Date: Fri, 4 May 2007 15:44:17 -0400
Reply-To: Peter Flom <firstname.lastname@example.org>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject: Re: interpreting PROC GLM output
Content-Type: text/plain; charset=UTF-8
Jerry Davis <jwd@UGA.EDU> wrote
>Suppose an investigator does a 3**3 manufacturing type experiment and
>wants to fit a response surface model. In practice they will often do
>an initial stepwise procedure to see what terms affect the response
>variable. In the interest of understanding and parsimony they want to
>drop terms that have no effect. Sometimes the resulting model is one
>that I would have selected. The parameter estimates, standard errors
>and p-values are what they are. They don't care what 10,000 simulations
>of a theoretical problem produce.
I don't know enough about this to comment intelligently
>I'm no expert on data mining but apparently data miners love automatic
>selection methods. Why do you think SAS developed GLMSELECT? Customers
>wanted variable selection methods with a class statement. I guess the
>marketplace will decide who is stupid and who is being practical.
SAS developed GLMSELECT in part because it's much better than STEPWISE.
The marketplace, though, will never determine who is stupid. Just who is rich.