Programming the statistical procedures from SAS

Testing Simple Slopes in Multilevel DAta

Reply
Occasional Contributor
Posts: 7

Testing Simple Slopes in Multilevel DAta

Hello everyone! Thanks so much for reading and helping if you have any ideas. Also, this might be a really obvious question, or it might be really hard. I have no idea!

 

We have a pretty simple paradigm where we showed people 49 slides, whose content ranged from very pleasant to very unpleasant. On each trial, people rated how they felt during the slide (level 1 predictor, person-centered), and then rated how willing they were to see the slide again (level 1 outcome). After the 49 trials, we had people rate their Neuroticism levels once (Level 2 moderator). All variables are continuous, and we ran a pretty simple initial MLM using proc mixed. It looks like this:


PROC Mixed NOITPRINT NOCLPRINT COVTEST MAXITER=250; CLASS Subject;

MODEL SeeAgain = Feeling_ZPC | ZB3_neuro

/ SOLUTION DDFM=BW COVB NOTEST;

RANDOM INTERCEPT Feeling_ZPC /SUB=Subject TYPE=UN; ODS EXCLUDE NOBS; RUN;

 

We found a cross-level interaction, and we made estimated means at high (+1 sd neuro) and low (-1sd neuro) levels of neuroticism to approximate the feeling x seeagain relationship at high and low levels of neuroticism. BUT, now we want to test these simple slopes.

 

I have code that can test the slope of the feeling x seeagain relationship at high and low neuroticism levels, and that looks like this:

 

proc mixed noclprint covtest method=reml; class subject;
model seeagain = Feeling_ZPC zb3_neuro Feeling_ZPC*Zb3_neuro
/ddfm=bw solution cl notest covb;
random intercept Feeling_ZPC /sub=subject type=un;
estimate 'simple intercept for personality' intercept 1 zb3_neuro -1/ cl;
estimate 'simple slope LOW personality'Feeling_ZPC 1 Feeling_ZPC*Zb3_neuro -1/ cl;
run;

proc mixed noclprint covtest method=reml; class subject;
model seeagain = Feeling_ZPC zb3_neuro Feeling_ZPC*Zb3_neuro
/ddfm=bw solution cl notest covb;
random intercept Feeling_ZPC /sub=subject type=un;
estimate 'simple intercept for personality' intercept 1 zb3_neuro 1/ cl;
estimate 'simple slope high personality'Feeling_ZPC 1 Feeling_ZPC*Zb3_neuro 1/ cl;
run;

 

However, we also need to somehow test the effect of neuroticsm on the feeling x action relationship at high and low feeling levels. Like did neuroticism matter more when the slides were really unpleasant than when they were pleasant. So, does anybody know how to model this?

 

Thanks for any thoughts!

Robbie at NDSU

 

Ask a Question
Discussion stats
  • 0 replies
  • 116 views
  • 0 likes
  • 1 in conversation