A Latent Markov Model for Poverty Analysis : the Case of the GSOEP

The paper is intended to be an empirical suggestion (the German Socio Economic Panel dat are analysed) relating to the traditional analysis of poverty, essentially conducted with poverty lines and equivalence scales. In fact, a model for the measurement error correction should be considered if this kind of analysis is to be conducted, particularly given the well-known low reliability of these measures. Such a model, also used for item non-response, must surely give a higher reliability for poverty measures obtained. For this purpose, a Mixed Markov Latent Chains model is applied either to correct the measurement errors that undoubtedly affect the monetary variables normally used in this approach (income or consumption expenditure) or to redefine the concept of poverty as as phenomenon that cannot be directly, this aspect appears to be especially necessary in a longitudinal analysis where the transitions between poverty and non-poverty states are considered. Furthermore, it is well known that, at the manifest level, these transition results are over-estimated, showing transitions that do not necessarily correspond to a real change in life conditions. The results of the empirical work seem to confirm the above statements.