Using Statistics Effectively in Wage and Hour Litigation: An Employer’s Offensive and Defensive Tactics

The plaintiffs’ bar has increasingly used class action litigation in recent years to reap large damages from employers. That increase has not been lacking in wage and hour litigation. Both the collective action under the Fair Labor Standards Act (“FLSA”) and the state-law-based class action have become powerful tools for employees to litigate wage and hour claims.

Employees may bring federal wage and hour claims under the FLSA. A plaintiff suing on FLSA claims may seek certification of a collective action of “similarly situated” employees, who “opt in” to the lawsuit after certification is granted.[1] Employees may also bring claims against an employer for violating state wage and hour laws. In contrast to the FLSA collective action, Federal Rule of Civil Procedure 23 and equivalent state class action rules allow a plaintiff to pursue a class action if certain prerequisites are met, including (1) numerosity of class members, the presence of common questions of fact or law, the typicality of the representative members’ claims in comparison to the class, and adequacy of class counsel, plus, usually, (2) the predominance of common questions of fact and law, and the superiority of the class action to other methods of adjudication.[2] Other plaintiffs do not “opt in” to a Rule 23 class action – instead, they “opt out” after receiving notice of the litigation.

Within both collective action and class action litigation, plaintiffs’ attorneys commonly use statistics to avoid issues of individual proof and to establish common liability at the class certification stage. The use of statistics in this context refers to the surveying of employee experiences – including job requirements, activities performed throughout the workday, wage and payment details, hours spent working, management practices, etc. – and the analysis of the results of those surveys.

While class litigation and the use of statistics within class litigation have grown increasingly over the last decades, recent Supreme Court precedent has somewhat hampered that growth. In Wal-Mart Stores, Inc. v. Dukes, the Supreme Court reversed certification of a nationwide class of 1.5 million female employees who alleged sex discrimination.[3] The proposed method of analyzing class claims, approved of by the 9th Circuit, was as follows:

A sample set of the class members would be selected, as to whom liability for sex discrimination and the backpay owing as a result would be determined in depositions supervised by a master. The percentage of claims determined to be valid would then be applied to the entire remaining class, and the number of (presumptively) valid claims thus derived would be multiplied by the average backpay award in the sample set to arrive at the entire class recovery – without further individualized proceedings.[4]

Writing for the Court, Justice Scalia “disapprove[d] that novel project,” emphasizing that “a class cannot be certified on the premise that Wal-Mart will not be entitled to litigate its statutory defenses to individual claims.”[5] Against this backdrop, courts around the country have equipped employers with methods of combatting class litigation through statistical analysis.

An Employer’s Defense: Opposing Statistical Models

In the wage and hour context, several recent decisions have undermined plaintiffs’ use of statistics in class litigation. Duran v. U.S. Bank involved wage-and-hour claims brought as a class action under California’s unfair competition law.[6] The plaintiffs claimed U.S. Bank misclassified 260 loan officers as exempt from overtime payments. Interestingly, the plaintiffs in Duran employed the same expert as in Dukes and attempted to use statistical sampling beyond certification to prove classwide liability. The California trial court permitted the plaintiffs to prove liability and damages on behalf of the entire 260-member class using a small sample of 19 class members and two named class representatives. Even more troubling, the trial court refused to allow U.S. Bank to present testimony of employees who claimed they spent more than 50 percent of their time on exempt duties. Based on the testimony of the sample group alone, the trial court determined U.S. Bank misclassified every class member. Damages were then approved based on a calculation derived from the sample group, leading to a $15 million award with interest.

The California Supreme Court reversed and ordered the class decertified. According to the court, the statistical analysis employed caused a “manifest” injustice to U.S. Bank and was “profoundly flawed.” Duran advised that “[t]he sample relied upon [to prove liability or damages in wage and hour litigation] must be representative and the results obtained must be sufficiently reliable to satisfy concerns of fundamental fairness.” The court provided three reasons why the sampling employed by the plaintiffs did not meet this criteria:

1. A sample size of 19 class members and two named representatives was too small relative to the variability of the class members. As explained by the court, “[i]t is impossible to determine an appropriate sample size without first learning about the variability in the population.” Variability – or the differences that exist in the total population – can be determined by an expert using existing data, time sheets, other personnel records or surveys. Ultimately, it is important to remember that sample size cannot be random – it must be based on the population’s distribution.
2. The statistics were plagued by nonresponse and selection bias. Nonresponse bias occurs where individuals who receive the survey but fail to answer differ in significant ways from those who participate. Selection bias occurs where individuals are selected by the survey administrator to be included or excluded from the survey. These biases cause the results to be unreliable. It is best to ensure participants are truly randomly selected (e.g., by a computer) and that an expert analyzes the data to ensure that nonrespondents do not differ meaningfully from those who do respond.
3. The plaintiffs’ statistical model was plagued by a high margin of error, as is common with small sample sizes. Such a high margin of error renders the results unreliable. As Duran noted, “the court must determine [with the help of experts] that a chosen sample size is statistically appropriate and capable of producing valid results within a reasonable margin of error.” Only then will the court meet its burden of ensuring that the proposed methodology will produce reliable results. To avoid an erroneously high margin of error, it is again important to ensure that the statistical model is appropriately developed with a proper sample size.

The first reason that Duran found the plaintiffs’ statistical model inappropriate – sample size – led to the decertification of a class by another court. In Espenscheid v. DirectSat USA, L.L.C.,[7] the 7th Circuit Court of Appeals affirmed decertification in a case alleging FLSA and state law wage violations. The proposed class consisted of 2,341 technicians employed by DirectSat to install and repair home satellite dishes, who alleged they were not paid for all hours worked and were unpaid for overtime. But the technicians were paid on a piece-rate system, so their work hours each week varied. They also performed different tasks and, thus, the variability of the class members was great. Ignorant to this variability, the plaintiffs proposed sampling of a mere 42 members of the class. As explained by the court, “[t]o extrapolate from the experience of the 42 to that of the 2,341 would require that all 2,341 have done roughly the same amount of work, including the same amount of overtime work, and had been paid the same wage. . . . No one thinks there was such uniformity.”[8] Based on the unreliability of plaintiffs’ proposed sample, the trial court determined decertification was appropriate, and the 7th Circuit affirmed.

Similarly, a federal district court in California rejected plaintiffs’ statistical methodologies at the class certification stage in Sirko v. IBM Corp.[9] The plaintiffs there sought to certify a Rule 23 class of exempt IT employees who were allegedly misclassified and denied overtime by IBM. In an effort to gain class certification, the plaintiffs concocted a 47-question survey concerning putative class members’ work duties. The court rejected the survey, determining it “lack[ed] basic indicators of reliability.” Specifically, the survey: (1) was devised and administered by plaintiffs’ counsel, not a statistician or expert; (2) included some questions that required non-binary answers, rather than a simple “yes” or “no,” which are not easily quantifiable through statistics; and (3) the survey likely included biased results since respondents were provided a cover letter that noted their potential ability to recover damages in the class action.

These cases provide helpful insight to employers who are confronted with statistical models provided by plaintiffs. Employers should consider the following factors in determining whether to combat the proposed models: (1) sample size used; (2) presence of nonresponse bias; (3) presence of selection bias; (4) potential for large margin of error; (5) whether the model was created by counsel or a nonexpert; (6) whether the model calls for non-binary responses; and (7) whether the survey will be or was provided with a cover letter that describes the potential for classwide payouts.

An Employer’s Offense: Using Statistics Affirmatively

While the above cases provide grounds for employers to object to the improper use of statistics by plaintiffs, reliable statistical models can greatly benefit defendants when opposing wage and hour class certification. In Zivali v. AT&T Mobility, LLC, the Southern District of New York granted the employer’s motion for decertification where statistical evidence was used to show varying experiences, ultimately proving the purported class members were not similarly situated as required for a collective action.[10] There, plaintiff Zivali brought claims on behalf of herself and others similarly situated, alleging her employer failed to pay wages and overtime compensation in violation of the FLSA and New York Labor Law. Specifically, Zivali asserted AT&T’s timekeeping system failed to capture all hours worked due to off-site work activities and opening and closing tasks performed in-store. AT&T effectively used statistics and surveying to oppose class certification. First, AT&T introduced statistics of timekeeping data, which indicated that the distribution of management edits varied significantly across the stores. This helped AT&T to prove the plaintiffs’ experiences varied since their managers applied the time-editing policy inconsistently. Phone record statistics were also beneficial. AT&T surveyed phone records from a sample of employees to analyze the use of phone calls for both business and personal purposes. The statistics established that potential off-the-clock activity ranged from de minimus values (mere minutes) to over two hours.

Likewise, in Tracy v. NVR, Inc., the employer used statistics of its employees to persuade the Western District of New York to decertify an FLSA collective action.[11] There, the plaintiff brought claims on behalf of himself and other similarly situated sales representatives, alleging NVR violated the FLSA and New York Labor Law in failing to properly compensate employees for overtime. The statistics established (1) NVR’s employees performed work outside the office with varying frequencies and for varying time periods, and (2) the performance of work activities outside the office by even the same employee varied widely. These results helped NVR establish that employees retained discretion in connection with out-of-work activities, and therefore NVR’s defense would necessarily be highly individualized.

Moreover, especially since the Supreme Court’s proclamations in Dukes, courts have become skeptical of requests for nationwide classes encompassing employees of multiple positions who work in different locations, under different supervisors and in different practice areas. Using statistics to show variance with respect to these employee-specific job characteristics will also support an argument that a class should not be certified. At the least, such statistics may cause a court to require a narrowed class description.

Concluding Remarks

Wage and hour litigation is certainly increasing. Meanwhile, class actions are also increasing, and therefore the use of statistics by plaintiffs to establish the appropriateness of a class model for both liability and damages is increasing. Employers need to understand how they can combat unreliable statistical models that may lead to erroneously large damages awards. It is equally important for employers also to recognize how the use of statistics effectively can prevent class litigation entirely. Keeping in mind the courts’ lessons about combating plaintiffs’ statistical models and using reliable statistics affirmatively will go far in evading class liability and damages in wage and hour litigation.