shihyingyao posted on Wednesday, March 07, 2012 - 11:36 am
Hi, I need to estimate different types of constrained latent class models using Bayesian estimator as below:
(a) Ordered latent class model that is equivalent to the Mokken model assuming monotonicity and homogeneity (Croon,1991; Hoijtink,1998) (b) Ordered latent class model that is equivalent to the Mokken model assuming double monotonicity (Hoijtink,1998)
For model (a), the probability of a correct response for item i in latent class c is constrained to be higher than the probability in latent class c-1. For model (b), in conjunction with the constraint for model (a) the order of item proportion correct within each class is also constrained to be equal.
After some research, I am happy to find that Mplus is capable of estimating some mixture models using Bayesian estimator. Can Mplus fit these two models using Bayesian estimator as well? I am particularly unsure about how these complex constraints can be set in Mplus.
Yes - you can estimate these models. Take a look at example 5.20 in the user's guide to see how the "model constraint:" command is used to include parameter ordering. You essentially will need to setup inequalities between the threshold parameters for (a). For (b) you just give the same label to all thresholds within the same class. It would be useful for such models to provide starting values for all parameters that are ordered.
shihyingyao posted on Wednesday, March 07, 2012 - 1:06 pm
Thanks for your reply. The version of Mplus in my hand is version 5. Is the version capable of fitting these models "Mplus version 6.11(with Mixture Add-On)"?