Date: Sat, 18 Mar 2006 07:38:01 -0800
Reply-To: Dale McLerran <stringplayer_2@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <stringplayer_2@YAHOO.COM>
Subject: Re: Proc Mixed (of course)
Content-Type: text/plain; charset=iso-8859-1
--- Wendy Buzy <wbuzy@PSYCH.UMARYLAND.EDU> wrote:
> I apologize for sending this question to the list, but I've read many
> the previous posts regarding Proc Mixed and my poor, pea-sized brain
> still struggling. I'm trained as a clinical psychologist and my
> statistical knowledge is somewhat limited, although I'm trying my
> best to
> I'm trying to analyze data from a task that subjects performed 6
> times --
> i.e., at 3 different levels of difficulty (Level) and at 2 different
> interstimulus intervals (ISI) within each difficulty level. Thus,
> subject performed the task at:
> * Level=1, ISI=3200
> * Level=1, ISI=2800
> * Level=2, ISI=3200
> * Level=2, ISI=2800
> * Level=3, ISI=3200
> * Level=3, ISI=2800
> in that order, on the same day. Some subjects have missing data,
> which is
> why we're trying to use Proc Mixed rather than Proc GLM.
> There are two groups of subjects (ADHD vs. NC).
> So far, we've struggled with when/whether to use the RANDOM statement
> versus the REPEATED statement, and also with which covariance
> to use. We've come up with the following code:
> proc sort data=usethis; by subject level descending isi; run;
> proc mixed data=usethis order=DATA;
> class group level isi subject;
> model accuracy = group | level | isi;
> repeated level isi / type=un@cs subject=subject;
> My questions are as follows:
> 1. Are we using the REPEATED statement correctly and/or should we be
> the RANDOM statement (instead)?
My guess is that you probably want both a RANDOM statement as
well as the REPEATED statement. The question is whether the mean
of the six responses varies from one individual to the next.
If there are subject-specific means, then you need the RANDOM
statement. But you may still need a REPEATED statement.
> 2. It seems that because there are only two levels of ISI, it doesn't
> matter whether we specify "type=un@cs" or "type=un@ar(1)", etc. Is
> that correct?
That is correct. The two covariance structures are identical
given that you have only 2 levels of ISI.
> 3. Is it appropriate to be using an unstructured covariance
An unstructured covariance structure is never inappropriate, but
it could be inefficient.
> 4. I would like to use intelligence (IQ) as a covariate. Is it
> correct to
> simply add IQ to the model statement (but not the class statement)?
Probably. I presume that the IQ measure is continuous. However,
it is not necessarily the case that the relationship between
the response and IQ is linear. If the relationship between the
two is nonlinear, then naming IQ as a CLASS variable CAN be
useful. Of course, another way to handle nonlinear relationships
is to use polynomial terms.
Fred Hutchinson Cancer Research Center
Ph: (206) 667-2926
Fax: (206) 667-5977
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around