Clustering multivariate spatial data based on local measures of spatial autocorrelation. An application to the labour market of Umbria

A growing interest in clustering spatial data is emerging in several areas, from local economic development to epidemiology, from remote sensing data to environment analyses. However, methods and procedures to face such problem are still lacking. Local measures of spatial autocorrelation aim at identifying patterns of spatial dependence within the study region. Mapping these measures provide the basic building block for identifying spatial clusters of units. If this may work satisfactorily in the univariate case, most of the real problems have a multidimensional nature. Thus, we need a clustering method based on both the multivariate data information and the spatial distribution of units. In this paper we propose a procedure for exploring and discover patterns of spatial clustering. We discuss an implementation of the popular partitioning algorithm known as K-means which incorporates the spatial structure of the data through the use of local measures of spatial autocorrelation. An example based on a set of variables related to the labour market of the Italian region Umbria is presented and deeply discussed.