I am fitting a cross-lagged path model with 2 binary observed dependent variables at 3 timepoints. I was planning to use the ML estimator for logit models by specifying numerical integration. It seems, however, that with ML, categorical dependent observed variables measured at the same time point cannot be correlated using the 'WITH' command (but they can using WLS estimator). We need to correlate our two dependent variables at each timepoint. We have a fair amount of missing data and are hoping to avoid multiple imputation (which we would do if using WLS to deal with missingness). Is there any way to capture the correlations/covariances between concurrently measured observed binary dependent variables in a cross-lagged model using ML?