I have a two-level growth model for a continuous outcome (three level-analysis), similar to the one in example 9.12. I want to explain between-cluster variance in the time slope with two cluster-level covariates w1 and w2 that are correlated. I tried to model this correlation by specifying w1 WITH w2 in the %BETWEEN% section of the model. However, the corresponding path is negative and significant, whereas a separate simple bivariate correlation between w1 and w2 (with one measure per cluster) is strongly positive and highly significant. Is it possible that using WITH In the %BETWEEN% part of the model produces not a regular bivariate correlation between w1 and w2 but something else? This is the model:
Linda, thanks for your quick reply. Using the SAMPSTAT option showed me that apparently, something had gone wrong when I exported the data set from R to Mplus, because in the sample statistics output the two variables also had a negative bivariate correlation on the between level. I fixed the error in R and now everything works as expected. Apologies for the mistake on my side.