June 23, 2014

50 For 50: Five Decades Of The Most Important Employment Discrimination Decisions

Number 39: Statistics In Discrimination Cases

Statistics have been a component of discrimination litigation since Title VII enforcement actions began to catch on.  Normally, statistics arise in a disparate impact discrimination claim – where a facially neutral employment practice falls more harshly on one group of employees than another, and the practice is not justified by any business justification.

In the 1971 case of Griggs v. Duke Power Co., the U.S. Supreme Court held that liability under Title VII does not require a showing that an employer acted intentionally to discriminate.  It could, instead, be based on a showing that an employment screening device, such as a requirement of a high school education or a minimum score on an aptitude test, which is applied to everyone actually disproportionately excludes a group protected by Title VII without the justification of business necessity.  The question was how could this be established.  The answer:  statistics.

Indeed, a prima facie case of this disparate impact discrimination is typically established when a plaintiff identifies a specific employment practice to be challenged and proves through relevant statistical analysis that the challenged practice has an adverse impact on a protected group.  In short, inferences of disparate impact discrimination can be drawn when there is a difference between “actual” results and “expected” results.  To be useful (or useable), however, the observed differences must be statistically significant, which the Supreme Court was concluded to be greater than two or three standard deviations (see, Hazelwood School District v. U.S.).  Upon such a showing, a court can draw a valid adverse inference of discrimination and shift the burden to the employer to produce evidence justifying the observed difference.  A form of this analytical framework was codified in the 1991 Civil Rights Act.  Under that law, an employee can make out a prima facie case of race or gender discrimination based on no more than a statistical disparity between the composition of the employer’s work force and the pool of available workers.  More importantly, to defend itself, the employer must then show that the disparity is job related or consistent with business necessity. Moreover, once disparate impact had been shown by the plaintiffs, the burden on the employer becomes one of persuasion, not merely production.

The use of statistics is far less common in intentional discrimination cases.  The primary explanation is that the relevant analytical pool is typically far smaller.  In the context of an employer’s single decision to hire, promote or terminate, the sample size is likely small.  With smaller sample sizes, the observed results are far more likely to be random as opposed to proof of intentional action.  Thus, statistical evidence has typically been reserved for large employers and class actions.  Some argue, however, that statistical proof should have a greater role than it currently does in such cases.  One major argument here is that statistics can be used to infer employer knowledge.  Specifically, statistics could be used to establish that the employer must have been aware of the exclusionary effect of a particular requirement.  Accordingly, the employer’s continued use of a selection process with a known exclusionary effect is a reckless disregard of rights, and that disregard satisfies the intent requirement and would require the employer to then demonstrate a good faith belief in the validity of the requirement.  It is still difficult to draw meaningful conclusions with small sample sizes, but the refinement of statistical analysis will likely spur continued efforts.

In both disparate impact and disparate treatment litigation, the pedigree of the statistics expert, the construction of the statistical models, the underlying assumptions, secondary sampling and the like can dominate the litigation.  Even with the backdrop of skepticism concerning statistics – lies, damn lies, and statistics – there is still a power of perceived objective analysis such data can provide, especially when, as is frequently the case, direct evidence of discriminatory intent is absent.

Category: EEO,