I am running a multilevel analysis and have a question about whether to identify certain individual-level variables as “within.” In my dataset individual students are nested within 37 schools. I have several school-level variables (e.g., NAG=school norms for aggression). I have several individual-level variables measured at wave 1 including students’ frequency of aggressive behavior (AGG1) and perceptions of their parents’ attitudes toward aggression (PA1). My dependent variable is individual students’ frequency of aggression at wave 2 (AGG2). I have additional variables, and some more complex models involving random slopes, but this is a simplified version of my basic model.
I use syntax like the following:
Cluster is school; Between= NAG; Analysis: Type=twolevel missing H1; Model: %BETWEEN% AGG2 on NAG; %Within% AGG2 on AGG1 PA1;
My question is whether I should include the following statement which designates my exogeneous individual-level variables as within.
Within = AGG1 PA1;
When I include this statment I get a chi-square with zero dfs. If I do not include this statement I get higher ICCs for the dependent variables, and what seem to be more interpretable findings.
From reading other Q & As, my understanding is that this decision should be based on whether I expect between variation. Does this mean that if I expect non-zero ICCs for my individual level exogenous variables, and variability in these variables across schools I should not list them as within? If I do not list them as within how does that effect my interpretation?
I am uncertain how to proceed and any guidance you can give me would be greatly appreciated?
I would not put these 2 variables on the Within = list because they may have sizeable between variation. Email us and we will send a note on the difference these modeling choices make. It has to do with how to form a cluster-level version of these 2 variables.