A RETROSPECTIVE DIVE INTO PUFFER V. ALLSTATE
by Eric Dunleavy Ph.D., Principal Consultant, DCI Consulting
As Art described in an earlier post, the recent 7th circuit ruling in Puffer v. Allstate emphasized some nuanced class certification issues and the differentiated disparate impact claims from pattern or practice claims. In reviewing the three judge panel decision, we felt it was worth going back to the original ruling from Magistrate Judge Schenkier in January of 2009. Of particular interest to us are the merit aspects of the ruling directly related to pay equity analysis, which were given substantial weight in the decision to deny the plaintiff’s motion for class certification under a pattern or practice theory of discrimination.
The battle of the experts was complex, with labor economists focusing on pay disparities, I/O Psychologists focusing on the likelihood of bias in employment decision processes, and sociologists considering whether the culture of the company could foster discrimination. We focus here on the statistical analyses of pay disparity. Judge Schenker considered the limitations of the plaintiff expert analysis to be serious enough to affect class certification. An initial issue of interest relates to constructing similarly situated employee groupings for analysis, which can affect commonality in class certification consideration. According to Judge Schenkier:
“To satisfy the commonality requirement, Allstate’s employment policies or practices must have affected the putative class members in the same or similar ways……Allstate argues that plaintiff cannot show commonality here, because “[t]he staggering heterogeneity of plaintiff’s proposed class requires answering questions about decisions made by hundreds of managers, across varied business divisions and segments, with employees who perform entirely different functions, who hold hundreds of different jobs, and who are scattered across a multitude of regions and office locations” (Def.’s Opp’n to Class Cert. at 25-26). We agree.
………As explained above, the putative class members were supervised and reviewed by many different people; had a wide variety of salary levels, jobs, and responsibilities; and had annual goals individually tailored to them.
………Some courts have held that a conflict of interest may arise where a class contains both supervisory and non-supervisory [**55] employees, and the supervisors implemented the employment system which the class litigation challenged.
A second issue of interest is how judges interpret simple statistical indicators. Judge Schenkier suggested caution:
“While we are mindful of the importance of statistical evidence, we likewise must acknowledge that “[i]t is said that you can find a statistic to prove anything.” As Mark Twain once said, “[t]here are three kinds of lies–lies, damned lies and statistics.” Id. Thus, the Court recognizes both the potential power of statistics, and their potential frailty. [HN11]……”Gross statistical disparities” alone may constitute prima facie proof of a pattern or practice of discrimination in a proper case, but statistical evidence alone is rarely sufficient. Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 307-08, 97 S. Ct. 2736, 53 L. Ed. 2d 768 (1977).”
However, Judge Schenkier also seemed to place substantial weight on the appropriateness of the multiple regression analyses to determine whether disparities existed across group after accounting for legitimate factors that may affect pay. More specifically, the plaintiff expert analysis was criticized along a number of dimensions, including the following:
“For example, “[s]tatistical evidence which fails to properly take into account nondiscriminatory explanations does not permit an inference of discrimination.” Radue v. Kimberly-Clark Corp., 219 F.3d 612, 616-17 (7th Cir. 2000). Although an expert need not include all measurable variables [**27] in conducting a statistical analysis, see Bazemore v. Friday, 478 U.S. 385, 400, 106 S. Ct. 3000, 92 L. Ed. 2d 315 (1986), certain factors are crucial to statistical evidence in Title VII cases, such as identifying not only those individuals who are qualified for a position, but also those who are potentially interested in it. Bennett, 295 F.3d at 697. The Seventh Circuit has explained that “[i]n order to be considered, the statistics must look at the same part of the company where the plaintiff worked; include only other employees who were similarly situated with respect to performance, qualifications, and conduct; [and] the plaintiff and the other similarly situated employees must have [*462] shared a common supervisor. . . .” Balderston v. Fairbanks Morse Engine Div. of Coltec Indus., 328 F.3d 309, 319-20 (7th Cir. 2003). “[W]hen the statistical evidence does not adequately account for the diverse and specialized qualifications necessary for the positions in question, strong evidence of individual instances of discrimination becomes vital to the plaintiff’s case.” E.E.O.C. v. Sears, Roebuck & Co., 839 F.2d 302, 311 (7th Cir. 1988) (internal citations and quotations omitted).
……..Dr. Madden performed a multiple regression analysis using five different models, controlling for various combinations of age, tenure at Allstate, education, salary grade, job code, location, and time in grade. The Court finds limited probative value in Dr. Madden’s report on the issue of class certification.
…….Plaintiff concedes the possibility that evidence may show “that the measured gender differences can be explained in terms of legitimate measures of skills or productivity” (Pl.’s Class Cert. Mem. at 7 (citing Madden Report at 21)). Dr. Madden reiterates in her rebuttal report that gender differences in salary are not evidence of discrimination unless [*466] productivity and other characteristics associated with productivity have been “appropriately controlled” (Madden Rebuttal at 6-8, 13). While Dr. Madden states that she “investigated all of the data that are available on productivity,” that does not include employees’ major [**42] job responsibilities, individual performance goals, Performance Development Summaries, Quality Leadership Management Surveys, Critical Success Factors, or any other actual measure of the quantity or quality of an employee’s performance (id. at 8).
……..Although plaintiff is not required to include all measurable variables in her statistical analysis, Dr. Madden’s analysis is only minimally probative of commonality without these important variables. See Sears, 839 F.2d at 334 (upholding district court’s finding that the plaintiff’s statistics were flawed because they were based on the false assumption that all plaintiffs were equally qualified and equally interested in the promotions at issue); see also Radue, 219 F.3d at 616-17 ( [HN15] “Statistical evidence which fails to properly take into account nondiscriminatory explanations does not permit an inference of discrimination”). 13 For this reason, together with the ones we have already discussed, Dr. Madden’s reports do not show commonality in plaintiff’s putative class.”
In addition, the issue of analyzing discrete pay decisions versus analyzing overall pay metrics (e.g., base salary) was also considered by Judge Schenkier. For example:
“Dr. Madden originally calculated what the Court will refer to as the “total difference” between male and female employees’ salary each year, rather than the “incremental difference” in compensation each year. For example, if in 2001, men in a certain salary grade earned $ 100 and women of the same grade earned $ [**31] 90, the “total difference” in compensation is $ 10. If in 2002, the men and women both received a 10% increase in salary (such that the men now earned $ 110 and the women now earned $ 99), the “total difference” in compensation would increase to $ 11. The “incremental difference” in compensation from 2001 to 2002, however, is only $1. Dr. Madden’s original method of regression analysis thus included [*463] the effect of wage differentials that existed between male and female employees as a result of employment decisions that predated the class period sought by plaintiff.
That method of analysis runs headlong into Ledbetter v. Goodyear Tire & Rubber Co., Inc., 550 U.S. 618, 127 S. Ct. 2162, 167 L. Ed. 2d 982 (U.S. 2007). [HN12] In Ledbetter, the Supreme Court held that each paycheck issued after a prior discriminatory pay decision does not constitute an actionable act of discrimination unless the individual paycheck was accompanied by new discriminatory intent. Id. at 2174-75. See also Lewis v. City of Chicago, 528 F.3d 488, 492 (7th Cir. 2008) (distinguishing between injury resulting from a “fresh act” of discrimination and injury resulting from the “automatic consequence” of an earlier act of alleged discrimination). By basing [**32] her analyses on “total difference” between salaries of male and female, Dr. Madden improperly incorporated pay-setting decisions that took place before the class period, and thus used an analysis that is contrary to Ledbetter. An “incremental difference” analysis, on the other hand, avoids this problem by focusing on pay decisions made within the asserted class period.”
One other point is worth noting. Judge Schenkier did not find sociological research from the plaintiff’s expert related to the quality of subjective decision making to be persuasive. More specifically:
“Allstate has filed a motion to strike Dr. Reskin’s report under Daubert (doc. # 207). We need not decide that question because, even assuming that Dr. Reskin’s report passes muster under Daubert, we do not find it persuasive on the issue of commonality. Most of Dr. Reskin’s opinion relies on laboratory studies, “experimental research,” and various historical trends and ideas which lead her to the conclusion that subjective decisionmaking is inherently flawed: a conclusion [**50] that flies in the face of the governing case law. See Blise, 409 F.3d at 868 (holding that subjective decision making plays an important and legitimate role in employment decisions).”
Obviously there is a lot to chew on here. The goal of this blog was to highlight sections of the ruling that seem applicable to recent OFCCP pay equity investigations alleging a pattern or practice of discrimination in pay. While some aspects of the ruling were a bit surprising to us (e.g., separating discrete pay outcomes from base pay, requiring similar probabilistic disparities (i.e., standard deviations) across units of analysis), most of the ruling is consistent with basic Title VII Principles that anchor the OFCCP’s 2006 Compensation Standards (e.g., similarly situated groupings and fully specified regression analyses). These principles are intuitive and have been around for a long time. We continue to wonder why OFCCP appears to be moving forward with rescission of those Standards.