i am trying to estimate a large model with both observed and latent time-invariant predictors effecting the intercept factor of a multiple indicator quadratic growth curve (as in, each variable in the time series is a latent). when i introduce an interaction between a latent and observed predictor using the "TYPE=RANDOM; x | y XWITH z;" syntax, the model won't run and i get an error message stating, erroneously, that all variables have no non-missing values.
i have built the model up systematically so i know the problem is with the interaction. troubleshooting, i tried specifying ALGORITHM = INTEGRATION, but this did not resolve the problem. also, perhaps relevant, i am using the MODEL=NOCOVARIANCE function.