An Integrated Spatial Statistics Package for Map Data Analysis

In the analysis of geocoded statistical data, common practice has been to treat the data in isolation from its locational or spatial characteristics. This results in a potentially critical loss of the spatial information that is contained in the mapped representation of the statisti cal data but not in the application of aspatial statistical techniques such as cross-sectional regression. One explanation for the domination of aspatial methods is the absence of an integrated software package to carry out explicitly spatial data analysis on geocoded databases. In this paper we describe the functional elements of such a package, based on our development of a typology of general-purpose statistical map data analysis needs. The design of the integrated spatial statistics package includes a module for transforming digital coordinate data (e.g. polygonal boundary files) into concise mathematical representations of relative location for use in the analytic procedures. We distinguish between several categories of locationally-referenced data including point patterns, point-sampled data, grid-cell data, irregular polygon-based data and network and flow data. Within these classes we consider a hierarchy of spatial analytic objec tives: quantitative pattern description and randomness testing, univariate modeling, spatial interpolation, space series analysis, multivariate spatial analysis, spatial forecasting and control, and dynamic and space/time analysis. Examples of the application of the package to typical spatial statistical problems are discussed, with emphasis on the hierarchial nature of the analytic process.