On measuring polarization for ordinal data: an approach based on the decomposition of the Leti index

This paper deals with the measurement of polarization for ordinal data. Polarization in the distribution of an ordinal variable is measured by using the decomposition of the Leti heterogeneity index. The ratio of the between-group component of the index to the within-group component is used to measure the degree of polarization for an ordinal variable. This polarization measure does not require imposing cardinality on ordered categories to quantify the degree of polarization in the distribution of an ordinal variable. We address the practical issue of identifying groups by using classification trees for ordinal variables. This tree-based approach uncovers the most homogeneous groups from observed data, discovering the patterns of polarization in a data-driven way. An application to Italian survey data on self-reported health status is shown.