I am fitting a growth model to a binary outcome at three time points and have a number of time varying covriates (this is equivalent to a fixed effects x-sectional time series model in econometric parlance). the model converges and provides sensible estimates when I use MLR but when I use WLSMV, I get no parameter estimates. I receive no warning or error messages, there is just no sign of any model estimates. Please help!
I have now managed to get parameter estimates for both MLR and WLSMV. I now have a further question. These different estimators give very similar results but for a couple of predictors, the significance tests come out different, so in one model a predictor is significant at 95% while in the other it is not. for presentational purposes, then, the choice of estimator is crucial. should we have reason to prefer MLR over WlSMV or vice versa in terms of efficiency/consistency? Can you point me to some references on this? thanks again,
Both estimators are consistent. Maximum likelihood is asymptotically efficient. How they behave in practice can depend on the sample size, the particulars of the data, and the model. The only way to know which works best for your data and your model is to do a Monte Carlo simulation study. I don't think this topic has been studied much.
I have a type=complex growth model predicting a binary outcome. The Mplus default estimator is WLSMV. However, as this does not give an odds ratio, I also tried MLR. In this situation, Mplus requires montecarlo integration.
I understand from the post above that we don't know which estimator is preferable in which situation. Is this still the case, or have there been new studies on the topic?
And, if not, would it be 'okay' to choose MLR instead of the default to get odds ratio's? I would like odds ratio's as these are easilly interpreted. However, perhaps WLSMV also provides some output which is easy to interpret?