[prev in list] [next in list] [prev in thread] [next in thread] 

List:       sas-l
Subject:    Re: STEPWISE REGRESSION WITH UNEQUAL OBSERVATIONS
From:       "Paul A. Thompson" <pat () PO ! CWRU ! EDU>
Date:       1997-11-28 17:15:41
[Download RAW message or body]

In a previous article, MARINI%VTVM1.BITNET@VTBIT.CC.VT.EDU (MICHELE MARINI)
says:

>Good Morning SAS-L'ers!  Such a lovely winter morning here in Southwest
>Virginia!
>
>My problem this morning is this - I need an appropriate stepwise regression
>technique for when I have unequal numbers of observations, say X, Y, and
>Z where maybe some of my Z values are missing - and I do not want to delete
>those observations in their entirety from the data set.  I know that there
>is a macro in IML for doing regression on subsets of variables, but am not
>sure that it is appropriate as in this example (happens to be example 6 in
>the SAS/IML Software Usage and Reference manual) all of the data are complete.
>Any help would be most appreciated.  Thanks in advance.

There are a couple things you can try.  Since what you are trying to
do is questionable, none of these techniques will work properly in many
datasets.
1) Impute values for the missing values.
2) Work with the covariance/correlation matrix.  It is almost sure not
to be Gaussian, however.

In this circumstance, the very questionable aspects of stepwise regression
are questionable ^2.  It is very easy to imagine a situation in which
in subset 1 variable X is the best predictor and in subset 2 variable Y is
the best predictor.

Thus, I would not do these things.  I would bite the bullet and select a
subset which is consistent in missing values.  Any other approach is
almost certain to fall apart with a possible dataset.
>
>*************************************************************************
>                           Michele Marini
>                           Statistics Department
>                           Virginia Tech
>                           (540) 231-8315
>                           email:marini@vtvm1.cc.vt.edu
>*************************************************************************
>
--
Paul Thompson, Ph.D.     |  This is a coded .sig file.  Stare hard at
Department of Psychiatry |  the following terms, and tell me what you see:
Case Western Reserve Univ|  KJA:HJSDoJL*H**&o*&*S&kjalkJnUKJtL**a&&&r*&*.
Cleveland, OH  44106     |  JHJHBJSHeDHJGRHHoWUYoUYUdYWJNHJHoMNMwXJJ!YJY!

[prev in list] [next in list] [prev in thread] [next in thread]