'Disparate Impact': Regulators Need a Lesson in Statistics

In April the Consumer Financial Protection Bureau issued a statement that it was adopting the "disparate impact" concept in its enforcement of the anti-discrimination provisions of the Equal Credit Opportunity Act. It is doing so in a manner consistent with an Interagency Policy Statement issued in 1994 by the Department of Justice and other federal agencies involved with the enforcement of fair lending laws.

This is one new chapter in a remarkably perverse episode of federal law enforcement. The problem is that federal agencies' failure to recognize a fundamental statistical concept results in encouraging lenders to take actions that make them more likely targets for litigation.

The disparate impact concept has its origin in employment discrimination cases going back to the late 1960s, where plaintiffs challenged tests and educational requirements that disproportionately disadvantaged black applicants or employees. The concept was formally legitimized for employment cases in the Civil Rights Act of 1991.

In the employment setting, disparate impact doctrine essentially holds that, even though an employer does not intend to discriminate against a protected group, it cannot use a device or practice that disproportionately disadvantages such group unless the device or practice serves a sound business interest. Further, there must be no less discriminatory alternative that equally serves that interest.

Beginning in 1989, lenders were required to keep records reflecting the race of applicants for home mortgages. Studies immediately appeared showing large differences between rates at which minorities and whites were denied mortgages. Often rejection rates for minorities were several times those of whites. Initially, these rejection rate differences were attributed to willful discrimination by lenders. But in the early 1990s, there emerged a recognition that differences in rejection rates resulted in significant part from that fact that minorities were less able to meet standard lending criteria just as minorities were often less able to meet certain employment criteria.

In March 1994, that recognition led ten federal agencies involved with monitoring or enforcing fair lending laws to issue an Interagency Policy Statement citing the disparate impact lenders' policies may have on minorities and stating that the agencies regarded unjustified disparate impacts to violate federal lending laws. The statement cited minimum loan amounts as an example of a practice that might have greater adverse impact on minorities than whites, and it generally called into question all unnecessarily stringent lending criteria.

The discouragement of unnecessarily stringent lending criteria accorded with thinking in the employment testing context where one universally recognized way of reducing a test's disparate impact on a protected group was to lower cutoff scores. That worked like this. Suppose that at a particular cutoff point pass rates are 80% for an advantaged group and 63% for a disadvantaged group. At this cutoff the advantaged group's pass rate is 27% higher than the disadvantaged group's pass rate. If the cutoff is lowered to the point where 95% of the advantaged group passes the test, assuming normal test score distributions, the disadvantaged group's pass rate would be about 87%. Thus, with the lower cutoff, the advantaged group's pass rate would be only 9.2% higher than the disadvantaged group’s pass rate.

Lending criteria operate just like test cutoffs, and, as with the lowering of test cutoffs, relaxing those criteria tends to reduce relative differences in meeting them. The extent to which lenders relaxed their criteria in response to the Interagency Policy Statement or other discouragements of unduly stringent lending criteria is unknown. But assuming some lenders did so, one can also assume that relative differences between rates at which whites and minorities secured mortgages decreased.

Simple enough so far. But, while lowering the cutoff tends to reduce relative differences in passing rates, it also tends to increase relative differences in non-passing rates.

In the situation just described, the disadvantaged group’s non-passing rate was initially 1.85 times the advantaged group's non-passing rate (37%/20%). With the lower cutoff, the disadvantaged group's non-passing rate would be 2.6 times the advantaged group's non-passing rate (13%/5%).

This pattern is not peculiar to test score data or the numbers I chose to illustrate it. Lowering a credit score requirement will reduce relative differences in meeting the requirement but increase relative differences in not meeting it.

Nevertheless, regulators and others concerned about disparities in lending outcomes continued to measure the size of those disparities in terms of relative differences in mortgage rejection rates. Thus, lenders that were most responsive to the encouragement to relax criteria — hence, more so than other lenders, tending to reduce relative differences in approval rates while increasing relative differences in rejection rates — were regarded as the most discriminatory lenders. The actions of the DOJ and other regulators were akin to pressuring employers to lower test cutoffs and then singling out for litigation the employers who lower their test cutoffs the most.

One would think a perverse situation where, in enforcing fair lending laws, the federal government encourages lenders to engage in conduct that makes them more likely to be sued for discrimination could not persist for long. But the situation has existed for 18 years with no sign of abating.

The DOJ's complaint filed in conjunction with the record $335 million settlement of lending discrimination claims against Bank of America's Countrywide Financial Corporation last December sent lenders one clear message: once aware that their practices lead to racial differences in adverse lending outcomes, lenders must seek out less discriminatory alternatives.

And the complaint specifically criticized the defendant for practices that increased the frequency of such things as assignment to subprime loans. As indicated above, however, measures that reduce adverse outcome rates, while reducing differences in favorable outcome rates, tend to increase differences in adverse outcome rates on which lenders are judged.

There exist many situations where the failure to understand fundamental statistical concepts leads to distorted interpretations of data on group differences.  But, as reflected by the size of the Countrywide settlement, it is in the lending industry that the financial consequences of the failure may be the most severe.

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