5 Regionalisation Tools for the Exploratory Spatial Analysis of Health Data Steve Wise, Robert Haining and Jingsheng Ma Sheffield Centre for Geographic Information and Spatial Analysis, Department of Geography, University ofSheffield, Sheffield SIO 21N, UK. 5.1 Introduction This paper considers issues associated with the construction of regions as part of a programrne of exploratory spatial data analysis in the case of what Cressie (1991) refers to as "Iattice data". Lattice data arise where a study area has been partitioned into a set of zones or regions attached to each of which is a vector that describes the set of attributes for that zone. The focus of this paper will be the analysis of health data so the attributes in question may be health related but may also include demographic, socio-economic and environmental attributes. Exploratory spatial data analysis (ESDA) comprises a set of statistically robust techniques that can be used to identify different forms of spatial variation in spatial data. ESDA represents an extension of exploratory data analysis (EDA) to handle spatially referenced data where in addition to the need for techniques to identify distributional properties of a set of data there is also a need for techniques to identify spatial distributional properties of the data. Typically these techniques comprise numerical summaries (e.g. measures of central tendency, measures of dispersion, regression) and graphic displays (e.g. boxplots, histograms, scatterplots) but in the case of ESDA cartographic displays make a vital and distinctive additional contribution enabling the analyst to see each attribute value in its geographical context relative to other attribute values. Like EDA therefQre, ESDA exploits different methods ofvisualisation. Moreover like EDA, ESDA is not associated with any one stage in the process of data analysis for the techniques can be appropriate both for preliminary stages of analysis (pattern detection; hypothesis formulation) and later stages (model assessment). EDA is underpinned by a conceptual data model in which data values comprise an element that is "smooth" (sometimes called the "fit") and an element that is "rough" (sometimes called the "residual"). In the case of ESDA the "smooth" is oft:en associated with large scale patterns (such as spatial trends, patterns of autocovariation or concentration) whilst the "rough" may be outliers, that is individual areas with values that are higher ("hot") or lower ("cold") than found in the neighbouring areas. To identify these data properties in the case of ESDA a number oftechniques have been brought together, either adapted from EDA (e.g. median polish to detect trends (Cressie 1984» or custom developed for the identification of spatial properties (e.g. M. M. Fischer et al. (eds.), Recent Developments in Spatial Analysis © Springer-Verlag Berlin Heidelberg 1997