Guidance for Statistical Determination of Appropriate Percent Minority and Percent Poverty Distributional Cutoff Values Using Census Data for an EPA Region II Environmental Justice Project.

摘要

The goal of this project is to identify a GIS and a statistical procedure which will objectively, reproducibly, and statistically identify a 'Community of Concern' (COC) which is defined as a community with a 'minority' or 'below-poverty' population. We shall demonstrate the procedure using the census data for the state of New Jersey and New York located in EPA's Region 11. This exercise in classification sounds straightforward and doable, but the choice of threshold values or cutoff values and changes of scale (e.g., census block groups to counties) changes the number and location of the COC, and may raise questions and criticism. An objective statistical algorithm is needed for identifying and locating the COC on the map of the Region. This is a non-trivial statistical problem. Because the data have time and space dimensions and skewed probability distributions, hypothesis testing, confidence intervals, and ratios and proportions are inappropriate and hence have the potential to mislead decision-makers. Descriptive analyses of the probability distribution of the data when aggregated to the appropriate scale (census block or group, census tract, town, township, county, state, or region) is an appropriate approach for the data and will give the desired quality for identification of a COC. Decisions will be made from the probability of the cutoff, not from arbitrary cutoff. In this context, it is important to define units and scale. The basic (indivisible) sampling unit of data or information is the census 'block group.' The decision unit changes (e.g., census block group, census tract, township, county, or state) and is chosen by the specific question to be answered. To change scale to a different decision unit other than the census block group (sampling unit), all of the spatially included sampling units in the new decision unit must have the counts of their characteristics summed over the desired decision unit and the desired percentages recomputed. The counts or frequencies are additive but the percentages or relative frequencies (probabilities) are not.