Anonymous posted on Thursday, July 21, 2005 - 4:19 pm
I'm planning on using M+ for a Monte Carlo study with ordered categorical variables.
The model is basically a mixture factor model with ordinal indicators, similar to Example 7.26.
However, it's unclear to me under what conditions the Monte Carlo data would be generated under a probit versus logit model.
I've noticed in earlier posts that the WLS family of estimators assumes a probit model, and the ML family assumes a logit, but that wouldn't apply to the data generation phase of a Monte Carlo simulation, because nothing's being estimated at that point.
bmuthen posted on Thursday, July 21, 2005 - 4:39 pm
Type = mixture uses only logit, not probit. The rule for the current Mplus is "if you use ML then it is logit, if you use WLSMV it is probit". Mixture modeling only uses ML.
The data generation in Monte Carlo obeys these estimator-based rules in order for you to get model estimation that fits the data. That's not necessary in principle, but that's how the current version works.
A minor twist is that missing data is always generated via logit.