Alice Frye posted on Wednesday, October 08, 2008 - 12:36 pm
LCA uses the MLR estimator. Does this mean that there is no additional value to be gained by using a negative binomial or zero inflated negative binomial regression for categorical variables within a latent profile analysis, because the MLR is sufficiently impervious to violations of distributional assumptions?
I'd be grateful for any comments or references people might have on this topic.
MLR may not be sufficient for continuous variables where deviations from normality are extreme and for categorical and count variables with strong floor or ceiling effects. Special modeling like the negative binomial model for count variables may be needed in these situations.
Q1-Q2: See UG ex7.22. The references given there show that this is an acceptable model.
Q3-Q4: You cannot use a regular chi-square difference test to check on the number of classes because the assumptions behind that test are not met. See
Nylund, K.L., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling. A Monte Carlo simulation study. Structural Equation Modeling, 14, 535-569. download paper show abstract