Date:         Wed, 6 Oct 1999 13:41:49 GMT
Reply-To:     jpwinter@UMSL.EDU
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         jpwinter@UMSL.EDU
Organization: University of Missouri - St. Louis
Subject:      Re: PROC MIXED 2

On 5 Oct 99 17:40:17 GMT, sarthur67@YAHOO.COM (Stephen Arthur) wrote:

: If I have one continuous outcome variable and 15 to 50
: discrete independent variables (each of which can be
: at four levels) and "several" other continuous
: variables.
:
: What I want to do is find the simpliest model for
: these predictor variables.  I am thinking I would need
: to use PROC GLM (which I have used) or PROC MIXED
: (which I have yet to use).

The simplest "model" of the outcome variable is it's mean.  The next
simplest model would include one variable in a regression.  No model
with 15 variables is simple.  Does this answer your question?

Also, I don't see how PROC MIXED is appropriate for your data, one
should always look for more basic answers before going to the more
complex routines.

: Please don't flame me too hard on this one, I am just
: thinking (maybe not enough) about testing some ideas.
:
: Any thoughts?  In general, can PC-SAS handle using
: these statistical procedures with 60+ variables and
: still remain functional?

SAS' computational limits aren't the issue.  Regression and model
building are not black-box computer wizardry.  You have to understand
something about statistics in general and regression in particular
before proceding.

As people often point out, your questions are similar to posting the
following in a medical newsgroup:

"I have really strange lump on my body.  I want to remove the lump.
At my place of employment we have really sharp knives.  Can anyone
provide me some insight into removing the lump?"

:>