This might come as a shock: Employees in large corporations sometimes mistakenly believe that they have been discriminated against. Admittedly, discrimination does occur, both in society and in the workplace. And as most attorneys know, many discrimination cases concern claims of either adverse treatment or adverse impact. In both types of litigation, employees believe that they have been discriminated against because of their minority status. In disparate treatment cases, plaintiffs must show that they were treated differently because of that status, and incriminating statements-express or implied-must be admitted as evidence to suggest a discriminatory intent. In contrast, disparate impact cases typically address the discriminatory impact of an ostensibly neutral policy, decision, or program, so plaintiffs rely upon objective data from the entire corporation to prove a discriminatory outcome. In essence, disparate treatment cases often (though not invariably) rely on the admissibility of prior statements or admissions to demonstrate discriminatory intent, whereas disparate impact cases typically rely on statistical analysis of quantitative data to demonstrate a discriminatory outcome that cannot be explained by chance or external societal factors (like gender-based differences in strength, education, etc.) alone. Disparate impact cases, unlike disparate treatment cases, do not require proof of the employer's motive, as International Brotherhood of Teamsters v. U.S. (S.Ct. 1977) shows.

This article is designed as a guide for corporate counsel when refuting an erroneous discrimination claim at a large corporation (i.e., having between one thousand and several hundred thousand employees). How do you formulate specific guidelines for using statistics in such litigation so that you can, if and only if it is indeed justified, prove that your corporation acted properly? How do you know that the expert you are hiring has used the right methods for obtaining and analyzing the data? This article can guide you when working with a statistician or psychologist, from either inside or outside your corporation, as you build your case. These guidelines have made it possible to analyze very large datasets from the workplace¹, using quantitative data from both printed² and electronic³ sources.

A Few Words About Statistical Evidence

Numerous cases have established standards and precedents for the use of statistical evidence in disparate impact litigation. In fact, as Watson v. Fort Worth Bank & Trust (S.Ct. 1988) shows, adverse impact plaintiffs must identify the specific procedure causing the alleged disparity, and "...must offer statistical evidence of a kind and degree sufficient" to show that members of the protected subgroup were negatively impacted by the policy or program in question. That is, plaintiffs must use statistical evidence to prove their case, and must do so with precision, as New York City Transit Authority v. Beazer (S.Ct.1979) shows. Moreover, defendants can insist on holding proof of that impact's cause in their hands during litigation, as Holder v. City of Raleigh (4th Cir. 1989) shows; impact cannot be assumed just because it seems obvious, or logical, or likely.

Yet some experts overlook an important aspect of Watson v. Fort Worth Bank & Trust (S.Ct.1988), which adds to the specificity requirement mentioned above: The plaintiff's burden to establish a prima facie case goes beyond the need to show the presence of specific "statistical disparities." Plaintiffs must also show that those observed disparities were not caused by innocuous or unavoidable factors associated with external forces, as EEOC v. Joe's Stone Crab, Inc. (11th Cir. 2000) subsequently confirmed. That is, to use statistics appropriately, it is necessary to rule out alternative explanations.

The only way to do an adequate job of "... isolating and identifying the specific employment practices that are allegedly responsible for any observed statistical disparities" (as Watson specifies), is to build a comprehensive statistical model of independent variables (also called "predictor variables") and dependent variables (also called "outcome variables"). This model must be compelling enough to withstand scrutiny by academic colleagues, adversarial experts, and decision makers in court. As much research in social psychology shows, any such models must include variables that control for factors such as socioeconomic status, years of education, years of experience, job tenure, skill, and the like. Without the inclusion of these potentially confounding variables (also called "covariates"), none of the analyses will stand up adequately in court.

Selecting an Expert

It is imperative to select a statistical expert who has extensive first-hand experience so that your case need not rely exclusively on findings from published research. Such evidence is vulnerable to a hearsay objection unless it has been read with the benefit of a specialist's expertise, as shown in United States v. Dukagjini (2nd Cir.2003). Accordingly, it is wise to select an expert who combines solid knowledge of published research with practical first-hand experience analyzing workplace data.

Social psychologists are in a particularly good position to help jurors, judges, plaintiffs, defendants, and attorneys by using rigorous statistics to identify and measure the causes of adverse impacts in the workplace. Even a brief description of disparate impact litigation introduces notions pertaining to societal norms, subtle unintended consequences, and the need to distinguish between those two statistically. Social psychologists are uniquely equipped to address these issues, in part because their tradition of using statistics to analyze impacts goes back to the late 1890s.

It was the social psychologist Norman Triplett in 1898 who first quantified the impact of gender, age, and an audience's presence on athletic performance⁴. In a set of carefully controlled analyses he isolated the impact of bystanders on an athlete's bicycling speed. The work is germane in this discussion for one simple reason: Triplett used straightforward statistical analyses to disentangle subtle inter-connected factors (such as age, gender, encouragement, anxiety, and mental fatigue) and to measure their impact on an objectively determined outcome.

Similarly, in disparate impact cases, particular importance attaches to statistical methodology and the complex interaction of social and psychological factors. Now the courts are becoming especially receptive to social psychologists' quantitative statistical approach, in part because of their ability to meet the "will assist" clause of the Federal Rules of Evidence (FRE) 702. Just as Triplett provided a helpful analysis of the factors leading to a win or a loss in races more than a century ago, social psychologists today can use advanced statistical tools to disentangle complex causes leading to a promotion or termination in the workplace.

Social psychologists are also in an exceptionally good position to measure impacts in EEO lawsuits because-ever since Sewall Wright's work developing statistical models for experimental research in 1921⁵-their flagship journals like the Journal of Experimental Psychology, which first appeared in 1916, and their references like Cohen and Cohen's 1983 text on applied multiple regression⁶, have promulgated guidelines ensuring agreement about what it means to run analyses using "sufficient facts or data," using "reliable principles and methods," and "applying those principles and methods reliably to the facts" just as FRE 702 specifies. Such agreement ensures that "statistical validity" is preserved and maximized-just as the Supreme Court required in its 1993 decision Daubert v. Merrell Dow Pharmaceuticals, Inc. (S.Ct. 1993).

Moreover, many social psychologists have experience with large datasets containing literally millions of rows and hundreds of variables. They also typically have experience examining the impact of social-psychological variables in the large complex datasets that businesses conventionally collect on production, absenteeism, pay, training, promotions, selection, and the like-just as FRE 803(6) allows. Their approach is especially consistent with the court's requirement that experts use a clear statistical rule of exclusion to minimize the role of random chance while analyzing an observed disparity, a key element in the discrimination case Hazelwood School District v. United States, (S.Ct. 1977). In addition, just as FRE 902 (11) or (12) requires, they can also run those analyses properly without tipping their hand-even when they provide opposing counsel with a copy of those certified datasets and written notification that they intend to analyze them. In short, social psychologists are eminently appropriate and qualified to serve as expert witnesses in EEO lawsuits by virtue of what FRE 702 requires in the way of "knowledge, skill, experience, training, or education."

A Baker's Dozen: Thirteen Guidelines for Using Statistics to Prove Impact

. . .Continue to read original article (PDF).