Date:         Wed, 28 Jan 2009 13:01:58 +0100
Reply-To:     Thomas Fröjd <thomas.frojd@NEURO.UU.SE>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Thomas Fröjd <thomas.frojd@NEURO.UU.SE>
Subject:      Proc Mixed,
              Confidence intervals for combinations of estimates in a mixed
              spline model?
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

Hi,

I am currently analyzing the response on a psychometric scale from a
series of questionnaires. The (relatively few, N=30) repondants have
answered the questions on different (but not evenly spaced) points in
time during many years. I am interested in if there is any diffrence in
how the score change in time between men and women.

For both sexes there is a sharp drop in the score between the first two
measures and then a much smaller slope for the rest of the measures. My
plan is therefore to use a mixed spline model with a knot on timepoint 2.

The code i come up with look like this. "t" below is the same variable
as time only designated as class variable, this is recommended as good
practice by Fitzmaurice in his book on longitudinal studies. Also
time_1= time - 1.

proc mixed data=nyt ;
class  ID GENDER t;
model score=time time_1 GENDER GENDER*time GENDER*time_1 /s chisq;
repeated t  / type=UN subject=ID R RCORR;
run;

My first question is  if this model seems reasonable?

Secondly I am wondering how I can calculate standard errors or
confidence intervals for combinations of estimates. Say for example i
want to know the estimate of a change in time that is not the first
timepoint for men. That would be the sum of the estimate for time and
time_1 so if the estimate of time = 14 and time_1 = -10 it will be 4.
But how do I caluculate the standard deviation or confidence interval
for this estimate? How does it differ if i need som sum up more estimates?

I have a feeling that I should be able to do this with some statement in
proc mixed.

Thanks for you answers in advance.

Best regards.