Poverty dynamics according to direct, indirect and subjective measures : modelling Markovian processes in a discrete time and space with error

Abstract

A debate is going on about how transitory poverty really is in affluent welfare states. Research findings indicate that usually poverty spells are of a relatively short duration. On the other hand, studies have also shown that poverty is very persistent among a specific part of the population. This study tries to find an explanation for these two seemingly contradictory findings. Modelling panel data from ten EU countries with measurement models, this study is able to reveal that the classical Mover-Stayer model can explain the dynamics of poverty and that, if measurement error is ignored, the mobility in poverty and deprivation transition tables is over-estimated. The mover group and the measurement error explain why there are two seemingly conflicting pictures of poverty dynamics. The financial poverty, housing deprivation and subjective deprivation indicators from the European Community Household Panel (ECHP) were used in four repeated measurements. The descriptive analysis revealed that the population is heterogeneous when relating to poverty dynamics and that the dynamics are similar across the countries and across the direct, indirect and subjective indicators. With the time-heterogeneous partially Latent Mover-Stayer model, we were able to identify these groups: the first group are stayers in poverty, the second are stayers not in poverty and the third group are the movers. With different poverty classifications in different countries, we have different fractions o f population classified into these three groups, but the groups o f stayers and movers are identified in every transition table. The preliminary finding on common poverty dynamics were confirmed with the Latent Constant Fluidity model, which was fitted into the layered transition tables. The three poverty measures in the ten countries have very similar poverty transition probabilities, especially when random error is corrected to, as the error operates at different levels between countries and indicators. Three main conclusions can be drawn from the results. First, there is high poverty mobility but poverty spells seem to concentrate to the same group of people. Second, poverty mobility is over-estimated in panel data if the random error is ignored. Third, poverty dynamics, both the absolute mobility as well as transition probabilities, seem to have a striking affinity across countries and indicators, despite the large differences in the cross-sectional poverty and deprivation rates. We studied only three classifications in ten EU countries, but we can expect that other poverty and deprivation classifications would lead us similar conclusions about poverty dynamics if turned into longitudinal measures.