Course Text: Page 26
Mixture latent Markov (MLM) models are latent class models containing both time-constant and time-varying discrete latent variables. They can be extremely useful in analyzing data arising from longitudinal surveys, clinical trials and related designs. Such models often fit data better than latent growth models since the autocorrelation structure in data often satisfies the Markov assumption made by the MLM model.
MLM models are introduced and illustrated in three tutorials with real world data examples using the Menu/Graphical User Interface (GUI) which can easily accommodate even hundreds of time points.
Common variations of the MLM can be tested easily such as restricting the transition structure (change) to be time homogeneous, and restricting the mover-stayer structure so that 1 latent class (stayer class) does not change. We will see that ‘longitudinal bivariate residuals’, which were added in version 5.0 of LatentGOLD, can assist in selecting the most appropriate variation.
The three examples are:
Tutorial 1: Simple latent Markov model for brand switching
Tutorial 2: Analysis of Satisfaction Data
Tutorial 3: Mover-Stayer latent Markov model for studying change in drug usage
These examples were discussed as part of this Markov power point presentation