Latent class Markov models for addressing measurement problems in poverty dynamics

The traditional approach to poverty measurement utilises only monetary variables as indicators of individuals’ intensity of the state of deprivation, causing measurement errors of the phenomenon under investigation. Moreover, when adopted in a longitudinal context, this approach tends to overestimate transition poverty. Since poverty is not directly observable, a latent definition can be adopted: in such a conception is possible to use Markov chain models in their latent acceptation. This paper proposes to use Latent class Markov models which allow taking into account more observed (manifest) variables. We define those variables via monetary and non-monetary fuzzy indicators.