Poverty risk and territorial disparities in Europe: a clusterwise lasso regression approach

Abstract

In 2024, more than 93 million people in the EU were at risk of poverty or social exclusion, highlighting persistent territorial disparities and the urgent need for accurate and robust measurement. This article proposes an innovative two-stage methodology to improve the estimation of poverty risk across European territories by leveraging high-dimensional socio-economic data. The approach combines a clustering phase, which captures territorial heterogeneity, with Lasso regression for variable selection and model simplification, ensuring parsimony and interpretability. The application to different degrees of urbanisation reveals that the determinants of poverty vary spatially, with significant differences both in the composition of clusters and in the relevance of explanatory variables. The results provide new insights for the design of effective, territorially targeted anti-poverty policies and contribute to the European debate on regional inequalities through a comparative and statistically sound approach.