I'm doing a mediation analysis with longitudinal data where the predictor is binary (control/treatment) and one of the mediators is a variable (recall (ryes#)) participants were randomized to receive once during the study. We are looking at different lag patterns for various mediators and I'm having trouble getting the model to run since there is no correlation between the recall method at different weeks.
categorical=qa3 qa4 qa5 qa6 ryes3 ryes4 ryes5 ryes6; MISSING are all (-999);
ANALYSIS: estimator = WLSMV; !integration=montecarlo; parameterization=theta; coverage=0; !Bootstrap=1000; Model: !Measurement model (eliminated for space)
!start si week 3 si3 on group; elab3 on si3; norm3 on si3; ryes3 on si3; qa3 on elab3; qa3 on norm3; qa3 on ryes3; qa3 on group; qa3 on group elab3 norm3 ryes3;
!start si week 4 si4 on group; elab4 on si4; norm4 on si4; ryes4 on si4; qa4 on elab4; qa4 on norm4; qa4 on ryes4; qa4 on group; qa4 on group elab4 norm4 ryes4;
same for weeks 5 and 6 elim for space Errors that I get COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR RYES4 AND RYES3. THE MISSING DATA COVARIANCE COVERAGE FOR THIS PAIR IS ZERO.