Geographic Enhancement of Choropleth Data.

Abstract

For a variety of reasons it is often necessary to convert data which have been collected within a set of geographical regions so that observations lie at the nodes of a regular square grid. This conversion is one example of a cartographic data transformation, and involves enhancing the areal or choropleth data to recover the information lost during spatial generalization. This study examines the factors which contribute to the absolute and relative error associated with six different choropleth to grid lattice conversion techniques. The six techniques were tested under controlled circumstances using theoretically derived data sampled by choropleth units of various sizes and shapes. The study determined that resolution, characteristics of the choropleth units, and shape had little impact upon conversion error, except in combinations of extreme circumstances. Surface complexity, however, was a major determinant of error. These factors were in many cases of varying importance to the different techniques. On average, the techniques which modified the choropleth matrix itself outperformed the point interpolation methods. A model of error was constructed to predict absolute error based on factors other than complexity. In test conditions, this model explained between 16 and 43 percent of aggregate error, though both better and worse results were obtained in particular applications. The study recommends that the investigator should not apply techniques which make specific assumptions about the underlying surface when the nature of this surface is unknown. In these cases, the simplest techniques seem to produce the best results.