A multivariate latent Markov model for the analysis of criminal trajectories

摘要

We introduce a multivariate version of the latent Markov model for the investigation of criminal trajectories whose transition matrix may be suitably constrained in order to formulate hypotheses of interest on the criminal behaviour. For the maximum likelihood estimation of the model and its constrained versions we outline an EM-type algorithm. We also illustrate a simple procedure based on the likelihood ratio for choosing the number of states and testing restrictions on the transition matrix.