Hi, I am trying to test a mediation hypotheses in data from a randomized clinical trial with two intervention groups. Both the outcome (y) and the proposed mediator (m) were measured four times, therefore I am using a multilevel approach. I tried to transfer example F (http://www.statmodel.com/download/Preacher.pdf) onto this longitudinal setting. I conceptualized the Cross-Level-Interaction of the treatment variable ("cond", 0/1) and time in regard of the mediator as the "a" path in my mediation model. As the "b" path I conceptualized the mean of the random slope of y on m. However, this produces an error. Where is the error in my concept (or syntax)?
This is my input:
WITHIN = time; BETWEEN = cond; CLUSTER = id; STRATIFICATION = Center; ANALYSIS: TYPE = COMPLEX TWOLEVEL RANDOM; ALGORITHM=INTEGRATION; INTEGRATION = MONTECARLO; MODEL: %WITHIN% m y; sy|y on time; sm|m on time; sb|y on m; %BETWEEN% m y; sm on cond(a); sy on cond(c); sb WITH cond m y; [sb](b);
MODEL CONSTRAINT: NEW(ind); ind=a*b;
OUTPUT: TECH1 TECH8 CINTERVAL;
The error: THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-ZERO DERIVATIVE OF THE OBSERVED-DATA LOGLIKELIHOOD.