This article explains why pre-litigation statistical audits of employment processes may be useful, the steps for conducting audits of this nature and examples of how audit outcomes have been utilized by employers.
Why Conduct A Pre-Litigation Statistical Audit?
On March 7, 2007, the Equal Employment Opportunity Commission ("EEOC") issued a press release announcing that it had filed a nationwide class action against Walgreen Co., claiming that it discriminates on the basis of race against African-American employees by assigning them to low-performing stores and stores located in predominantly African-American neighborhoods. One month earlier, in Dukes v. Wal-Mart, Inc., 474 F.3d 1214 (9th Cir. 2007), the Ninth Circuit upheld the certification of what it described as a "broad and diverse" nationwide class comprised of approximately 1.5 million female salaried and hourly employees who worked at one or more of the thousands of Wal-Mart stores across the country since 1998. Id. at 1224. The plaintiffs in Dukes claim gender discrimination in pay and promotions.
These cases highlight the potential exposure that employers face in class action litigation filed by either the EEOC or on behalf of a private class of plaintiffs (or both) claiming broad-based employment discrimination in multiple locations over a multiyear timeframe. The employment practices that may be subject to litigation of this nature include recruitment, hiring, job assignment, promotion, compensation and termination (typically in the context of a reduction in force). While class actions may focus upon a single department, facility or region, nationwide claims appear to be increasing in frequency.
The risk of nationwide or other class action litigation raises the question: What can employers do to assess their potential exposure to liability if a claim of this nature is brought? Since class action employment discrimination suits typically rely heavily upon statistical analysis of the employment practice(s) at issue (central to both the class certification and liability stages of the litigation), the simple answer is to conduct a statistical audit of those employment processes most likely to draw the attention of the plaintiffs' bar or EEOC.
The audit is a tool not only for assessing potential exposure to liability but also for preparing defenses and/or modifying potentially problematic policies or practices well in advance of the actual filing of any claim. A properly conducted audit provides the employer with a preview of the statistical evidence that a class statistical expert could present. Perhaps even more importantly, it may serve as a vehicle for determining whether there are credible explanations for protected group disparities that may exist and addressing any potential problem areas. In the appropriate circumstance, audit results might even be selectively shared with class counsel in the early stages of litigation to demonstrate that an alleged disparity does not exist and that the defendant employer is well-prepared to stand by its practices.
Steps For Conducting A Pre-Litigation Statistical Audit
In order for a statistical audit to have value, it must be properly designed and executed. These are the basic steps: First, one assembles a team normally consisting of representatives from the employer's legal, human resources and information technology departments, plus experienced outside employment counsel and a knowledgeable statistical and/or labor economics expert. It may also be beneficial to include a management representative from the areas of the business upon which the audit will focus. Moreover, it is advisable at the very outset of the process to assess whether the results of the audit can be protected by attorney-client privilege and to take steps to maximize the protection where possible.
Second, the team must decide which employment processes and protected groups to study. An assessment is made of the class claims currently in favor by the plaintiffs' bar (including any that have been filed against competitors), as well as existing or former employee claims raised internally with the employer or externally with government commissions. The audit team may also have information about potential problem areas that would be suitable for study.
Third, corporate policy manuals are reviewed to understand how the employment processes should work in theory and relevant managers are interviewed to understand how those processes work in practice (to the extent that expertise is not present on the audit team). It is important to understand not only how the processes work but also the documents generated and factors considered in making employment-related decisions during the course of the processes being studied.
Fourth, one determines if sufficient relevant data exist to study the subject processes. (The absence of some relevant data does not mean the plaintiff class will abstain from measuring protected group differences based upon whatever data may exist.) In the event that all relevant data does not exist, steps may be taken to generate the data during the course of the study, if practicable, or to implement procedures so that the data is gathered in the future in the ordinary course of business.
Fifth, an analysis plan is designed to answer three basic questions for each process and protected group that is being studied: (a) what is the magnitude and statistical "significance" of the protected group disparity the plaintiffs' expert would likely find and introduce at trial?; (b) to what extent do refinements to the statistical analysis (changes in the way one defines "similarly situated" employees/applicants such as through the inclusion of additional factors) mitigate (or exacerbate) that disparity?; and (c) are either the "gross" or "refined" protected group differences consistent across time, business unit, job group, geography, etc.? (The latter is useful for assessing the issue of "commonality" and for identifying where follow-up effort might best be focused).
It is important to emphasize that the reasons for steps three and four are to permit the study of the relevant factors that may have an important influence upon the statistical result. It is often meaningless to do a bottom-line analysis based upon only protected group membership. Unless the factors that significantly influence process outcomes are included in the analysis, it is simply not possible to determine whether any disparities are the basis of legitimate non-discriminatory factors. Commonly used statistical tools for accommodating such factors are multiple linear regression analysis (for compensation studies) and multiple pools analysis (for employee selection studies).
Finally, the statistical consultant is provided with all salient data and tasked with preparing the agreed-upon studies. The findings are presented to the audit team and a follow-up plan is developed. The manner in which audit findings can lead to productive employer responses is best illustrated by actual examples.
How Audit Outcomes May Assist Employers
The following are a few examples of findings from actual audits and resulting employer actions:
A multiple regression analysis of base pay rates showed a consistent disparity adverse to African-Americans that could only be explained by accounting for current and past performance ratings. The ratings explained the pay differentials but raised the issue of whether the evaluation system was properly designed and implemented. An industrial psychologist was retained to validate the performance rating process.
Females were disproportionately hired into cashier positions as opposed to sales positions (the gateway to management). The standard employment application did not require the applicant to articulate the specific position sought, thus exposing the employer to a claim of stereotypical "job steering." The application form was appropriately modified with the result that those seeking the cashier position were indeed largely female (i.e., a defensive database was created that explained the pattern of actual job assignments). Further, a policy was enacted that prohibited an applicant from being assigned to any job other than that requested.
Adverse pay disparities by gender were substantially moderated by accounting for pay rate at hire, suggesting either a prior experience differential by gender or discrimination in initial pay rates. A sample of employees was drawn for the purpose of building a prior experience database from resumes and job applications. Starting salary differences by gender were fully explained by those prior experience measures and, consequently, a process for the capture of prior experience information for all employees was implemented on a going-forward basis.
An analysis of promotions to management found substantial female promotion shortfalls. It was theorized that females tended not to seek such positions due to a "willingness to relocate" requirement (or that decision-makers held that possibly stereotypical view). A post-and-bid system was installed (whereby all job vacancies were communicated to employees and interested parties formally bid on those vacancies). In addition, the "willingness to relocate" requirement was reviewed and ultimately modified.
A compensation study showed a substantial non-white shortfall in one job category (Business Analyst). The statistician prepared a listing of all incumbents in that job showing the salary rate predicted from the statistical model, the actual salary rate and the consequent salary rate difference. Follow-up work with that list revealed that those appearing to be paid above the model prediction tended to have job assignments requiring high-level computer skills (and tended to be white). The employer decided to abandon the overly broad Business Analyst title and create two different job titles reflective of the different types of work being performed by these incumbents. A subsequent analysis of the data on this modified basis showed no meaningful race disparity.
A reduction in force based on a forced ranking process was to be implemented in the coming weeks. A workforce "snapshot" data file was provided to the statistician indicating peer group, race, gender, age and whether tentatively selected for inclusion in the RIF. Adverse impact studies indicated an over-selection by age-protected status in one business unit. Drill-downs to the peer groups in that business unit allowed the audit team to interview the relevant decision-makers to ensure that selections were not age-motivated (interview notes were archived with in-house counsel). Where no clear rationale for a selection was forthcoming, the selection was either negated or submitted for examination by the business unit manager.
It is important to emphasize that a pre-litigation statistical audit is not required by law and may be a useful tool for some but not all employers. However, for certain employers, the audit can be a valuable risk management tool that:
quantifies the type of evidence a plaintiff class will ultimately find within the employer's data;
allows exploration of alternative avenues of statistical rebuttal to the class evidence;
leads to a better understanding of how employment processes work in practice;
identifies facially neutral practices that have adverse impact and may require validation;
leads to new areas of data collection/capture that may be critical to litigation defense; and
suggests important procedural and policy changes.
Jonathan D. Wetchler is a Partner in the Philadelphia office of WolfBlock, where he counsels clients concerning all aspects of employment law. David W. Griffin, Ph.D. is a Director in the Philadelphia office of LECG, an international economic consulting firm. For the past 25 years he has been a statistical consultant to the legal community on employment discrimination matters and has testified numerous times as an expert witness in federal and state courts.