Display of small-area variation in health-related data: a methodology using resistant statistics

Health care planning requires characterization of the population to be served. Examination of available demographic and epidemiologic data is one early step in this process. However, aggregate data for the entire geographic area of concern often fail to reveal important differences among geographically defined sub-populations--differences that influence the form an effective delivery system should take. We present a methodology based on exploratory data analysis (EDA) techniques that we have found useful in examining health-related data for our ambulatory care catchment area. Our examples use three population characteristics that have major implications for health care planning for the elderly: 1970-1980 change in population aged 65+; the percent of the population aged 65+ below poverty level; and the percent of single-person households among households with one or more persons aged 65+. With these data for the 25 municipalities of Middlesex County, New Jersey, we illustrate a two-step process: (1) the construction of stem-and-leaf displays that permit examination of a data distribution for asymmetry, concentrations around specific values, gaps in values, and outliers; and (2) the use of the median, the fourth-spread, and other information from the stem-and-leaf display in the systematic selection of data value classes to be given distinct shadings on a map of the selected geographic area. Discussion emphasizes the usefulness of graphic display of data in detecting similarities and unusual data values. Comparison of maps based on the EDA techniques and maps based on several traditional methods of value classing for the same data illustrates the influence of classing choices on the interpretation of cartographic displays of health-related data.