James P. Scanlan, Attorney at Law

Adjustment Issues – Lending Disparities

(Feb. 20, 2012)

One of the remarkable things about lending disparity studies (though a like point could also be made about a great range of studies on group differences in the law and the social and medical sciences) is the manner in which such studies purport to address group differences in outcome-related characteristics by dividing the groups being compared into categories of such characteristics, or adjusting outcome rate differences according to the group’s varying distributions among those categories, without discussion of the inadequacy of the categories to fully address group differences in characteristics.  Anyone with the least understanding of normal distributions will recognize that even when the categories appear to be narrowly defined (and of course more so when they are broadly defined), the group that on average has weaker qualifications relating to securing the outcome will be disproportionately represented within the lower reaches of each adjustment category.

In the case of lending disparities studies, the categories are typically defined by income, with the groups that are being compared divided into four (low, moderate, middle, and high)  [i] or three (low and moderate, middle, and high) [ii]  income categories.  As it happens, it is with respect to income categories that the way in which groups that are disproportionately concentrated in the lower categories will tend to be disproportionately concentrated in the lower levels of each income category.

Table 1 below is based on Table 1 of “Can We Actually Measure Health Disparities” (Chance 2006), which is based on published census data from 2005.  The fourth last column of the article’s Table 1 shows that way that blacks comprise a higher proportion of the combined black and white population falling below each point defined by a percentage of the poverty line, a pattern that is also illustrated in the article’s Figure 1.  Table 1 below shows the proportion blacks comprise of the population falling between each of 13 points, thus within 12 categories.  And it shows that if the twelve categories were grouped into six (e.g., between 1000% and 500% of the poverty line, etc.), blacks would comprise a larger proportion of the lower part of the categories.

Few people with an understanding of normal distributions would question that, if the data were made available, it would likely show that within each of the 12 categories blacks comprise a larger proportion of the group with lower incomes within the category.  The same would presumably be the case with 24 or 48 categories, though, to be sure, at increasing levels of granularity one might observe occasional departures from the patterns.  But certainly it makes no sense to say that dividing the groups into three or four income categories adequately addresses income differences. 

Table 1.  Black Proportion of Combined Black and White Populations Falling Between Points Defined by Percentage of the Poverty Line

  Similar patterns can be illustrated with data on credit scores by race, as is done in the Credit Score Illustrations sub-page of the Scanlan’s Rule page.