A Statistical Method for the Identification of Spatial Clusters

Abstract

A method is presented for the identification and location of clusters in data reported in small areal units. The areal units are aggregated into zones which are not prespecified at the outset and which may be of any shape or size. A statistical distribution is described against which any apparent clusters in the zones can be assessed for statistical significance. Monte Carlo methods are used to derive the distribution, and it is argued that complex null hypotheses can be represented in this way. In this approach the critical distinction between prespecified and constructed areas or zones of interest is recognised, and is shown to circumvent some of the difficulties usually associated with modifiable areal units.