Revisiting traditional Welfare State taxonomies from an unsupervised clustering approach: the case of the European Union countries

The aim of this paper is to revisit some of the well-established Welfare State taxonomies adopting a data-driven approach. Particularly important in the field is the typology of Welfare States proposed by Esping-Andersen (1990), which has served as a basis to numerous works. To fulfil the aim of the paper, we compile a comprehensive list of EUROMOD welfare indicators, specifically those being commonly used in the related literature. Then, we applied machine- learning unsupervised cluster techniques to determine the number of groups found and the countries belonging to each group. Previously, when needed, we reduce the dimensionality problem by applying Principal Components Analysis (PCA) techniques to prevent the inclusion of redundant information in the model. Once we have extracted the subjacent trends, unsupervised techniques of clustering are applied to obtain alternatives classifications of Welfare States. In this regard, the contribution of this paper is twofold. First, we broaden the perspective from which Welfare State is quantified by including indicators that encompass simultaneously all its fundamental features. Particularly, following the previous studies, we have selected indicators to quantify the degree of redistribution, the progressivity of the tax-benefit systems, the relative weight of each instrument, the poverty reduction achieved. Second, we obtain for each pair of countries the likelihood of belonging to the same cluster/group, depending on the set of features included.