Linear Mixed Effects Regression

#1
Hello,

Please can anyone help me with the above topic. I'm using SAS and am unsure on how to program this. I think that a PROC MIXED is to be used, however, I am unsure about what output is of interest and how to interpret the results. :confused: Any help would be greatly appreciated.

Many thanks,
Lisa

:)
 

Dason

Ambassador to the humans
#2
... why don't you explain what you're trying to do. Mixed models is a very general subject. Proc mixed is one way to fit a mixed model (but not the only way). What output are you talking about? You really need to be more specific for us to be any help at all.
 
#3
Sorry for the brief post. I'm working on clinical trials data and have been asked to produce a linear mixed effects regression model on some data in which the parameter of interest is on a change in the pharmacokinetic measurement of the patient. Fixed effects are to be the treatment group, baseline pharmacokinetic measurement of the patient, random effect is the patient. I am using PROC MIXED in SAS as I researched on the internet and found a paper that advises to use this. I unsure of what options I should be putting into the PROC MIXED and wonder which parts of the output would be considered important and how to interpret it. I'm still experimenting. I've searched on the internet on this regression method and can find so little on it.:shakehead
 

Mean Joe

TS Contributor
#4
Well this may be a little late to help sasa81, but I just read this post.

Try syntax like:
proc mixed;
model y = treatment;
random patient;
lsmeans treatment;
run;

Try checking the SAS help, example 56.3. It's also available on the web: http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_mixed_sect035.htm

This example does some extra stuff with parms...but it explains the output in some detail.

I don't use PROC MIXED myself. But at a glance, it looks like "Estimated Covariance Parameters", the last columns of "(Solutions of) Mixed Model Equations", "Test of Fixed Effects", "Least Squares Means for Treatment Effect" have the good stuff; the other ones you can take a cursory glance, for any other info.

To get all these tables, you'll have to add the options to the proc statement:
proc mixed asycov mmeq mmeqsol covtest;
 

Link

Ninja say what!?!
#5
If the observations are individual patients, I would change the syntax from mean joe to say:
proc mixed;
class patient;
model y = treatment;
random int / type="type of correlation" sub=patient;
lsmeans treatment;
run;
 
#6
Thank you for your replies, I haven't been logging on to this site as I'm so busy working on other projects. I am back on this project so I will test them as suggested. Thanks again!