Date:         Thu, 4 May 2000 08:27:43 -0400
Reply-To:     Paige Miller <paige.miller@KODAK.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Paige Miller <paige.miller@KODAK.COM>
Organization: Eastman Kodak Company
Subject:      Re: GLM question - trend and contrast statement
Content-Type: text/plain; charset=us-ascii

"Meldrum, Sean" wrote:

> >I'm doing an ANCOVA using Proc GLM.  I want to do a trend
> >analysis on my IV.  It has four categories.  I'd like to
> >test for a linear trend.  The variable is named att_style.
> >The four categories are Anxious, Avoidant, Secure, & Traumatic.
> >
> >The SAS code I've been using is
> >
> >proc glm ;
> >   class att_style ;
> >   model ABSORPTN_i = att_style  bslinel;
> >   means att_style ;
> >   means att_style / tukey ;
> >   run;
> >
> >where absorptn_i is the dependent variable
> >bslinel is the covariate
> >the means statements examine multiple comparisons, but I
> >want to examine for a linear trend.  I think the rank
> >ordering of the categories is in this order:
> >Traumatic Avoidant Secure Anxious.
> >
> >Would I use a contrast statement?  Do you know how to
> >specify the effects for a linear trend?

Before you go ahead and test for a linear trend, are you absolutely sure
you have the right ordering of categories? I'm no expert in your field,
whatever that field may be, but I have a hard time seeing how "secure"
can fit linearly in between "avoidant" and "anxious". Secondly, even if
that is the correct ordering, are you sure that the categories are
approximately equally and linearly spaced? Again, I lack subject matter
expertise, but "traumatic" seems to be far separated from the others in
my limited understanding.

So unless you can clearly state that the categories are properly ordered
and approximately linearly separated, I wouldn't even try the test you
are proposing.

--
Paige Miller
Eastman Kodak Company
paige.miller@kodak.com
"It's nothing until I call it!" -- Bill Klem, NL Umpire