hand side of the equation. This is really two errors combined
(assuming each of the X variables are precise, error free measurements), one
for the error in the Y measurement, and two for the fact that left-out
variables affect Y, but not being measured, are included.

If I leave out one variable, all the other variables have to "pick up the
load", and appropriately, the coefficient values change.

One way is to regress each variable, one at a time, starting with the
variable that has the greatest effect (R squared). Regress now on the
residuals from the first regression using the next variable. Each measure is
orthogonal. Continue for all variables and the effect of each variable can
be determined from the R values, the value of the coefficient, and the
standard error of the coefficient.

One possible test is Fisher's R squared tests, in which his transformation
gives approximate z values. The sequence of the corresponding cumulative P
values provides a measure.

DAHeiser

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to