finding the plaintiffs’ expert report to be flawed where it failed to account for the variable of job title which, by their own theory, was a “major factor”
How later courts described this case
- finding the plaintiffs’ expert report to be flawed where it failed to account for the variable of job title which, by their own theory, was a “major factor”
- finding plaintiff's regression analysis so incomplete as to be irrelevant because it failed to account for key variables in the applicant pool
- finding the McDonnell Douglas framework appropriate in a § 1981 suit involving intentional wage discrimination
- granting summary judgment for defendant employer where plaintiffs' statistical analysis was flawed as a matter of law
Written by the judges who cited it.
The opinion
SENTELLE, Circuit Judge,
concurring in part and concurring in the judgment:
I concur with the majority’s result, but do not join all its reasoning in Part III of its opinion.
The majority correctly judges Coward’s regression analysis, which controlled only for race and years of service, “so incomplete as to be inadmissible as irrelevant.” Maj. Op. at 274 (quoting Bazemore v. Friday, 478 U.S. 385 , 400 n. 10, 106 S.Ct. 3000 , 106 S.Ct. at 3009 n. 10, 92 L.Ed.2d 315 (1986)). The district court had concluded that the regression failed to include “essential variables” like education and experience. The majority today finds a different flaw: the omission of “job title or any other variable representing type of work performed.” Maj. Op. at 274.
From either perspective, the bottom line is this: the regression has no probative value because it does not compare employees who can reasonably be deemed “similarly situated.” Besides race, the regression controls only for years of service at ADT. Thus, it groups together all employees with equivalent seniority, whether mailroom clerks or high-level managers, so long as they have been with the company an equivalent length of time. Such a classification, applied to the entire body of ADT employees, is too broad: Under no reasonable definition of the term could employees be considered similarly situated on the basis of seniority alone. As we have said before, when a “regression model ignores information central to understanding the causal relationships at issue,” it does not adequately raise an inference that discrimination accounts for salary differences among employees. Valentino v. United States Postal Service, 674 F.2d 56, 71 (D.C.Cir.1982). Employee pay depends on far too many additional factors to allow this salary comparison based only on seniority to raise a presumption of race discrimination. Thus, the regression is irrelevant and was properly excluded by the district court.
The majority, like the district court, attempts to identify specifically what the regression should have included to render it probative. The flaw with the regression is neither that it excluded “major factors” like education or experience, nor that it failed to account for “job title” or “type of work performed.” There are a variety of different (perhaps even mutually exclusive) ways to *277 select a set of independent variables which will group together employees who are, more or less, similarly situated. The flaw with this regression is that the independent variable selected did not even begin to classify employees by similar training, experience, performance, duties, or function. Including variables for education and experience might go a long way toward identifying relevant similarities, but so might including variables for job title or type of work. Neither approach is necessarily the “correct” or best one. Either could lead to an admissible regression analysis; it would then be up to the experts and lawyers to argue the meaning and weight to be accorded to it.