Programming the statistical procedures from SAS

Getting marginal means that are way off the actual mean (RMPL vs RSPL)

Regular Learner
Posts: 1

Getting marginal means that are way off the actual mean (RMPL vs RSPL)



So I am working on an EMA dataset with different numbers of reports... in those reports participants report exposure to enviornmental variables on a yes no basis.   So I am using Proc GLIMMIX (SAS 9.4) to Model If class variables are associated with increases or decreases in the liklihood of exposure. Code is as follows:


proc glimmix maxopt=200 lognote;

class sub Var1 ;

model Exposure(event="1") = Var1 / dist=binary link=logit ddfm=bw solution oddsratio;

random intercept /sub=sub type=un g solution cl;

random _residual_ / type=ar(1) subject=sub;

lsmeans Var1 /  ilink cl;



Problem is that when modeled with RSPL, the least square means are all considerably lower than the actual means... this is magnified the further from 50/50 the exposure gets.  In other words if an exposure occures 10%-15% across the levels of Var1 in the sample.  I am getting Least square means in the 3-8% range. Specifiying Method=RMPL gets me aroud this problem and to means that are similar to the actual means.  But I am fairly confident that I want to specify "S"ubject-specific rather than "M"arginal expansions.  So the question is what could be causing this and is there away around it?

Ask a Question
Discussion stats
  • 0 replies
  • 1 in conversation