Evaluation of methods for classifying epidemiological data on choropleth maps in series

Our research goal was to determine which choropleth classification methods are most suitable for epidemiological rate maps. We compared seven methods using responses by fifty-six subjects in a two-part experiment involving nine series of U.S. mortality maps. Subjects answered a wide range of general map-reading questions that involved individual maps and comparisons among maps in a series. The questions addressed varied scales of map-reading, from individual enumeration units, to regions, to whole-map distributions. Quantiles and minimum boundary error classification methods were best suited for these general choropleth map-reading tasks. Natural breaks (Jenks) and a hybrid version of equal-intervals classing formed a second grouping in the results, both producing responses less than 70 percent as accurate as for quantiles. Using matched legends across a series of maps (when possible) increased map-comparison accuracy by approximately 28 percent. The advantages of careful optimization procedures in choropleth classification seem to offer no benefit over the simpler quantile method for the general map-reading tasks tested in the reported experiment.