Statistical Quirks Confound Lending Bias Claims

On July 12, the Department of Justice announced a $175 million settlement of claims that Wells Fargo discriminated against black and Hispanic home loan borrowers. The settlement was the Justice Department's largest in a fair lending case after the $335 million settlement with Bank of America's Countrywide Financial Corp. last December.

The department's complaint in the Wells Fargo case, like that in the Countrywide case, implicates the perverse anomaly in fair lending enforcement addressed in my earlier blog post: Lenders are pressured to do things that increase their likelihood of getting sued. It also implicates important related issues concerning the merits of such cases.

The Wells Fargo and Countrywide complaints both fault the defendants for anything that fails to minimize the proportion of loans that are subprime. But, as previously explained, for reasons inherent in normal distributions of factors related to some outcome, reducing the frequency of adverse outcomes, while tending to reduce relative differences in favorable outcomes, tends to increase relative differences in adverse outcomes. And regulators focus on the latter when monitoring the fairness of lending practices. Thus, lenders that do the most to make loans prime rather than subprime will tend to be the most likely targets for litigation.

Paragraph 2 in the Wells Fargo complaint highlights other issues involving the same poorly understood statistical forces that cause minimizing adverse lending outcomes to increase relative differences in those outcomes. The paragraph states that "between 2004 and 2008, highly qualified prime retail and wholesale applicants for Wells Fargo residential mortgage loans were more than four times as likely to receive a subprime loan if they were African-American and more than three times as likely if they were Hispanic than if they were white."

The provocative claim, which was picked up in media accounts of the settlement, certainly suggests it is highly unlikely that, absent discrimination, one would find such large disparities even among highly qualified applicants – where there is no apparent reason why any loan should be assigned subprime status much less why it should happen to minorities much more often than whites. Few observers, however, would appreciate that racial and ethnic differences in assignment to subprime status tend to be large among highly qualified applicants simply because rates of assignment to subprime status tend to be very low among highly qualified applicants. Differences in rates of assignment to prime status, however, tend to be small among such applicants.

The crucial issue in appraising whether the manifold differences in assignment to subprime status reflect differences in treatment based on race or ethnicity is whether white and minority applicants deemed by the DOJ to be highly qualified are truly comparable with respect to factors that affect assignment to subprime status. Fortunately, the complaint adds a footnote explaining that applicants deemed highly qualified included only persons whose credit scores were 680 or above.  

But features of normal distributions raise questions as to whether any claim that a broad, or even comparatively narrow, categorization of demographic groups can effectively account for differences in outcome-related characteristics of the groups. Disadvantaged groups will invariably be disproportionately concentrated within the lower ranges within each category.  ­­

Publicly available data from a case brought against Wells Fargo in California in 2008 illustrate this point. When borrowers with scores of 680 or above are broken down into eight ranges used in an expert report in the California case, the percentages blacks and Hispanics comprise of borrowers within the range consistently increase as the score range decreases. The percentages increase from 2.2% in the highest range to 8.4% in the lowest range for blacks and from 4.7% in the highest to 12.4% in the lowest for Hispanics. Thus, not only would blacks and Hispanics have lower average scores than whites within the DOJ's highly qualified category, they would have lower average scores within each of four more refined subcategories. And one would generally find that even within each of the eight ranges, groups with lower average overall scores tend to be disproportionately concentrated within the lower half of the range (as well as within the lower half of the lower half and so on).

While these data are from a different Wells Fargo matter, one ordinarily will find similar patterns, not simply among any group of Wells Fargo borrowers, but among borrowers anywhere, so long as substantial racial and ethnic differences in things like income and credit scores exist. At any rate, minorities and whites within the highly qualified group discussed in Paragraph 2 of the Wells Fargo complaint are hardly similarly situated in a way warranting a belief that the cited differences in subprime assignment rates probably result from discrimination.

Whether there actually is merit to the claims in the Countrywide or Wells Fargo cases or others like them is anyone's guess. But it will be a better guess if informed by a clear understanding of the statistical patterns described above.

James P. Scanlan is a lawyer in Washington. He specializes in the use of statistics in litigation.