Opinion
Case No. 18-CV-00133-NJR
04-26-2021
DONALD CONRAD, On Behalf of Himself & All Others Similarly Situated, Plaintiff, v. JIMMY JOHN'S FRANCHISE, LLC, JIMMY JOHN'S ENTERPRISES, LLC, and JIMMY JOHN'S LLC, Defendants.
MEMORANDUM AND ORDER ROSENSTENGEL, Chief Judge :
Plaintiff Donald Conrad asks the Court to reconsider its February 24 Memorandum and Order for four reasons. Unfortunately for Conrad, however, none of his reasons is persuasive. So for the reasons explained below, the Court denies his Motion for Reconsideration.
BACKGROUND
The Court incorporates the Background section of the February 24 Memorandum and Order (Doc. 223). Conrad's specific objections are set forth below.
LEGAL STANDARD
"Judges are permitted to reconsider their rulings in the course of a litigation." In re Text Messaging Antitrust Litig., 630 F.3d 622, 627 (7th Cir. 2010). But "[t]he district court's 'opinions are not intended as mere first drafts, subject to revision and reconsideration at a litigant's pleasure.'" A&C Constr. & Installation, Co. WLL v. Zurich Am. Ins. Co., 963 F.3d 705, 709 (7th Cir. 2020) (quoting Quaker Alloy Casting Co. v. Gulfco Indus., Inc., 123 F.R.D. 282, 288 (N.D. Ill. 1988)). A motion for reconsideration, therefore, "is not a forum to relitigate losing arguments; it may be granted only if the movant can 'demonstrate a manifest error of law or fact or present newly discovered evidence.'" Ohr v. Latino Express, Inc., 776 F.3d 469, 478 (7th Cir. 2015) (quoting Anderson v. Catholic Bishops of Chi., 759 F.3d 645, 653 (7th Cir. 2014)).
ANALYSIS
I. Argument 1: "The Court mistakenly thought Dr. Singer's summaries of the WSR data were actually the results of his regressions."
In its February 24 Memorandum and Order, the Court referred to a chart from Dr. Singer's report that showed how the average manager was reflected in the WSR data as earning [Redacted] per-hour, or almost [Redacted] per year—far below the actual average wage of about [Redacted] . (Mem. & Order at 45 (citing Singer Report at 34, Doc. 115-3)). Conrad objects to how the Court referred to this as a "finding" of Dr. Singer's regression. For example, the Court said that "the regression yields an average of [Redacted] " (Id. (emphasis added)). Instead, Conrad notes that the chart reflects "the wage data as it was produced by Jimmy John's." (Pl.'s' Mot. at 13). He also asserts that the Court contradicted itself by not passing on the quality of the WSR data while also excluding Dr. Singer's testimony because of the switcher problem. (Id. at 17). The Court disagrees.
Simply put, the Court did not misapprehend Dr. Singer's report. True, the chart refers to the raw data: it is not the product of Dr. Singer's regressions. Yet the switcher problem remains. Conrad agrees that no one thinks the average manager was paid [Redacted] per year. (Pl.'s Reply at 2). Indeed, as Dr. Ordover explained, that is not what the data reflects:
The WSR data used by Dr. Singer is consistent with Mr. Conrad's testimony and shows Mr. Conrad's wage rate switched from [Redacted] per hour to being recorded as being paid [Redacted] per shift in May 2018. However, Dr. Singer wrongly heats Mr. Conrad as being paid on an hourly basis throughout the time period—so Mr. Conrad's supposed 'hourly' wage rate increases from [Redacted] to [Redacted] in Dr. Singer's analysis over a period of a few days.(Ordover Report at 13, Doc. 133-56 (emphasis in original)). Dr. Singer made this "same error for other managers" whose wages changed from per-hour to per-shift during the period. (Id.). Take another example: "the data on compensation of Mr. Gerald Acosta, a manager at Jimmy John's store 2322." (Ordover Rebuttal at 11, Doc. 195-37).
Mr. Acosta worked at this location from October 2015 through June 2019 and for each week over that period the WSR data used by Dr. Singer reports that Mr. Acosta was paid [Redacted] . However, due to differences in whether Mr. Acosta's time was recorded in shifts or hours (and when in hours, the variation in the number of hours worked), Dr. Singer's calculated wage rate for Mr. Acosta ranges from [Redacted] to [Redacted] per "hour."(Id.).
Dr. Singer makes the same error for other managers that were paid a salary, which was recorded on a per-shift basis in the WSR. . . . This means that the estimated relationships between the wage rates and the independent variables, such as the presence of the No Poach Provision, will be mis-measured.(Ordover Report at 13). So the switcher problem is not merely an issue of faulty data. See Manpower, Inc. v. Insurance Co. of Pennsylvania, 732 F.3d 796, 809 (7th Cir. 2013). Rather, it marks a systemic failure of Dr. Singer's "models fail to adjust for those two percent of WSRs that do not consistently record employee wages as per-shift or per-hour." (Mem. & Order at 40). Small as it seems, that two percent of employees comprise 25 percent of managers. (Ordover Rebuttal at 11). Dr. Ordover recognized this and aptly demonstrated how "'separating Dr. Singer's regression by manager pay type results in a finding that managers paid on an hourly basis had an average wage suppression of approximately two percent, while salaried managers suffered no suppression at all.'" (See Mem. & Order at 13 (quoting Ordover Report at 21, Doc. 133-56)). Conrad wrote off the switcher problem as "'mathematically irrelevant'" and moved on. (See id. at 44 (quoting Singer Rebuttal at 28, Doc. 185-2)). But "on issues affecting class certification," the Court need not "simply assume the truth of the matters asserted by the plaintiff." Bell v. PNC Bank, Nat'l Ass'n, 800 F.3d 360, 377 (7th Cir. 2015). Indeed, "[t]he 'rigorous analysis' requirement 'applies to expert testimony critical to proving class certification requirements.'" Howard v. Cook Cty. Sherriff's Office, 989 F.3d 587, 601 (7th Cir. 2021) (quoting In re Blood Reagants Antitrust Litig., 783 F.3d 183, 187 (3d Cir. 2015)). Conrad's argument amounts to semantics aimed to distort the focus of the Court's opinion.
Conrad's argument that the switcher problem is benign because Dr. Singer's models reveal impact in percentage terms—rather than dollars and cents—is similarly unavailing. As Jimmy John's ably explains in its brief, "[e]xpressing the relationship in percentage terms (i.e., in log terms) would not eliminate the measurement errors created by recording [wages] in two different metrics." (Jimmy John's Resp. at 6 (describing how if, for example, one were to record snowfall in inches one day and centimeters the next, then calculating the percentage change in snowfall without adjusting for the different metrics would still lead to erroneous results)). At any rate, Conrad never raised this argument before, so it is not properly before the Court now. See Caisse Nationale de Credit Agricole v. CBI Indus., Inc., 90 F.3d 1264, 1270 (7th Cir. 1996) ("Reconsideration is not an appropriate forum for rehashing previously rejected arguments or arguing matters that could have been heard during the pendency of the previous motion.").
II. Argument 2: "Dr. Singer's regressions do not yield inflates estimates of antitrust impact."
Conrad claims that "[t]he Court overlooked thirteen pages of Dr. Singer's explanation why there was no econometric or statistical reason to suspect that these 2% would skew his results." (Pl.'s' Mot. at 14). But in those pages, Dr. Singer conceded that the switcher problem was not accounted for by the "worker fixed effects," (Singer Rebuttal at 24, Doc. 185-2); characterized it as a mere "[r]andom measurement error," (id. at 25); and used a graph, not specific to this case, purporting to show that the switcher problem is "effectively washed away when the regression has a large number of data points," (id. at 25-28). The Court rejected those arguments, recognizing that the potential for error caused by the switcher problem was heightened because it affected a significant subsection of employees (managers). (Mem & Order at 44-45).
Now, on reconsideration, Conrad presents another expert report, this time in the guise of a "supplemental declaration." He contends that Dr. Singer excluded the switcher data, redid his regressions, and confirmed his original finding of antitrust impact. Thus, Conrad asks the Courts to consider this new report and reverse course.
Consideration of Dr. Singer's supplemental declaration is inappropriate because it is not newly discovered evidence. Jimmy John's first raised the switcher problem in its July 2020 Motion to Exclude. Dr. Singer then prepared a rebuttal report, where he argued that it was a random measurement error. Conrad could have shown that then; but instead, he tried meeting his burden by relying on general principles. At the hearing, the Court specifically asked Conrad's counsel about the switcher problem; he summarily claimed that the two-percent measurement error was harmless. (Tr. at 31-33). He also conceded that the "fixed effects" only control for factors fixed over time. (Id.). Yet that would exclude 25 percent of managers whose wages were not consistently recorded as per-shift or per-hour. Conrad wants a redo by submitting a supplemental declaration, but the Court already gave him a fair chance to state his case. He failed to meet his burden.
Along those lines, the litigants, the public, and the Court all share an interest in finality. Naturally, Jimmy John's disputes the assertions in Dr. Singer's supplemental declaration. Indeed, Jimmy John's raises several arguments challenging his "fixes," paving the way for a Daubert challenge within a Daubert challenge. Adding to the confusion, this dispute is linked to a motion for class certification filed in December 2019 that relies heavily on Dr. Singer's first report—the one that the Court excluded. More active management is necessary now "to ensure that the certification decision is not unjustifiably delayed." Fed. R. Civ. P. 26(a) advisory committee's note 2003 amendment; see, e.g., Kempner Mobile Elecs., Inc. v. S.W. Bell Mobile Sys., 428 F.3d 706, 713 (7th Cir. 2005) (affirming denial of leave to file expert rebuttal report because it "would have a prolonged delay in the already protracted proceedings requiring additional discovery and increased costs"). The Court will not consider this not-so-new evidence.
III. Argument 3: "In the alternative, the appropriate answer to the Court's concerns is to exclude the 2% of switchers, not the entire opinion."
Conrad argues that the Court should not exclude Dr. Singer's report in whole cloth. Rather, he hopes to salvage "Dr. Singer's separate, stand-alone regressions for in- shoppers and drivers that did not use manager data, or the damages methodology." (Pl.'s' Mot. at 6). But those regressions alone cannot establish that common issues predominate over all Jimmy John's employees, including managers, as presented in Conrad's Motion for Class Certification. And when confronted with expert testimony, "the trial judge must determine at the outset" whether the "methodology can be applied to the facts in issue." Daubert v. Merrell Dow Pharms., Inc., 509 579, 923-93 (1993). Yet because Conrad himself would be excluded from the results, a piecemeal report cannot establish that antitrust impact is capable of proof through evidence that is common to the class.
In a footnote, Conrad suggests redefining the class. (Pl.'s' Reply at 5 n.7). He may very well choose to do so, but not through a motion for reconsideration in a Daubert challenge. --------
IV. Argument 4: "The Court applied contradictory legal standards to Dr. Ordover and Dr. Singer."
Conrad previously argued that Dr. Ordover erred by not controlling for county-level economic conditions. The Court rejected that argument because "'the exclusion of major variables or the inclusion of improper variables may diminish the probative value of a regression model'" but "'do not generally preclude admissibility.'" (Mem. & Order at 50 n.7 (quoting In re Urethane Antitrust Litig., 768 F.3d 1245, 1260-61 (10th Cir. 2014))). Now, Conrad points to a variable that Jimmy John's expert, Dr. Ordover, added to his regression to control for the switcher problem. He says that the Court contradicted itself by citing the same case when it chose not to exclude Dr. Ordover's over Conrad's objection about Dr. Ordover's dummy variable. Relatedly, Conrad says that it was contradictory for the Court to exclude Dr. Singer's report because he "omit[ed] from his own regression the new endogenous variable that Dr. Ordover created." (Pl.'s Mot. at 17).
To that end, Conrad rehashes the "endogeneity bias" argument. He again says that Dr. Ordover's models are unreliable because he "created a new right-hand side variable attempting to guess the unit of measurement." (Pl.'s' Mot. at 19). The Court rejected that argument, noting that Conrad provided "scant proof that the variable Dr. Ordover created by unpooling the wage data is merely a function of the dependent variable." (Mem. & Order at 47). Again, Conrad says Dr. Ordover's "fix" for the switcher problem was flawed because his "additional 'control' variables . . . are mechanically determined by the value of the left-hand-side variable, injecting circularity or endogeneity." (Pl.'s' Reply at 2). Again, the Court disagrees.
As alluded to in the Memorandum and Order, endogeneity bias "occurs when the dependent and independent variables affect each other." See John E. Lopatka, Economic Authority and the Limits of Expertise in Antitrust Cases, 90 Cornell L.R. 617, 698 n.498 (2005). It underscores the adage "Correlation does not equal causation." When a model suffers from endogeneity bias, it shrugs the possibility that an unobserved factor may be driving any correlation between the independent and dependent variables. To account for endogeneity bias, statisticians neutralize those unobserved factors to avoid conflating correlation with causation. Here, Dr. Ordover adjusted for the switcher problem by adding a dummy variable to the right-hand-side of the equation. Because the WSR data does not specify whether a given wage was per-shift or per-hour, Dr. Ordover inferred that wages below [Redacted] were per-shift, and that wages above [Redacted] were per hour (the few data points in between being discarded to avoid mistake). If the wage was recorded per-shift, then the dummy variable turned to zero; if the wage was recorded per-hour, then the dummy variable turned to one. Dummy variables like this are useful because they enable the use of a single regression to represent multiple groups of people.
Dr. Singer states that endogeneity bias arises "when one or more of the purportedly 'independent' variables is jointly determined along with the dependent variable." In this case, Dr. Singer claims that this endogeneity bias is due to a supposedly deterministic, formulaic relationship between the dummy variable that I used to capture differences in how managers' pay is recorded and the level of each manager's wage rate
* * *
Dr. Singer attempts to demonstrate endogeneity bias by means of a simple analogy where daily temperature is measured using either the Fahrenheit or Celsius scale. But that analogy underscored why his concern and attack are unfounded; for while there is a known and stable, formulaic relation between those two temperature scales, there is no such stable, formulaic relationship in the data between whether a manager is paid hour (or is salaried) and his wage rate.
* * *
To illustrate the issue, in my initial report I discussed the data on compensation of Mr. Gerald Acosta, a manager at Jimmy John's store 2322. Mr. Acosta worked at this location from October 2015 through June 2019 and for each week over that period the WSR data used by Dr. Singer reports that Mr. Acosta was paid [Redacted] . However, due to differences in whether Mr. Acosta's time was recorded in shifts or hours (and when in hours, the variation in the number of hours worked), Dr. Singer's calculated wage rate for Mr. Acosta ranges from [Redacted] to [Redacted] per "hour." The variation in Mr. Acosta's "wage rates" as calculated by Dr. Singer is economically meaningless, and yet is used to estimate the alleged "wage suppression" caused by the No Poach Provision. This is obviously nothing like Dr. Singer's temperature analogy, in which knowing that it is 32° Fahrenheit on a day allows one to know that it is 0°Celsius. In
other words, knowing that Mr. Acosta was paid on a shift basis when his wage was recorded as [Redacted] does not tell us his hourly rate.(Ordover Rebuttal at 14). The Court still agrees with this assessment: Dr. Ordover's models do not suffer from endogeneity bias because there is not a lockstep relationship between an employee's wages and whether that employee was paid per-hour or per-shift.
To illustrate, consider the distinguishable case that Conrad cites for support, In re Polypropylene Carpet Antitrust Litigation, which involved a supposed price-fixing scheme orchestrated by several carpet manufacturers. 93 F. Supp. 2d 1348, 1351 (N.D. Ga. 2000). At class certification, an expert for the putative class used "multiple regression analysis" to "identify any difference between the actual prices of polypropylene carpet and the forecasted competitive prices during that period." Id. Rather than using total manufacturing cost as an independent variable, however, the expert instead used the ratio of the price of carpet to fiber costs as the dependent variable. Id. As a result, the defendants claimed that the expert's analysis suffered from endogeneity bias because it assumed that increases in fiber costs corresponded with the overall price of polypropylene carpet. See id. at 1360. That correlation, they argued, had "nothing to do with collusive activity" and undermined the significance of other factors like variables for demand, changes in income, and the entry of competition. Id. But the district judge sided with the putative class, noting that "changes in fiber costs reflect changes in the variable costs of producing polypropylene carpet." Id. at 1362. Indeed, the parties agreed that fiber costs made up "seventy percent of total costs depending on the style," so changes in the "remaining variable costs . . . did not significantly affect the high correlation between changes in fiber costs and changes in total variable costs." Id. In short, the defendants' concern about endogeneity bias—that there might have been certain unobserved factors that impacted the results—was unfounded because there was evidence suggesting that "changes in fiber costs reflect changes in the variables costs of producing polypropylene carpet." See id. In this case, the risk posed by endogeneity bias is even more obtuse because whether an employee was paid per-shift or per-hour is not itself being used to predict wages. As discussed before, the endogeneity critique has no place here. What's more, the Court did not simply exclude Dr. Singer's testimony for not including Dr. Ordover's dummy variable—it excluded his testimony because he failed to adopt any reliable method for addressing the switcher problem.
CONCLUSION
In sum, Conrad has not demonstrated a manifest error of law or fact warranting reconsideration. The Court DENIES Plaintiff Donald Conrad's Motion for Reconsideration.
IT IS SO ORDERED.
DATED: April 26, 2021
/s/ _________
NANCY J. ROSENSTENGEL
Chief U.S. District Judge