ABSTRACT
Often when the topic of the accuracy of spatial databases arises one thinks in terms of the positional accuracy of the punctiform, linear, areal or volumetric features of the earth's surface. In many GIS applications, this type of focus is appropriate and necessary for the task at hand which may involve map overlay operations or terrain modelling in a natural resources context or may have legal implications in a cadastral database context. For most GIS applications in human geography, the research context involves the use of census data or data tabulated by census enumeration units. For these applications, which may involve dozens or even hundreds of data layers with the same polygonal boundaries (choropleth maps), the concept of the accuracy of the spatial database is quite different. This paper will briefly review the types of error on choropleth maps which are commonly held to be important, and introduce an overlooked component of choropleth map accuracy as the Small Number Problem. A short history of the Small Number Problem, and a review of various attempts to minimize it or solve it will be given with specific examples from epidemiologic, health care delivery and health care resource allocation contexts.
