Principal component analysis for geographical data: the role of spatial effects in the definition of composite indicators

Abstract

This paper investigates the role of spatial dependence, spatial heterogeneity and spatial scale in principal component analysis for geographically distributed data. It considers spatial heterogeneity by adopting geographically weighted principal component analysis at a fine spatial resolution. Moreover, it focuses on dependence by introducing a novel approach based on spatial filtering. These methods are applied in order to derive a composite indicator of socioeconomic deprivation in the Italian province of Rome while considering two spatial scales: municipalities and localities. The results show that considering spatial information uncovers a range of issues, including neighbourhood effects, which are useful in order to improve local policies.