CLASS INTERVAL SYSTEMS ON MAPS OF OBSERVED CORRELATED DISTRIBUTIONS

Abstract

The degree to which choropleth maps accurately reflect product-moment correlation between sets of data depends upon the statistical distributions of the data and the system by which they are divided into class intervals. Comparing results for observed distributions with earlier results for normal distributions, it is found that: (1) the expected values of rank correlations between maps (given the product-moment correlation between the data sets) appear to be similar regardless of class interval method and regardless of whether data are normal or not; (2) class intervals based on standard deviation units have relatively smaller degrees of variation about expected values when observed data are being compared, while quantiles have small variations on the normal data (quantiles being a special case of standard deviations in that situation); and (3) the choice of interval system remains important for higher numbers of observations when observed, rather than normal, data are being considered.