Regions and regional patterns on choropleth maps

Two types of maps, choropleth maps and maps of statistically-derived regions, are compared in order to investigate how well classed choropleth maps represent the regional structure of geographic distributions. Regional patterns on choropleth maps are formed by contiguous unit areas belonging to the same class. These regional patterns vary depending upon the nature of the geographic distribution, the method of classification, and the number of classes. Maps of statistically-derived homogeneous regions are constructed using various techniques of contiguity-constrained cluster analysis. Then choropleth maps are compared to region maps to determine if the manipulation of the number of classes on a choropleth map can produce regional patterns that accurately reflect the regional structure of the original geographic distribution.;The research using a computer simulation of geographic distributions and their maps. Four types of pseudorandom data distributions are used: uniform, normal, skewed, and bimodal. These are each placed on a base map of 100 hexagons, whose configuration of contiguities models the contiguity structure of irregular polygons in the real world. The data is located on the base map to create geographic distributions with various levels of spatial auto-correlation. Choropleth maps are made with various numbers of classes using Jenks' classification method. A number of different algorithms for contiguity-constrained cluster analysis are used to create the statistically-derived regions that account for 90% of the variance in the original data.;The simulations show that different numbers of classes result from different types of data distributions for choropleth mapping using Jenks' method and a Goodness of Fit (GVF) criterion. Ward's method of hierarchical agglomerative contiguity-constrained cluster analysis followed by an iterative swapping procedure, is the method that most consistently produces the least number of regions for a given GVF criterion. Comparison of the choropleth maps to the region maps indicates that the appropriate number of classes to use to portray regional patterns is the same as the least number of classes that accounts for 90% of the variance in the data, with one exception. When the number of classes is two, that number should be increased to three. In addition this research shows that region maps always produce a fewer number of regions than the number of regional patterns on a choropleth map, and this difference is dependent upon the spatial autocorrelation of the geographic distribution.