There is a documented significant mismatch between poverty indirectly measured using a monetary, and direct measures focusing on material deprivation (MD). The latter is regarded as a direct method of poverty measurement. This mismatch largely prevented the implementation of focused social policies since the target of the policy is often "muddy".
The main goal of the project is to identify and develop new measures and models for material deprivation. It is part of the more general problem of measurement of well-being in a multi-dimensional and inter-temporal perspective.
The main contribution of the project is both methodological and empirical. Methodologically we plan to develop a measure of material deprivation in a dynamic latent state modelling framework that classifies the population by deprivation status without resorting to arbitrarily specified thresholds. This extends in the discrete space previous results. Classification corresponds to membership to a latent deprivation status. Each individual is not identified with certainty but through an estimated posterior probability of belonging to the deprivation status. Our proposed approach addresses some issues that so far limited the operativeness of MD measurement. The project also aims at assessing the item discriminant validity, in terms of deprivation measurement, by identifying the most discriminant indicators among a plausible set. We plan to overcome the potential issue of equally weighted items by a configuration of a deprivation score with country-specific optimal weighting.
The analysis of the statistical information available in the European Union is also a key element of the project. Using the longitudinal component of the EU-Statistics on income and living conditions, we plan to analyse the deprivation pattern in European countries with particular emphasis on mobility between deprivation and non-deprivation classes and persistent deprivation.
The project plans to improve the measurement of deprivation, and generally measurement of multi-dimensional poverty, by developing a conceptually meaningful framework that classifies individuals in terms of persistent deprivation status and estimates the probability of transition between status of deprivation and not deprivation over time. The traditional classifications of agents within a society into groups based on their social and economic conditions could be questioned when they are based on deprivation counting, for instance because the boundaries between different categories are perceived as arbitrary.
From the perspective of statistical methodology the innovation is two-fold. First of all, we will develop latent Markov models with several sources of unobserved heterogeneity and dependence among measurements (here, repeated measurements on the same subjects are dependent over time and subjects belonging to the same nation are dependent over space and time) where the outcome is multidimensional. Secondly, we will develop a method for optimal dimension reduction of the outcome. The outcome will be summarized in one dimension finally obtaining a weighted sum of the outcomes which can be used as a simple score for deprivation. Optimal weights are those providing the smallest information loss. A peculiar feature of our setting is that dimensionality reduction might be lossless, a situation which we will formally characterize. Simultaneous estimation and dimensionality reduction of latent Markov models has never been considered in the literature so far. Third, we will develop formal methods for item selection by extending our dimensionality reduction approach to a sparse dimensionality reduction approach where weights can be zero.
The dynamic latent model we want to develop will be able to summarize the information of the multi-dimensional outcome, naturally providing error correction of measurements coming from a list of items. The model is innovative since it is able to detect the most discriminant items that are items more associated to the (directly unobservable) status of deprivation and, once estimated, is able to solve simultaneously the issue of the optimal number of items in the list and the issue of weighting the items. Individuals are classified, within this new framework, as deprived if their posterior probability of belonging to the unobserved status of deprivation is greater than a given value and it is possible to classify individuals by using a continuum of thresholds, overcoming the traditional problem relating to setting a fixed arbitrary value or a a fixed number of lacking items.
In this respect, the expected results of this research are highly innovative and can contribute to the state of knowledge in several fields.
Methodologically, it contributes to generalize latent Markov (LM) models for longitudinal data when we have multivariate outcomes. Moreover, the usual LM formulation uses a single random effect, while in this development we might have residual dependence related to the fact that individuals are clustered in countries of residence. The estimation of optimal weighting is a new task not explored in the literature so far.
For statistical offices, the potential stakeholders, the expected outcome of the project provides new measurement tools for complex and multi-dimensional phenomena. The importance of handling multi-dimensional phenomena in measuring inequalities and poverty has been recently enhanced by the World Bank in its Report of the Commission on Global Poverty (2016). One of the most relevant recommendation given by the Commission is the implementation of a multi-dimensional index of poverty that should not depend on the assumptions of the researcher. Our project goes in that direction.
In terms of policy relevance, policymakers can be informed on the saliency of chronic deprivation and well-being disentangled from transitory events, and how it varies along observables as age, gender, education, and how it overlaps with income-based measurements. The expected results might help re-orienting target social policies in European countries.