I am comparing three models comprised of 2 exogenous latent variables, 2 manifest, dichotomous mediators and a 4-category nominal outcome. The three comparisons include the addition of a manifest, dichotomous variable: 1st model, as a covariate; 2nd model as an intermediate step between the two latent variables and the two mediators; 3rd model as a moderator of the model relationships. Since I have to use an integration algorithm to estimate these models, I cannot do traditional multi-group modeling for the third model; instead, I am doing a mixture model. My question is this: Is it appropriate to compare the AIC/BIC from the three models (with the first two including the variable as manifest, not knownclass), or should I do all three models as mixture models for comparability? Thank you in advance for any guidance.