A comparison of some criteria for states selection in the latent Markov model for longitudinal data

TL;DR

This paper compares various criteria for selecting the number of latent states in a multivariate latent Markov model for longitudinal data, finding that likelihood-based criteria generally outperform classification-based ones.

Contribution

It provides a systematic comparison of different selection criteria, highlighting the effectiveness of likelihood-based indices over classification-based criteria in latent Markov models.

Findings

Likelihood-based criteria perform better in selecting the true number of states.

Classification-based criteria tend to underestimate the number of latent states.

Likelihood criteria show robustness in both univariate and multivariate cases.

Abstract

We compare different selection criteria to choose the number of latent states of a multivariate latent Markov model for longitudinal data. This model is based on an underlying Markov chain to represent the evolution of a latent characteristic of a group of individuals over time. Then, the response variables observed at the different occasions are assumed to be conditionally independent given this chain. Maximum likelihood of the model is carried out through an Expectation-Maximization algorithm based on forward-backward recursions which are well known in the hidden Markov literature for time series. The selection criteria we consider in our comparison are based on penalized versions of the maximum log-likelihood or on the posterior probabilities of belonging to each latent state, that is the conditional probability of the latent state given the observed data. A Monte Carlo simulation study shows that the indices referred to the log-likelihood based information criteria perform in general better with respect to those referred to the classification based criteria. This is due to the fact that the latter tend to underestimate the true number of latent states, especially in the univariate case.