Naula Ndugga v. Bloomberg L.P.

Opinion

UNITED STATES DISTRICT COURT SOUTHERN DISTRICT OF NEW YORK wn NAULA NDUGGA, : : ORDER Plaintiff, : 20 Civ. 7464 (GHW) (GWG) -V.-

BLOOMBERG L.P., :

Defendant. panne eK GABRIEL W. GORENSTEIN, United States Magistrate Judge Because, as stated in Docket # 438, the Court has ordered that far fewer redactions be made to the Opinion and Order (filed at Docket # 433) and the Report and Recommendation (filed at Docket #434), the Clerk is directed to substitute the attached versions of those documents for the ones now appearing at those docket entries. SO ORDERED. Dated: April 20, 2026 New York, New York Soe 1 Grose United States Magistrate Judge

UNITED STATES DISTRICT COURT SOUTHERN DISTRICT OF NEW YORK wn KX NAULA NDUGGA, : Plaintiff, : OPINION AND ORDER ~V.- : 20 Civ. 7464 (GHW) (GWG) BLOOMBERG L.P., : Defendant. : wn KX GABRIEL W. GORENSTEIN, United States Magistrate Judge: Plaintiff Naula Ndugga has sued defendant Bloomberg L.P. (“Bloomberg” or “BLP”’) for violations of Title VII of the Civil Rights Act of 1964, 42 ULS.C. § 20006¢ et seq. (“Title VII’) on behalf of a proposed class of news personnel who worked at Bloomberg in 2021, and for violations of the New York State Human Rights Law, N.Y. Exec. Law §§ 290-301 (““NYSHRL”) on behalf of a proposed subclass of news personnel who worked at Bloomberg in 2017 through 2020. See Fifth Amended Complaint, filed July 10, 2024 (Docket #296) (“Compl.”), J 165-96. Bloomberg moves to exclude the opinions and testimony of Ndugga’s expert witness David Neumark.! For the following reasons, Bloomberg’s motion is denied.

' See Notice of Motion, filed Feb. 20, 2025 (Docket # 344); Memorandum of Law in Support of Defendant’s Motion to Exclude the Opinions and Testimony of David Neumark, filed Feb. 20, 2025 (Docket # 345) (“Mem.”); Declaration of Elise M. Bloom, filed Feb. 20, 2025 (Docket # 346) (“Bloom Decl.”); Opposition to Defendant’s Motion to Exclude the Opinions and Testimony of David Neumark, filed Mar. 13, 2025 (Docket # 358) (“Opp.”); Reply Memorandum of Law in Further Support of Defendant’s Motion to Exclude the Opinions and Testimony of David Neumark, filed Apr. 2, 2025 (Docket # 365) (“Reply”). The foregoing record citations are to the original unredacted, sealed filings. Redacted versions of each document are filed on the ECF system at Docket ## 421-23

I. BACKGROUND A. Procedural Background At the time suit was brought, Ndugga was a “[n]ews producer” at Bloomberg. Compl. 41. On July 10, 2024, Ndugga filed the operative complaint in which she alleges “systemic sex discrimination in compensation . . . directed from the highest levels at Bloomberg.” Id. The complaint asserts numerous class claims, in addition to claims on her own behalf. See id. 4 165-222. On December 20, 2024, Ndugga filed a motion for class certification.” In the motion, Ndugga sought to certify, for purposes of New York state-law claims, a proposed class of “[a]ll female Reporters, Producers, and Editors who: (1) were not Team Leaders or in other supervisory positions; and (2) were subjected to Defendant’s compensation systems for work performed in New York at any time from August 9, 2017 through December 31, 2020” to pursue New York state law claims. Class Cert. Mem. at 25. She also sought to certify, in order to pursue parallel Title VII claims, a proposed class of “[a]ll female Reporters, Producers, and Editors who: (1) were not Team Leaders or in other supervisory positions, and (2) were subjected

See Notice of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 329); Memorandum of Law in Support of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 330) (“Class Cert. Mem.”); Declaration of Dana Busgang in Support of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 333) (“Busgang Decl.”); Memorandum of Law in Opposition to Plaintiff’s Motion for Class Certification, filed June 23, 2025 (Docket # 379) (“Class Cert. Opp.”); Reply Memorandum of Law in Support of Plaintiff’s Motion for Class Certification, filed July 14, 2025 (Docket # 385) (“Class Cert. Reply’); Sur- Reply Memorandum of Law in Opposition to Plaintiff’s Motion for Class Certification, filed July 18, 2025 (Docket # 395) (“Class Cert. Sur-Reply”); Letter from Mark W. Batten, dated Aug. 27, 2025 (Docket # 406); Letter from Christine E. Webber, dated Sept. 10, 2025 (Docket # 411). The foregoing record citations are to unredacted, sealed filings. Redacted versions of each filing are filed on the ECF system.

to Defendant’s compensation systems for work performed in the United States at any time from February 3, 2021 through December 31, 2021.” Id. at 26. In the memorandum of law accompanying the motion, Ndugga relied heavily on an expert report authored by Dr. David Neumark, see, e.g., id. at 19-22, particularly in order to “satisfy the commonality requirement” of Rule 23, see id. at 33-35.° On February 11, 2025, defendant requested permission to file a motion to exclude the opinions and testimony of Neumark from consideration in connection with Ndugga’s motion for class certification under the principles of Daubert v. Merrell Dow Pharms., Inc., 509 U.S. 579 (1993). See Letter, filed Feb. 11, 2025 (Docket # 340). Finding that Ndugga relied heavily on Neumark’s opinions in support of the motion for class certification, the Court initially determined that it would be best to resolve the admissibility of those opinions before the parties submitted any more briefs in connection with the class certification motion. See Order, dated Feb. 12, 2025 (Docket # 343). After the Daubert motion was fully briefed, however, the Court changed course and determined that it would be best to decide the Daubert motion in conjunction with the class certification motion. See Order, dated June 9, 2025 (Docket # 374). Both motions are now fully briefed. B. Dr. Neumark’s Opinion Neumark is a labor economist holding the position of “Distinguished Professor of Economics at the University of California-Irvine.” Neumark Decl. § 1. Neumark was retained by plaintiff “to evaluate evidence of pay discrimination against women . . . at Bloomberg.” Id. 4 2.

> See Expert Report of David Neumark, dated July 1, 2024, annexed as Ex. 66 to Busgang Decl. at *4-*60 (Docket # 333-66) (“Neumark Report”); Declaration of David Neumark, dated July 1, 2024, annexed as Ex. 66 to Busgang Decl. at *2-*3 (Docket # 333-66).

1. Neumark’s Original Report Neumark used “data produced by Bloomberg that contain employment history and compensation records for the employees in the proposed Classes” to “construct a dataset of annual employee records.” Neumark Report {ff 9, 10. Neumark analyzed that data to “compare | compensation at Bloomberg for similarly-situated female and male employees.” Id. 4 12. “In addition to pay, the data” used by Neumark “include[d] an indicator for the gender of an employee, and characteristics of the individuals and their jobs.” Id. Neumark analyzed the data using a regression model to “estimate[] the female pay penalty (if there is one) once we adjust for possible differences between female and male employees that could account for a pay gap between women and men.” Id. §/ 13. As Neumark explains, For example, suppose that we simply compare average pay of all female and male employees at Bloomberg, and find that average pay of female employees is 10% lower. It is possible that women do different jobs, and those jobs could pay less. It is also possible that women and men are in broadly similar jobs, but the women have lower performance. In either case, our intuition would be that the 10% estimate overstates the pay gap for comparable women and men in comparable jobs, and we should hence adjust for these differences between women and men before estimating the female pay penalty. Of course, the opposite is also possible, so adding controls for the individual or job could lead to a larger or smaller estimated female pay penalty. This 1s precisely what a regression model does. A regression model “holds constant” or “controls for” these other factors. These phrases mean that, in estimating a regression model, we adjust the pay gap for differences in the jobs employees hold, and their characteristics and performance, so that we are comparing pay between women and men with similar performance, education, and experience. Id. 13-14. Neumark further explains that “[w]hen a pay regression model includes controls for the different jobs people do, the only pay differences between women and men that

contribute to [the] estimated gender pay gaps are differences between women and men who are in the same job.” Id. § 18. Accordingly, “[t]o control for differences in characteristics and jobs between men and women, [Neumark] include[s] several variables: race, experience since hire at Bloomberg, potential experience prior to hire at Bloomberg, education degree, employment city, Job Profile, year-end performance rating score, Cost Center, ... Business Unit,” and whether someone had part-time status. Id. § 37. Most relevant to the motion before the Court are the variables “Cost Center” and “Business Unit.” According to Neumark, “Business Units are organizational sub-entities, some of which are under the Editorial and Research Departments. Business Units can cover different subject matters, and hence provide some information about a person’s job.” Id. § 29 (footnotes omitted). Neumark explains that “there is no need to disaggregate my analyses by Business Unit,” because “policies and practices related to compensation do not differ by Business Unit within the Editorial and Research Department.” Id. Neumark nevertheless “‘allow[s] for the possibility that pay differs by Business Unit because these may contain different types of jobs.” Id. The inclusion of Cost Center is challenged by defendant. See Mem. at 12-18. Cost Center is “a different organizational category to which costs can be charged.” Neumark Report 4 30. Neumark acknowledges that “Cost Center does not play a role in compensation guidelines, so again, there is no rationale to disaggregate by Cost Center in doing my analysis. However, raises and hence pay may differ by Cost Center since different Cost Centers may have different amounts available for raises.” Id. (footnote omitted).

Neumark performed two main regression analyses: one that did not include Performance Rating scores (Model 1) and one that did include Performance Rating Scores (Model 2). Id. 4 44. The Performance Rating is the “[y]Jear-end manager rating,” which, according to testimony from Bloomberg, primarily determines year-end salary and bonuses. Id. 4] 37(m), 38(e). According to Neumark, The results for the Model 1 regression for New York 2017-2020 show that Base Pay and Total Compensation of female employees were 3.9% below and 4.4% below similarly-situated male employees, respectively. These disparity estimates have standard deviations of 2.15 and 2.16, respectively, which imply statistical significance at the 5% level. Id. ¥ 45. The results for the Model 1 regression for U.S. 2021 show that Base Pay and Total Compensation of female employees were 2.9% below and 3.2% below similarly-situated male employees, respectively. These disparity estimates have standard deviations of 1.76 and 1.68, respectively, which imply statistical significance at the 10% level. Id. 46. The results for the Model 2 regression for New York 2017-2020 show that Base Pay and Total Compensation of female employees were 3.6% below and 4.0% below similarly-situated male employees, respectively. These disparity estimates both have standard deviations of 2.12, respectively, which imply statistical significance at the 5% level. Id. 49. The results for the Model 2 regression for U.S. 2021 show that Base Pay and Total Compensation of female employees were 2.8% below and 3.0% below similarly-situated male employees, respectively. These disparity estimates have standard deviations of 1.69 and 1.64, respectively, which imply statistical significance at the 10% level. Id. ¥ 50.

Neumark concluded “to a reasonable degree of expert certainty that women reporters, editors, and producers at Bloomberg News were paid less than similarly-situated men during the analyzed periods.” Id. § 56. 2. Martin’s Criticism and Neumark’s Supplemental Tables Subsequently, in a report dated August 30, 2024, Dr. Denise Martin, defendant’s expert witness, provided a response to Neumark’s report. See Expert Report of Denise Neumann Martin, dated Aug. 30, 2024, annexed as Ex. 65 to Busgang Decl. (Docket # 333-65) (“Martin Report”). As relevant to the present motion, Martin pointed out that Neumark “drops from his analysis . .. any employee compensated in a foreign currency.” Id. § 10 n.28. This “introduces errors into his analysis.” Id. For example, if an employee moves from New York in 2019 to London in 2020, their 2019 bonus — paid in February 2020 — will be paid in GBP. Because he drops all foreign currency records from his analysis, he drops this bonus from the 2019 total compensation of this employee and treats them as receiving no bonus, thereby underestimating their compensation. If the employee returned to the United States in 2021, he improperly attaches their 2021 bonus to their 2019 base pay to calculate their 2019 total compensation. Id. In his deposition, Neumark acknowledged this “error.” See Deposition of David Neumark, dated Oct. 22, 2024, annexed as Ex. E. to Bloom Decl. (Docket # 346-5) and filed in unredacted form (Docket # 352-4) (“Neumark Dep.”), at 191:7. However, Neumark also highlighted that it was unclear how best to deal with a circumstance in which an employee was paid in the U.S. but who subsequently moved abroad and received their bonus in a foreign currency, presumably “based on [their] work there.” Id. at 191:21-192:13. Accordingly, Neumark prepared three alternate analyses. Id. at 192:14-193:11. In the first — labeled as V2 — Neumark “just exclude[d] those years rather than . . . assigning” no

bonus. Id. at 192:13-21. In the second alternative — V3 — Neumark “exclude[d] the bonus, but” also included a “dummy variable for those years which will sort of pick up the fact that you don’t have a bonus . . . which could help explain why your total comp is lower.” Id. at 192:25- 193:5. Lastly, in the third alternative — V4 — Neumark included both the bonus and the dummy variable. Id. at 193:6-11. Neumark offers V2 as the principal alternative. See Supplemental Tables, dated Sept. 26, 2024, annexed as Ex. 67 to Busgang Decl. (Docket # 333- 67) (“Neumark Supp.”), at *2 tbl. 1 (highlights the rows providing the results from V2); Neumark Dep. at 193:15-20 (“I prefer V2.”). As part of her motion for class certification, Ndugga submitted the supplemental analysis conducted by Neumark in response to Martin’s criticism. See Class Cert. Mem. at 20 n.83 (writing that, in response to Martin’s criticism, “Neumark refined his analysis and produced supplemental tables”). Neumark’s supplemental analysis provided only alternative versions of Model 2. See Neumark Supp. at *1 tbl. 1, *4 tbl. 2, *6 tbl. 3. V2 of Neumark’s Model 2 regression for the New York 2017-2020 class now shows that Base Pay of female employees was 4.2% below similarly-situated male employees with a standard deviation of 2.39. Id. at *4 tbl. 2. The analysis further shows that Total Compensation of female employees in this class was 4.4% below similarly-situated male employees with a standard deviation of 2.29. Id. V2 of Neumark’s Model 2 regression for the U.S. 2021 class now shows that Base Pay of female employees was 3.1% below similarly-situated male employees with a standard deviation of 1.86. Id. The analysis further shows that Total Compensation of female employees in this class was 3.1% below similarly-situated male employees with a standard deviation of 1.64. Id.

Il. LEGAL STANDARDS A. Legal Standard under Rule 702 “The district court’s determination whether to admit expert testimony is guided by Fed. R. Evid. 702.” United States v. Gatto, 986 F.3d 104, 117 (2d Cir. 2021). Rule 702, as amended in 2023, provides as follows: A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise if the proponent demonstrates to the court that it is more likely than not that: (a) the expert’s scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) — the testimony is the product of reliable principles and methods; and (d) the expert’s opinion reflects a reliable application of the principles and methods to the facts of the case. Fed. R. Evid. 702. The 2023 amendment resulted from proposed changes to “clarify and emphasize that expert testimony may not be admitted unless the proponent demonstrates to the court that it is more likely than not that the proffered testimony meets the admissibility requirements set forth in the rule.” Fed. R. Evid. 702 advisory committee’s note to 2023 amendment. Furthermore, “[o]nce the court has found it more likely than not that the admissibility requirement has been met, any attack by the opponent will go only to the weight of the evidence.” Id. Under Fed. R. Evid. 702, the district court “acts as a gatekeeper and ‘is charged with the task of ensuring that an expert’s testimony both rests on a reliable foundation and is relevant to the task at hand.’” Bustamante v. KIND, LLC, 100 F.4th 419, 427 (2d Cir. 2024) (internal quotation marks omitted) (quoting Amorgianos v. Nat’ R.R. Passenger Corp., 303 F.3d 256, 265 (2d Cir. 2002)). The question of relevance is tied to Rule 702’s requirement that the expert testimony “will help the trier of fact to understand the evidence or to determine a fact in issue.”

See Daubert, 509 U.S. at 591 (explaining that the first condition of Rule 702 “goes primarily to relevance’’). As part of this analysis, “‘a judge assessing a proffer of expert scientific testimony under Rule 702 should also be mindful of other applicable rules.” Id. at 595; accord Bustamante, 100 E.4th at 427. “Specifically, ‘the trial court should look to the standards of Rule 401 in analyzing whether proffered expert testimony is relevant.’” Bustamante, 100 F.4th at 427 (quoting Amorgianos, 303 F.3d at 265). Evidence is relevant if “it has any tendency to make a fact more or less probable than it would be without the evidence” and “the fact is of consequence in determining the action.” Fed. R. Evid. 401. The Supreme Court has described this inquiry as “one of ‘fit.”” Daubert, 509 U.S. at 591 (quoting United States v. Downing, 753 F.2d 1224, 1242 (3d Cir. 1985)). This analysis “requires a valid scientific connection to the pertinent inquiry as a precondition to admissibility.” Id. at 592. For example, the study of the phases of the moon. . . may provide valid scientific “knowledge” about whether a certain night was dark, and if darkness is a fact in issue, the knowledge will assist the trier of fact. However (absent creditable grounds supporting such a link), evidence that the moon was full on a certain night will not assist the trier of fact in determining whether an individual was unusually likely to have behaved irrationally on that night. Id. at 591. “If the evidence is relevant, a district court ‘must next determine whether the proffered testimony has a sufficiently reliable foundation to permit it to be considered.’” Bustamante, 100 F.4th at 427 (quoting Amorgianos, 303 F.3d at 265). “In this inquiry, the district court should consider the indicia of reliability identified in Rule 702, namely, (1) that the testimony is grounded on sufficient facts or data; (2) that the testimony is the product of reliable principles and methods; and (3) that the witness has applied the principles and methods reliably to the facts of the case.” Amorgianos, 303 F.3d at 265 (citation and internal quotation marks omitted). This

inquiry is a “flexible one,” Daubert, 509 U.S. at 594, and “must be tied to the facts of a particular case,” Kumho Tire Co. v. Carmichael, 526 U.S. 137, 150 (1999) (cleaned up). Accordingly, as one court has described, “[f]lexible methodologies . . . can be implemented in multiple ways; despite the fact that the methodology is generally reliable, each application is distinct and should be analyzed for reliability.” In re Acetaminophen - ASD-ADHD Prods. Liab. Litig., 707 F. Supp. 3d 309, 337 (S.D.N.Y. 2023) (quoting In re Zoloft (Sertraline Hydrochloride) Prods. Liab. Litig., 858 F.3d 787, 795 (3d Cir. 2017)); accord In re Keurig Green Mountain Single-Serve Coffee Antitrust Litig., 2025 WL 354671, at *4 (S.D.N.Y. Jan. 30, 2025). Generally, the focus of the court’s analysis “must be solely on principles and methodology, not on the conclusions that they generate.” Daubert, 509 U.S. at 595. Here, as described above, Neumark provides a regression analysis. Case law holds that “Tijn the context of regression analysis testimony, some (slightly) more specific standards have emerged in the case law.” Reed Constr. Data Inc. v. McGraw-Hill Cos., 49 F. Supp. 3d 385, 399 (S.D.N.Y. 2014), aff'd, 638 F. App’x 43 (2d Cir. 2016). Thus, to be admissible, a regression analysis must (1) “examine an appropriate selection of data”; (2) “be the product of a consistently followed methodology”; and (3) “control for the ‘major factors’ that might influence the dependent variable.” Id. at 400. As to the third requirement, “[nJormally, failure to include variables will affect the analysis’ probativeness, not its admissibility.” Bazemore v. Friday, 478 US. 385. 400 (1986) (concurrence in part joined by all Justices); accord Sobel v. Yeshiva Univ., 839 F.2d 18, 36 (2d Cir. 1988) (“The failure, by either side, to include a relevant variable (or the inclusion of an irrelevant or multicollinear variable) will go to the probative value of the analysis, not its admissibility.”). The exception to this rule is the failure to account for “major factors.”

Bazemore, 478 US. at 400 (concurrence in part joined by all Justices); id. at 400 n.10 (“[T]here may .. . be some regressions so incomplete as to be inadmissible as irrelevant.”). The contrasting cases of Bazemore and Bickerstaff v. Vassar Coll., 196 F.3d 435 (2d Cir. 1999), are “instructive on this point.” Reed Constr. Data Inc., 49 F. Supp. 3d at 401. In Bickerstaff, a lecturer sued Vassar College for purportedly discriminating against her on the basis of race and gender in failing to promote her to the position of full professor. See Bickerstaff, 196 F.3d at 440-41. In support, Bickerstaff offered a statistical report including a multiple regression analysis which “controlled for the independent variables of experience, rank, productivity, and discipline.” Id. at 448. This “regression analysis, however, [did] not even purport to account for two . . . major variables — teaching and service,” which were two of the three criteria Vassar used to decide whether to grant a salary increase. Id. at 449. Because “[t]hese variables [were] too significant not to be accounted for in the regression analysis in this case,” this omission was fatal to the analysis. Id. On the other hand, in Bazemore, the plaintiffs, alleging racial discrimination in employment, sought to introduce a multiple regression analysis as evidence of discrimination. Bazemore, 478 U.S. at 386-87 (per curiam). Although the regression analysis failed to control for county-to-county differences in salary increases, Bazemore found the regression analysis admissible, concluding that “it is clear that a regression analysis that includes less than ‘all measurable variables’ may serve to prove a plaintiff's case.” Id. at 400 (concurrence in part joined by all Justices). “Because the burden of proving helpfulness and relevance rests on the proponent of a regression analysis, it is the proponent who must establish that the major factors have been accounted for in a regression analysis.” Freeland v. AT & T Corp., 238 F.R.D. 130, 145 (S.D.N.Y. 2006); accord United States v. Teva Pharms. USA, Inc., 2019 WL 13244252, at *2

(S.D.N.Y. July 1, 2019); In re Keurig Green Mountain Single-Serve Coffee Antitrust Litig., 2025 WL 354671, at *17. “At the same time, a defendant challenging a regression analysis must at least identify the significant, missing variables.” Teva Pharms. USA, Inc., 2019 WL 13244252, at *2 (citing Sobel, 839 F.2d at 34); accord In re Keurig Green Mountain Single-Serve Coffee Antitrust Litig., 2025 WL 354671, at *17. “While the burden to show relevance and admissibility always rests with the proponent, ‘a mere conjecture or assertion on the defendant’s part’ is insufficient.” Teva Pharms. USA, Inc., 2019 WL 13244252, at *2 (quoting Sobel, 839 F.2d at 34); accord In re Keurig Green Mountain Single-Serve Coffee Antitrust Litig., 2025 WL 354671, at *17. “Rather, ‘such an attack should be specific and make a showing of relevance for each particular variable it contends plaintiffs ought to include.’” In re Keurig Green Mountain Single-Serve Coffee Antitrust Litig., 2025 WL 354671, at *17 (quoting Sobel, 839 F.2d at 34). As one district court has explained, this burden is “minimal.” Teva Pharms. USA, Inc., 2019 WL 13244252, at *2. B. Daubert Inquiry at the Class Certification Stage Although “[t]he Supreme Court has not definitively ruled on the extent to which a district court must undertake a Daubert analysis at the class certification stage . . . the Court [has] offered limited dicta suggesting that a Daubert analysis may be required at least in some circumstances.” In re U.S. Foodservice Inc. Pricing Litig., 729 F.3d 108, 129 (2d Cir. 2013) (citation omitted). Accordingly, “courts in the Second Circuit regularly “subject expert testimony to Daubert’s rigorous standards insofar as that testimony is relevant to the Rule 23 class certification analysis.’” Bowling v. Johnson & Johnson, 2019 WL 1760162, at *7 (S.D.N.Y. Apr. 22, 2019) (quoting Scott v. Chipotle Mexican Grill, Inc., 315 F.R.D. 33, 55 (S.D.N.Y. 2016)); accord Valelly v. Lynch, 2023 WL 2918982, at *4 (S.D.N.Y. Apr. 12, 2023).

“However, the scope of the Daubert analysis 1s cabined by its purpose at this stage: ‘the inquiry is limited to whether or not the expert reports are admissible to establish the requirements of Rule 23.’” Chen-Oster v. Goldman, Sachs & Co., 114 F. Supp. 3d 110, 115 (S.D.N.Y. 2015) (“Chen-Oster I’) (quoting Ge Dandong v. Pinnacle Performance Ltd., 2013 WL 5658790, at *13 (S.D.N.Y. Oct. 17, 2013)); accord Bowling, 2019 WL 1760162, at *7. Importantly, “because the challenged expert evidence is offered in support of class certification, as to which the District Judge is both gatekeeper and factfinder, ‘there is no risk of jury confusion,’ and ‘all doubts should be resolved in favor of admissibility.’” Molly C. v. Oxford Health Ins., Inc., 2024 WL 4850813, at *2 (S.D.N.Y. Nov. 21, 2024) (citations omitted). Il. DISCUSSION Defendant argues that Neumark’s opinion is inadmissible and should be excluded for two interrelated reasons: (1) it is irrelevant because it relies on unfounded assumptions and therefore does not “fit” the facts of this case, and (ii) it is unreliable because it fails to account for factors that are integral to BLP’s compensation process. Mem. at 11. We address these arguments next. A. Relevance Defendant argues that Neumark’s opinion is not relevant because “it relies on unfounded assumptions — specifically, that Cost Center affects compensation and that BLP’s compensation and performance evaluation decisions may be analyzed on an aggregate basis — which find no support in the record or Dr. Neumark’s analysis.” Id. at 12. 1. Neumark’s Inclusion of the “Cost Center” Variable Defendant takes issue with Neumark’s “inclusion of Cost Center, an accounting category, as a control to capture differences across jobs and ensure that he is comparing similarly situated employees.” Id. Defendant argues that “Dr. Neumark’s decision to include Cost Center as a

control in his model lacks any foundation in the record” and that “Neumark had no basis to assume that Cost Center affected employee compensation.” Id. at 13, 18. Accordingly, defendant contends that Neumark’s opinion “is a mismatch for the facts of this case” and “fails Daubert’s ‘fit’ requirement.” Id. at 18. “Given that Dr. Neumark’s entire conclusion that there was a statistically significant gender-based pay disparity evaporates if Cost Center is omitted as a control, Dr. Neumark’s opinion is irrelevant and must be excluded.” Id. Defendant points out that Neumark “himself acknowledges that ‘Cost Center does not play a role in compensation guidelines’” but includes it “based on the speculation that ‘raises and hence pay may differ by Cost Center since different Cost Centers may have different amounts available for raises.’” Id. at 13 (citing Neumark Report § 30). Defendant asserts that “[t]o support this assumption, Dr. Neumark relied entirely on references to Cost Center in two compensation guidelines memos for 2017 and 2021, respectively, which address funding for year-end bonuses and salary increases.” Id. (citing Neumark Report § 30 n.77). Regarding the 2021 memo, defendant argues that it “is entirely irrelevant because it provides compensation guidelines to be applied for 2022, which is outside the relevant time period for either putative class.” Id. Regarding the 2017 memo, defendant acknowledges that the memo “announces that six “Business Units’ (of 29 total in Dr. Neumark’s model) would have a guideline pay increase of mm instead of the mm available for other Business Units” and “notes that these Business Units ‘show up under News .. . but fall under Media cost center view.’” Id. at 14 (footnote omitted). However, defendant argues that “this oblique reference does not suggest that funding for compensation increases varied by Cost Center so as to justify including Cost Center as a control.” Id. In any case, “even if the 2017 memo provided a basis for controlling for a single

‘Media’ Cost Center, nothing in it justified controlling for the dozens of other Cost Centers listed by Dr. Neumark.” Id. Seeming to address the 2017 memo’s reference to the “Media cost center,” defendant challenges any broader connection between Business Units and Cost Centers: “there is no one- to-one mapping of Business Units to Cost Centers: there are multiple Cost Centers associated with the six business units allocated greater funding in the 2017 memo [which] are also associated with other Business Units that did not receive greater funding.” Id. Lastly, defendant argues that “Neumark also could not account for the 2018, 2019, and 2020 versions of the same memos, which apply the same guideline increases for raises across the board to all News employees without regard to Business Unit or Cost Center.” Id. at 15. In fact, defendant argues, “the 2018, 2019, and 2020 memos (and related deposition testimony) established that the same guideline increase would apply across the board in News regardless of Business Unit or Cost Center.” Id. at 16. Yet, Neumark “included Cost Center as a control variable in those years.” Id. at 15. Defendant contends that the true reason Neumark included Cost Center as a variable was “because it turned out that Cost Center generated statistically significant results.” Id. at 14-15 (emphasis omitted). In support, defendant highlights portions of Neumark’s testimony to argue that “Neumark contends that any variable that generates statistically significant results should be included in the analysis, even if there is no theoretical basis to include it.” Id. at 15. Defendant argues that such a “extraordinary endorsement of reasoning backwards from desired result to theoretical justification renders Dr. Neumark’s analysis unsound and therefore inadmissible.” Id. According to defendant, these issues compel the conclusion that Neumark’s opinion fails Daubert’s ‘fit’ requirement. It is a mismatch for the facts of this case because Dr. Neumark’s decision to include Cost Center as a control in

his model lacks any foundation in the record; even he acknowledged that he did not understand the meaning of the document on which he relied to include it. Id. at 18.7 In response, Ndugga argues that Bloomberg’s own witnesses’ testimony make clear that funding for annual compensation can vary by Cost Center. See Opp. at 18. In other words, contrary to defendant’s contentions, Neumark’s justification to include Cost Center as a variable “stems from BLP witnesses’ testimony” and was “identified before running regression analyses.” Id. Specifically, Ndugga contends that “BLP’s witnesses testified” that “Cost Center data provided information about where each employees’ pay is funded from.” Id. Ndugga explains, “Tt]his matters because different Cost Centers can have different budgets available.” Id. In support, Ndugga cites to the same witness testimony relied upon by Neumark in his decision to include Cost Center as a control variable. See id. at 18-19; Neumark Report § 30 n.77. Specifically, Ndugga cites to the deposition of Krista Sullivan, a “Compensation Team Leader” for Bloomberg, who makes it clear that “within News, to identify the budget that a particular individual is covered by, one must look to Cost Center”: Q. [W]as that typical that there can be different budgets set for different parts of editorial and research? A. Yes, there could be. There could be different budgets for the different areas of News and Research. Q. And in identifying, sort of, which individuals are going to be covered by which budget, specifically distinguishing between media and news and news editorial; would you be identifying those — or would one identify those by looking to the cost center and seeing which of the people who report up to editorial and research are on a media cost center? A. Yes. The way we break out news and editorial is by cost centers,

* Defendant also highlights that while Neumark conducted sensitivity tests — which, according to defendant, are “a method for evaluating whether a model’s results are sensitive to small changes” — for several variables, “he failed to conduct a sensitivity test for Cost Center in his original report.” Id. at 16. However, defendant acknowledges that in Neumark’s supplemental tables, Neumark “added a test of ‘joint significance’ for Cost Center” and “concluded that the Cost Center variable passed the test.” Id. at 17. Thus, it appears to us that to the extent there was an initial failure to conduct a sensitivity test, that omission has now been cured.

they both roll up to John Micklethwait, but the cost center that is attached to the employee is what breaks them out between news and editorial and media news. Opp. at 18-19 (citation omitted) (alteration in original). Ndugga also cites to the testimony of Lisa Jennings, a “Global Business Partner and 30(b)(6) designee” for Bloomberg: Q. It looks like even though these individuals had a reporting relationship up to Mr. Micklethwait, that because they were in a different cost center, that made more money available to give raises to them; is that right? A. That is how it appears in the memo, that they were getting the full funding. Q. Okay. So, even if cost center didn’t dictate sort of who your ultimate manager was or necessarily distinguish your job in other ways, it could impact the amount of funding available for salary increases and bonuses at yearend? A. So, it seems that’s the case for 2017, the way the funding was distributed. I can’t speak to any other year because I haven’t seen a similar memo.” Id. at 19 n.67 (citation omitted). Ndugga also challenges defendant’s other factual assertions — that Neumark relied solely on two memoranda in support of his justification, that the “memoranda cannot establish a pattern of Cost Center being a potential source of different compensation budgets throughout the class period,” and that “Neumark admitted the memoranda do not support his conclusion that Cost Center could impact compensation” — and provides citations to the record in an attempt to controvert them. Id. at 19-21. Thus, Ndugga argues that because “labor economic theory instructs that the amount of money available for compensation is relevant to explaining compensation,” where “budgets vary, that should be controlled for.” Id. at 21. Accordingly, “Neumark had a more than sufficient basis to believe Cost Center was an important variable to control for” and this is “enough for admissibility.” Id. (citing cases). In reply, defendant argues that

Plaintiff fails to satisfy her burden of demonstrating that Dr. Neumark’s opinions — which hinge entirely on his unfounded speculation that Cost Center “may” have affected putative class pay in each year between 2017 and 2021 — are based in sufficient fact. Cost Center was not a factor that was considered in determining employee compensation, and Plaintiff does not show otherwise. Since there is no theoretical justification for Dr. Neumark’s inclusion of 89 separate Cost Centers as control variables in his regression model, his opinions are irrelevant and must be excluded. Reply at 3-4. In support, defendant disputes many of Ndugga’s factual claims, arguing that Neumark has failed to establish “that compensation decisions at BLP were made based on Cost Center during the relevant time period.” Id. at 6. Regarding the testimony of Sullivan, cited above, defendant writes that the full context of the transcript makes clear that Sullivan’s answer “relates to a portion of the 2021 compensation guidelines memo regarding raises, Le., ‘total compensation increases,’ that Dr. Neumark has acknowledged are irrelevant because they were implemented in 2022 after the putative class period.” Id. at 4-5 (footnotes omitted). Regarding the testimony of Jennings, cited above, defendant argues that Jennings’s responses were tentative

— testifying as to how it “appears” and “seems” — and only spoke to the 2017 compensation guidelines. Id. at 5. Resolution of these factual disputes is unnecessary because it would not affect the admissibility of Neumark’s regression analysis. The Second Circuit has held that “[t]he failure, by either side, to include a relevant variable (or the inclusion of an irrelevant or multicollinear variable) will go to the probative value of the analysis, not its admissibility,” Sobel, 839 F.2d at 36 (emphasis added), so long as the analysis does not neglect “major factors,” Bazemore, 478 U.S. at 400. Accord Chen-Oster v. Goldman, Sachs & Co., 2022 WL 814074, at *11 (S.D.N-Y. Mar. 17, 2022) (“Chen-Oster IT”). While Ndugga relies on these very citations, see Opp. at 22, defendant does not respond to them in the portion of its brief addressing the purportedly erroneous inclusion of Cost Center,

see Reply 3-8.° In fact, defendant’s briefs provide no citations to case law addressing the inclusion of a purportedly irrelevant variable in a regression analysis. See Mem.; Reply. Thus, while “[t]here may . . . be some regressions so incomplete as to be inadmissible as irrelevant,” Bazemore, 478 U.S. at 400 n.10 (concurrence in part joined by all Justices), defendant does not explain how the inclusion of Cost Center makes Neumark’s regression inadmissible. Accordingly, Neumark’s inclusion of the Cost Center variable, even if defendant prevailed on its claim that its inclusion is speculative, does not render Neumark’s regression inadmissible. Rather, the inclusion of Cost Center goes to the probative value of Neumark’s analysis, not its relevance. The case law relied upon by defendant does not change this outcome. Defendant cites to Anderson v. Westinghouse Savannah River Co., 406 F.3d 248, 261-63 (4th Cir. 2005), as “affirming the exclusion of statistical expert’s opinions” and holding that “inclusion of ‘EEO job groupings’ as control variables rendered expert’s analyses deficient because job groupings were not used in employer’s performance evaluation processes.” Reply at 6. However, in Anderson, the Fourth Circuit excluded the analysis because it “was based on comparisons that were not relevant to [the plaintiffs] claims,” Anderson, 406 F.3d at 263, not because an irrelevant variable was included. Insofar as the Fourth Circuit took issue with the use of “EEO job groupings,” it was because they lacked adequate control factors. See id. (“[T]here is simply too much disparity in the groups that have been used to control, and an absence of the use of a control factor that would control for the actual job title or the job duties.”) (emphasis added).

> Defendant does address Sobel in another portion of its reply. However, there, defendant only challenges Ndugga’s reliance on Sobel in purportedly attempting to shift the burden of proof. See Reply at 16 n.36.

Defendant also relies on case law establishing that expert testimony must “rest on knowledge, a term that connotes more than subjective belief or unsupported speculation.” See Mem. at 16 (citing Northway Med. Ctr. Condo v. Hartford Fin. Servs. Grp., Inc., 2024 WL 3876380, at *6 (S.D.N.Y. Aug. 20, 2024)); Reply at 6 (citing Atl. Specialty Ins. v. AE Outfitters Retail Co., 970 F. Supp. 2d 278, 289 (S.D.N.Y. 2013)). While this is a correct statement of law, it does not change the fact that the Second Circuit has held that the inclusion of a purportedly irrelevant variable “will go to the probative value of the analysis, not its admissibility.” Sobel, 839 F.2d at 36. The knowledge requirement does not mandate that no step of an expert’s analysis may be based on subjective decision-making — after all, such subjectivity is commonplace in regression analysis. See B & R Supermarket, Inc. v. Visa Inc., 2024 WL 4252031, at *20 (E.D.N-Y. Sept. 20, 2024) (“Subjective decision-making about which variables to include or how to control for other variables is a daily part of the scientific and statistical process.”); Reed Constr. Data Inc., 49 F. Supp. 3d at 400-01 (recognizing “subjectivity” in an expert’s selection of variables but noting that “[n]Jormally, failure to include variables will affect the analysis’ probativeness, not its admissibility”) (quoting Bazemore, 478 U.S. at 400 (concurrence in part joined by all Justices)). Accordingly, Neumark’s inclusion of Cost Center as a control variable does not affect the relevance of Neumark’s report and testimony. 2. Neumark’s Purported Failure to Evaluate the Degree to Which Compensation Decisions Across Classes May Be Properly Aggregated Bloomberg argues that Neumark’s analysis is flawed because it aggregates data from all job categories instead of analyzing each category separately. Mem. at 18-19. “To justify his aggregated analysis,” defendant contends “Neumark assumed that decisions about compensation and performance ratings for the hundreds of employees within the putative classes are ‘highly

centralized.’” Id. at 19. Defendant argues that this assumption was baseless. See id. at 19-20. Specifically, Neumark “made no effort to control for the identity of the manager making decisions about ratings or compensation,” and thus “cannot say whether few, some, or all managers’ compensation decisions were infected by gender bias, or if so which ones, or to what extent it affected any individual class member.” Id. at 19-20. Defendant argues that Neumark’s purported decision to “ignore variations in performance ratings and compensation decisions made by over 100 individual managers and aggregate them all into one unitary result” was based on the assumption that “Deputy Editor-in-Chief Reto Gregori” had “final authority on core pay decisions.” Id. at 20 (footnote omitted). Neumark, however, “did no analysis to support that premise.” Id. Thus, Neumark “did not examine commonality: he assumed it.” Id. at 19. Defendant argues that Neumark’s aggregated analysis of individuals’ pay without “any statistical assessment of whether aggregation is appropriate” may not be used to prove commonality. Id. at 22. Quoting Bolden v. Walsh Construction Co., 688 F.3d 893, 896 (7th Cir. 2012), defendant argues that the “‘sort of statistical evidence that plaintiffs present has the same problem as the statistical evidence in [Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011)]: it begs the question. Plaintiffs’ expert . . . assumed that the appropriate unit of analysis 1s all of [Defendant’s] Chicago-area sites. He did not try to demonstrate that proposition.” Id. (quoting Bolden, 688 F.3d at 896) (ellipses and alteration in original). Thus, defendant argues, because “Neumark’s opinion is only relevant at this stage if it tends to show that a question common to the class can be answered for all class members on a class-wide basis,” Neumark’s opinions are irrelevant and should be excluded. Id. at 18; see id. at 22. The Court is not persuaded by these arguments.

In the portion of her motion for class certification addressing the “commonality” requirement of Fed. R. Civ. P. 23(a)(2), Ndugga provides (1) evidence of a common mode of exercising discretion — Gregori’s involvement throughout the compensation decision-making process — and (2) evidence of class-wide discrimination. See Class Cert. Mem. at 28-35. Ndugga does not rely on or cite to Neumark’s report or testimony in the section purporting to establish the former claim — that Gregori’s involvement pervaded Bloomberg News’s compensation decision-making process. See id. at 9-19. Instead, Ndugga relies on Neumark’s report only as evidence of class-wide discrimination. See id. at 33-35. Specifically, Ndugga argues, Plaintiff has identified a common mode of exercising discretion: deposition testimony and document discovery establish that Gregori had a hand on every stage of BLP’s compensation process. Her expert’s statistical analyses, which show statistically significant pay gaps between men and women, offer the “significant proof” to establish commonality for Plaintiff’s disparate treatment and impact claims. Id. (citation omitted). Accordingly, because Ndugga does not rely on Neumark’s opinion to establish the existence of a centralized decision-maker, Neumark’s opinion need not be relevant to the question of whether a centralized decision-maker exists in order to be admissible. Instead, Neumark’s opinion must be relevant to the inquiry into whether, assuming there is a centralized decision-maker, there is evidence of discrimination that produces a class-based disparity. Framed in this context, Neumark’s report is relevant even if it is based on the assumption that there is a “common practice” of employment decisions — his testimony, after all, would be used only “to answer the common question of whether there is a disparity.” Opp. at 26. While Neumark’s assumption of a common practice of employment decisions may well be incorrect, this does not render his report inadmissible. See United States v. M/Y Amadea, 2025 WL

460030, at *1 (S.D.N.Y. Jan. 15, 2025) (“[A]ssertions that an expert witness’s testimony is based upon unfounded assumptions go to the weight, not the admissibility, of the testimony.”). Defendant faults Neumark for being unable to “say, for example, that female Producers (Plaintiff’s job) were paid statistically significantly less than male Producers.” Mem. at 19 (citation omitted). However, if there is a common practice of employment decisions across job profiles — a matter that must be independently proved for purposes of class certification — Neumark’s report did not need to specifically compare pay differences amongst producers. The cases relied upon by defendant do not change this outcome. Defendant argues that Neumark’s analysis must “conform[] to the level of decision for the challenged practices,” id. at 22 (alteration in original), by citing Kassman, 416 F. Supp. 3d at 282, and (for a similar proposition, see Mem. at 21) Dukes v. Wal-Mart Stores, Inc., 964 F. Supp. 2d 1115, 1120 (N.D. Cal. 2013). However, these cases are inapposite. Here, Neumark’s report does conform to the level of decision for the challenged practice: Gregori’s alleged class-wide involvement. As Kassman explains, “if plaintiffs demonstrate that practices are uniform across the company, then courts have found ‘good reason to rely on [company-wide] statistics.”” Kassman, 416 F. Supp. 3d at 282 (citing Ellis v. Costco Wholesale Corp., 285 F.R.D. 492, 523 (N.D. Cal. 2012)). Here, Ndugga is arguing that Gregori’s involvement was common across the class. Class Cert. Mem. at 9-19. While Ndugga will ultimately be responsible for “demonstrate[ing]” this proposition, Kassman, 416 F. Supp. 3d at 282, Neumark’s report could properly assume it to be true. Defendant’s other citations are not on point. For instance, Bolden (cited in Mem. at 22), did not address whether the expert report at issue was admissible. See Bolden, 688 F.3d at 897 (“We need not determine whether Smith’s study should have been excluded under Fed. R. Evid. 702.”). Instead, Bolden addressed only the weight of the report. See id.

In sum, to be relevant to Ndugga’s motion for class certification, Neumark could properly assume the centrality of Gregori’s decision-making for purposes of conducting his analysis. B. Reliability Defendant argues that Neumark’s opinion 1s not reliable “because he fails to account for starting pay” and “unjustifiably excludes certain observations from his model” and “because his class damages calculations rely on his irredeemably flawed analyses.” Mem. at 23. We address these arguments in turn. 1. Neumark’s Purported Failure to Account for Starting Pay Defendant faults Neumark for failing to control for starting pay. Id. at 23-25. But this argument is essentially disposed of by the case law we have already discussed regarding inclusion or exclusion of specific variables in a regression analysis. See B & R Supermarket, Inc., 2024 WL 4252031, at *20 (“Subjective decision-making about which variables to include or how to control for other variables is a daily part of the scientific and statistical process.”); Reed Constr. Data Inc., 49 F. Supp. 3d at 400-01 (recognizing “subjectivity” in an expert’s selection of variables but noting that “[n]Jormally, failure to include variables will affect the analysis’ probativeness, not its admissibility”) (quoting Bazemore, 478 U.S. at 400 (concurrence in part joined by all Justices)). Certainly, an expert’s statistical analysis can be irrelevant as a matter of law if it fails to account for the “major factors.” Bazemore, 478 U.S. at 400 (concurrence in part joined by all Justices); see id. at 400 n.10 (“[T]here may . . . be some regressions so incomplete as to be inadmissible as irrelevant.”). But we see no reason why starting pay — which, as defendant has admitted, was determined by defendant, see Class Cert. Opp. at 15 (“Starting salaries were

determined by the hiring manager and recruiter, with input from the compensation team as needed.”) — is such a major factor that it must be controlled for in the regression analysis. After all, starting pay is not some external confounding variable that could provide a neutral explanation for differences in pay between men and women. Rather, Bloomberg itself set the starting pay. If, as defendant would have it, starting pay was controlled for, and it turned out that differences in starting pay explain the pay disparity between men and women, such a result could itself show that Bloomberg discriminated against women.° Accordingly, it makes no sense to control for a variable that Bloomberg itself was completely responsible for setting and which was itself potentially the result of discriminatory practices at Bloomberg. Indeed, Neumark explained in his deposition that starting pay would be “a bad control” or a “tainted variable” because it “can itself reflect discrimination.” Neumark Dep. at 46:8-10, 47:2-3. We find that Neumark “reasonably suspected that [this] variable[] [was] tainted, that is, that [it itself is] affected by gender bias.” Chen-Oster I, 114 F. Supp. 3d at 118. “Thus, there was a legitimate basis for him to omit [this variable] from his analysis on the ground that [it was itself] subject to bias and would therefore mask discrimination.” Id.; accord Morgan v. United Parcel Serv. of Am., Inc., 380 F.3d 459, 470 (8th Cir. 2004) (“Regression analyses in discrimination cases attempt to control for the legitimate reasons for pay disparities through the use of explanatory variables. But, illegitimate reasons — reasons themselves representative of the unlawful discrimination at issue — should be excluded from the regression (or otherwise dealt with) to avoid underestimating the significance of a disparity.”).

6 This is because there is no obvious non-discriminatory reason why there would be a difference between men and women in starting pay other than as a result of the effect of variables that Neumark’s study already controls for, such as education, performance rating, and experience. See Neumark Report § 37.

Defendant asserts that “starting pay decisions may be influenced or determined by factors other than discrimination.” Mem. at 24-25. However, insofar as total compensation 1s influenced by discrimination at Bloomberg, such discrimination would also plausibly influence starting pay. Thus, because Neumark’s analysis attempts to account for factors other than discrimination when examining the total compensation received by the putative plaintiff class, controlling for starting pay is not obviously required. Our conclusion that starting pay was appropriately omitted as a potentially confounding variable is in line with the rulings of other courts. See Chen-Oster I, 114 F. Supp. 3d at 118 (collecting cases). Plaintiff mentions in her opposition that “the practice of setting starting pay is not a separate practice which Plaintiff now moves the Court to find caused disparate impact.” Opp. at 32. Defendants seize on this statement to argue that, if plaintiff is “no longer alleging that sex bias influenced starting pay decisions, then starting pay cannot be said to be tainted.” Reply at 14. But the mere fact that plaintiff is not pursuing the practice of setting starting pay as a “separate practice” does not mean that plaintiff is conceding that starting pay was set in a non- discriminatory manner. Plaintiff is entitled to challenge the compensation decision-making at Bloomberg without specifically challenging each component of this process. Accordingly, the Court finds that Neumark’s testimony was not rendered unreliable because it did not account for starting pay. See Thor Equities, LLC v. Factory Mut. Ins. Co., 627 F. Supp. 3d 330, 339 (S.D.N.Y. 2022) (“An expert’s statistical analysis is generally not irrelevant merely because it fails to account for all variables that might impact the expert’s conclusion. Where the expert’s analysis accounts for all ‘major factors,’ the failure to include additional variables ‘will affect the analysis’ probativeness, not its admissibility.’”) (citation omitted); E.E.O.C. v. Bloomberg L.P., 2010 WL 3466370, at *8-9 (S.D.N.Y. Aug. 31, 2010) (“That

[Bloomberg’s expert] did not account for every possible variable that may have impacted the class members’ compensation does not render his report inadmissible. ... Here, because [Bloomberg’s expert] did not omit major variables from his analysis . . . the Court finds that his opinion is sufficiently reliable to be admissible under Rule 702.” (citation omitted)). 2. Neumark’s Exclusion of Employee-Years Defendant argues that, “[a]s Dr. Martin pointed out, and as Dr. Neumark subsequently conceded, Dr. Neumark introduced errors into his analysis when he excluded any employees compensated in a foreign currency, which he did in order to filter out employees who are not in the putative New York and USS. classes.” Mem. at 25. Specifically, in Neumark’s original report, “Dr. Neumark assumed that [employees compensated in a foreign currency] did not work in the United States and therefore are not putative class members.” Id. at 25. However, this “assumption was incorrect because it failed to account for employees who moved abroad during the year and received a bonus in a foreign currency for work performed in the United States during the previous year.” Id. at 25-26. After Martin pointed out this error, Neumark “claimed that he was able to ‘fix’ his admitted mistake by providing supplemental tables that simply excluded any years in which an employee left one of the putative classes.” Id. at 26. However, defendant alleges that “Neumark’s ‘supplemental tables’ failed to cure this deficiency in his analysis . . . because he lacked any factual foundation for his ‘solution’ of simply excluding the employee-years at issue.” Id. Defendant argues that When attempting to explain how he accounted for bonuses for employees who moved abroad mid-year, Dr. Neumark candidly admitted: “I honestly don’t have a firm answer and haven’t seen one about exactly how bonuses get determined and is it based on — hypothetically it could be based on your performance before you left. That doesn’t seem that plausible. It could be based on your performance where you are now. It could is be [sic] based on some combination. I don’t

know.” Lacking any idea how to account for bonuses in such a scenario, Dr. Neumark decided to “just exclude those years rather than . . . assigning the wrong one.” Id. (citations omitted). Defendant further states that Neumark’s purported exclusion of employee-years relies on an assumption that a bonus awarded in February of Year 2 is unconnected to performance in a Year 1, notwithstanding his stated understanding that “bonuses paid in February of each year reflect payments for the work performed in the prior calendar year.” He made no effort to reconcile these conflicts in his own opinions and instead opted to summarily drop over 100 employee-year observations from his regression models without ever attempting to consider potential differences between employees who left early in Year | versus late in Year 1. Id. at 26-27 (citations omitted). Defendant argues that “[t]his is not a reliable analysis by any definition.” Id. at 27. Ndugga responds:. Neumark in his supplemental tables corrected for issues of connecting bonus decisions made after the end of the year with the base salary determination made at the start of the year, particularly where an individual had moved Jobs or locations mid-year. Dr. Neumark explained that when an employee moved between a class and a non-class position midyear — either because of a change in role or geography — he did not have sufficient information to tie the bonus amount to the work done while in the class position. He therefore stated it was best practice to exclude these observations — fewer than 100 observations out of nearly 2,000. However, Dr. Neumark also demonstrated very similar results if those 89 observations were included. Opp. at 35-36 (footnotes omitted). In the first of these alternative analyses, Neumark “included those 89 observations only with base salary in a class-covered job and added a variable to note that bonus was omitted because it was awarded after the person moved out of the class.” Id. at 36. This version is referred to as V3 in the supplemental tables. See Neumark Supp. *2 tbl. 1. In the second alternative analysis, referred to as V4 in the supplemental tables, see id., Neumark “included both salary and bonus for the 89 observations, with an added variable to note that the

bonus was included despite being awarded after the employee had moved out of the class.” Opp. at 36. Ndugga contends that these “alternative analyses, as well as Dr. Neumark’s preferred model, were all similar, with all statistically significant for the New York class at the 5% level, and for the US Class at the 10% level.” Id. “Thus, Dr. Neumark devised appropriate and reliable analyses for dealing with imperfect data.” Id. In response, defendant argues that it “is plainly false” that the alternative analyses are statistically significant. Reply at 17. Specifically, defendant argues that “[i]n his ‘alternative’ V3 and V4 models that do not drop any employee-year observations, Dr. Neumark does not find any statistically significant results for the putative U.S. class with respect to total compensation.” Id. Defendant asserts that Ndugga “has not established that [Neumark’s] ‘solution’ to address his lack of understanding — summarily dropping over a hundred employee-year observations from his analysis — was based on any facts, much less ‘sufficiently grounded in reliable facts.’” Id. at 18 (citing Atl. Specialty Ins., 970 F. Supp. 2d at 285). The Court will not exclude Neumark’s report on this basis. Neumark’s exclusion of some observations in one of his models does not render his opinion unreliable. First, the Court notes that defendant does not cite to any case law to help guide the Court’s analysis of the circumstances when the exclusion of some observations in a regression analysis renders the analysis unreliable. See Mem.; Reply. Thus, while defendant invokes Rule 702’s mandate that an expert’s testimony must be “based on sufficient facts or data,” Fed. R. Evid. 702(b); see Mem. at 9 (citing Fed. R. Evid. 702), defendant does not explain why the exclusion of 89 observations (out of 1,983) in the New York class and of 29 observations (out of 783) in the U.S. class, Neumark Supp. at *2 tbl. 1, renders the data insufficient.

Second, the sparse case law that addresses these circumstances suggests that Neumark’s exclusion of some observations is not enough to find his conclusions unreliable. Thus, one court concluded that “‘simply discard[ing]’ any observations with missing values[] remains ‘the standard treatment of missing data’ and ‘the most common method in the absence of readily available alternatives.’” In re Pool Prods. Distrib. Mkt. Antitrust Litig., 2016 WL 2756437, at *9 (E.D. La. May 12, 2016) (concluding that the expert’s “approach to the missing data is therefore not unreliable, and the Court will not exclude his opinion on this basis”). Third, as Ndugga points out, see Opp. at 36, “[t]he courts draw a distinction between insufficient data and imperfect data. If a proposed expert lacks the basic information required by professionals in his field to form an opinion, his testimony will be excluded. Imperfect data, however, goes to the weight of the expert’s opinion, not its admissibility, and is not grounds for exclusion.” Molly C., 2024 WL 4850813, at *12 (citations omitted). Here, defendant makes no argument that — in the case of employees who worked part of the year domestically before moving to a foreign location and received a bonus abroad — data existed as to which portion of the bonus was given for domestic work and which portion of the bonus was given for foreign work. Indeed, Neumark explained in his deposition that he did not have data on which portions of the bonuses paid to employees who spend part of the year domestically and part of the year abroad were paid for domestic work (and vice versa). Neumark Dep. at 191:21-192:14. Ifthe missing data, which involves a tiny percentage of the data points, makes any difference to the regression analysis, this may affect the weight that Neumark’s report deserves. It does not suggest that his methodology was itself unreliable. As one court has explained, “[t]he question for Daubert purposes is not whether the expert employed the best possible input data to form her testimony, but whether ‘the testimony is based on sufficient facts or data.’” In re Int. Rate Swaps

Antitrust Litig., 2023 WL 8675625, at *4 (S.D.N.Y. Dec. 15, 2023) (citing Fed. R. Evid. 702). Defendant has made no showing that the missing observations render Neumark’s opinion insufficient. Thus, the Court rejects defendant’s contention that Neumark’s decision to omit a number of observations renders his opinion unreliable. 3. Unreliability in Neumark’s Class Damages Calculations Lastly, defendant argues that Neumark also purports to calculate damages for the two proposed classes by multiplying the pay disparities he claims to have found by the average class member’s total compensation for the respective class periods and adding statutory interest. Because these figures rely on irredeemably flawed statistics produced by Dr. Neumark’s irrelevant and unreliable analyses, his class damages calculations should be excluded from any future proceedings in this case. Mem. at 26 (citations omitted). However, while Ndugga argues that the predominance requirement of Fed. R. Civ. P. 23(b\(3) “can be calculated using BLP’s records for dates of employment and using the average disparities based on Dr. Neumark’ tabulations,” Class Cert. Mot. at 41, Ndugga’s class certification motion does not raise any substantive issues regarding damages. Thus, we have no occasion to address whether Neumark’s analysis is reliable as to damages.

IV. CONCLUSION For the above reasons, Bloomberg’s motion to exclude the opinions and testimony of Ndugga’s expert witness David Neumark for purposes of the class certification motion (Docket # 344) is denied. Dated: March 19, 2026 New York, New York Soe 0 Grose United States Magistrate Judge

UNITED STATES DISTRICT COURT SOUTHERN DISTRICT OF NEW YORK wn KX NAULA NDUGGA, : Plaintiff, : REPORT and RECOMMENDATION ~V.- : 20 Civ. 7464 (GHW) (GWG) BLOOMBERG L.P., : Defendant. : wn KX GABRIEL W. GORENSTEIN, United States Magistrate Judge Plaintiff Naula Ndugga has sued defendant Bloomberg L.P. (“Bloomberg” or “BLP”) for violations of Title VII of the Civil Rights Act of 1964, 42 ULS.C. § 20006¢ et seq. (“Title VII’) on behalf of a proposed class of news personnel who worked at Bloomberg in 2021, and for violations of the New York State Human Rights Law, N.Y. Exec. Law §§ 290-301 (““NYSHRL”) on behalf of a proposed subclass of news personnel who worked at Bloomberg in 2017 through 2020. See Fifth Amended Complaint, filed July 10, 2024 (Docket #296) (“Compl.”), J 165-96. Ndugga now moves for class certification.! For the following reasons, class certification should be denied.

' Notice of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 329) (“Mot.”); Memorandum of Law in Support of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 330) (““Mem.”); Declaration of Dana Busgang in Support of Plaintiff’s Motion for Class Certification, filed Dec. 20, 2024 (Docket # 333) (“Busgang Decl.”); Expert Report of Denise Neumann Martin, dated Aug. 30, 2024, annexed as Ex. 65 to Busgang Decl. (Docket # 333-65) (“Martin Report”); Expert Report of David Neumark, dated July 1, 2024, annexed as Ex. 66 to Busgang Decl. at *4-*60 (Docket # 333-66) (“Neumark Report”); Memorandum of Law in Opposition to Plaintiff’s Motion for Class Certification, filed June 23, 2025 (Docket # 379) (“Opp.”); Declaration of Elise M. Bloom, filed June 23, 2025 (Docket # 380) (“Bloom Decl.”); Reply Memorandum of Law in Support of Plaintiff’s Motion for Class Certification, filed July 14, 2025 (Docket # 385) (“Reply”); Sur-Reply Memorandum of Law in Opposition to Plaintiff’s Motion for Class Certification, filed July 18, 2025 (Docket # 395) (“Sur-Reply”); Letter from Mark W. Batten, dated Aug. 27, 2025 (Docket # 406) (“Batten

I. BACKGROUND A. Defendant Bloomberg Bloomberg is “a global financial, software, data and media company.” Opp. at 2. Bloomberg News 1s “‘a global financial and business news organization” within Bloomberg “with over 3,000 employees.” Id. Bloomberg News is known internally as “Editorial and Research.” Deposition of Lisa Stephanie Jennings, dated Apr_8, 2024, annexed as Ex. C to Bloom Decl. (Docket # 380-3) (“Jennings Dep.: Bloomberg Excerpts”), at 21:10-14. Bloomberg News was overseen by the Editorial and Research Management Committee (the “Committee”)” during the relevant periods. See Jennings Dep.: Bloomberg Excerpts at 23:9-25. John Micklethwait, Bloomberg News’ editor-in-chief, and Reto Gregori, Bloomberg News’ deputy editor, were members of the Committee during the relevant periods. See Deposition of Reto Gregori, dated Apr_17, 2024, annexed as Ex. 1 to Busgang Decl. (Docket # 333-1) (“Gregori Dep.: Ndugga Excerpts”), at 10:7-9; 11:8-11; 17:7-18:24. The Committee consisted of Micklethwait, Gregori, and a “Chief Content Officer” until 2020. See id. at 18:17- 19:20, 22:10-15. That year, it was expanded to include additional individuals — in Gregori’s words, “to make sure that we had a slightly more diverse management committee at the top,” Le., to include “[m]Jore women.” Id. at 23:15-17, 19.

Letter”); Letter from Christine E. Webber, dated Sept. 10, 2025 (Docket # 411) (“Webber Letter”). The foregoing record citations are to the original unredacted, sealed filings. Redacted versions of each document are filed on the ECF system at Docket ## 421-23. ? Previously, the Committee had been called the Editorial Management Committee. See Jennings Dep.: Bloomberg Excerpts at 22:9-22.

Bloomberg News was organized into “Business Units” covering “different subject matters.” Jennings Dep.: Bloomberg Excerpts at 49:16-17. Business Units were overseen by “senior manager[s].” Deposition of Shelby Siegel, dated May 30, 2024, annexed as Ex. F to Bloom Decl. (Docket # 380-6) (“Siegel Dep.: Bloomberg Excerpts”), at 30:25-31:3. Employment roles at Bloomberg News were referred to as “Job Profiles.” Deposition of Lisa Stephanie Jennings, dated Apr_8, 2024, annexed as Ex. 2 to Busgang Decl. (Docket # 333- 2) (‘Jennings Dep.: Ndugga Excerpts’), at 50:10-51:13. Each employee was assigned to a Business Unit and, separately, to a “peer group.” See id. at 47:4-25, 74:2-9, 77:16-18. B. Plaintiff Ndugga At the time she brought this action, Ndugga was employed at Bloomberg News. See Compl. ¥ 1. Ndugga is Black and female. Id. § 10. She began working at Bloomberg News as a paid intern in September 2017. Id. Ndugga accepted a full-time offer of employment as a news producer with Bloomberg News’ “QuickTake department” (a Business Unit) on January 29, 2018. Id. § 102. Ndugga alleges that, “[uJpon becoming hired as a full-time employee, she was paid an annual salary of $65,000,” while “male producers who were hired out of the same internship program ... were paid a starting salary of $75,000.” Id. § 103. Ndugga also alleges that she was “repeatedly . . . overlooked for raises, promotions, favorable assignments, professional growth, and opportunities that were given to her male peers” despite “receiving positive feedback from her Manager.” Id. § 105. Further, she “received only small cost of living increases in her first three years, bringing her salary to approximately $66,500 annually.” Id. § 106. While Ndugga received “a merit raise” in 2021, “increasing her pay to approximately $70,000,” this was “still

less than the 2018 starting salary of her male peers, and far below her male peers’ contemporaneous pay.” Id. C. The Proposed Classes The operative complaint proposes two classes of similarly situated women. See id. at 4 56-57. As relevant here, Ndugga alleges that she and the proposed class[es] were subject to compensation decisions that were intended to, and had the effect of, adversely affecting women, which were made by the Editorial Management Committee, a small group consisting solely of white men, based in New York, which systematically favored white men like the committee members over women. .. . Id. 1. However, in her memorandum of law in support of class certification, Ndugga explains: In light of confidential information gleaned during discovery, Plaintiff has narrowed her arguments supporting class certification. Specifically, she has refined her understanding of how the EMC operated as a single decisionmaker with respect to compensation. Namely, the single-decisionmaker was the ever- present Deputy Editor-in-Chief and EMC member Reto Gregori. Mem. at 26 (footnote omitted). Ndugga argues that Gregori “has had his fingerprints on every stage of the evaluation and compensation process for BLP’s reporters, editors, and producers. The result has been lower pay for women than their male peers.” Id. at 25. Ndugga is seeking to certify a proposed class (the “U.S. Class”) of “[a]ll female Reporters, Producers, and Editors who: (1) were not Team Leaders or in other supervisory positions, and (2) were subjected to Defendant’s compensation systems for work performed in the United States at any time from February 3, 2021 through December 31, 2021.” Id. at 26. This class is estimated to number 338 individuals. See Supplemental Tables, dated Sept. 26, 2024, annexed as Ex. 67 to Busgang Decl. (Docket # 333-67) (“Neumark Supp.”), at *8 tbl. 4. On behalf of the proposed U.S. Class, Ndugga asserts disparate treatment and impact claims under Title VII. See Compl. 4§ 165-180.

Additionally, Ndugga is seeking to certify a proposed subclass (the “New York Class”) of “Tall female Reporters, Producers, and Editors who: (1) were not Team Leaders or in other supervisory positions; and (2) were subjected to Defendant’s compensation systems for work performed in New York at any time from August 9, 2017 through December 31, 2020.” Mem. at 25. This class is estimated to number 315 individuals. Neumark Supp. at *8 tbl. 4. On behalf of the proposed New York Class, Ndugga asserts disparate treatment and impact claims under the NYSHRL. See Compl. 181-196. “Members of the putative classes worked in nearly 30 cities across the country, in more than 30 different Business Units, and held more than 30 different Job Profiles.” Opp. at 4 (citing Neumark Report at 32 tbl. A-1, 35 tbl. A-4; Martin Report § 10()). They were also “assigned to more than 40 different peer groups.” Id. at 8 (citing Neumark Report at 34 tbl. A-3). D. Bloomberg’s Compensation Process Plaintiff’s arguments in favor of class certification center on her contention that compensation at Bloomberg was determined by a “single decisionmaker”: namely, Gregori. Mem. at 26. As Ndugga puts it, Gregori “micromanaged, at both systemic and individual levels, every stage of [Bloomberg News’]| multipart evaluation and compensation systems.” Id. at 1. Bloomberg counters that “subjective decisions about performance ratings and compensation ... were made by hundreds of different managers, each of whom was exercising their own discretion and independent judgment to make highly individualized determinations based on each employee’s performance and contributions.” Opp. at 1. We next examine Bloomberg’s compensation process and Gregori’s role in it.

1. Performance Ratings Bloomberg News evaluated the performance of employees at mid-year and year-end. See Deposition of Reto Gregori, dated Apr_17, 2024, annexed as Ex. D to Bloom Decl. (Docket # 380-4) (“Gregori Dep.: Bloomberg Excerpts”), at 48:14-16, 90:16-23. The policies and practices with respect to performance evaluations were set by Bloomberg’s human resources department and applied throughout the company. See Jennings Dep.: Ndugga Excerpts at 38:8-39:5. Employees were rated by their managers on a scale of 1 (best) to 5 (worst) on “cultural” and “business-specific” performance metrics. Deposition of Shelby Siegel, dated May 30, 2024, annexed as Ex. 6 to Busgang Decl. (Docket # 333-6) (“Siegel Dep.: Ndugga Excerpts’), at 185:5-14; see Gregori Dep.: Ndugga Excerpts at 64:2-8. These metrics were the same for employees who shared a Job Profile or belonged to the same peer group. See Siegel Dep.: Ndugga Excerpts at 186:9-15. Managers would input their ratings into a tool called “EVAL” which would then create an overall performance rating for each employee by weighting the various metrics. See id. at 186:16-187:16. Managers were to aim for a certain distribution of ratings across the employees they supervised. See Gregori Dep.: Bloomberg Excerpts at 60:12-61:5. This distribution was determined by Gregori; Lisa Jennings, a “Global Human Resources Partner” at Bloomberg; and Jennings’s “team.” See Jennings Dep.: Bloomberg Excerpts at 312:17-21. However, Gregori had the final say over the distribution. See id. at 312:22-24. Gregori testified that the distribution was a “somewhat elastic guideline[,] so I’m not going to force managers to come up exactly with this[,] but directionally we want to make sure that we have a curve that reflects the performance of any given group.” Gregori Dep.: Bloomberg Excerpts at 61:18-62:3.

Managers’ ratings would be reviewed by several layers of senior management. Bloomberg points to an e-mail, dated November 18, 2019, sent by Emma O’Brien, as representative of this process. See Opp. at 6 n.34 (citing E-mail from Emma O’Brien, dated Nov. 18, 2019, annexed as Ex. S to Bloom Decl. (Docket # 380-19) (“O’Brien E-mail”)). In that e-mail, managers were instructed to submit their ratings to managing editors, who in turn were instructed to submit the ratings (with any changes) to executive editors, who were to submit the ratings (with any further changes) to Chris Collins, the “head of Breaking News and Markets.” See O’Brien E-mail; Gregori Dep.: Bloomberg Excerpts at 55:7-56:11. Collins would then submit the ratings to Gregori and Micklethwait. Gregori Dep.: Bloomberg Excerpts at 56:12-15. At his deposition, Gregori was asked about his role in reviewing ratings: Q. So, Mr. Gregori, when you reviewed ratings that was not just for, am I correct that it’s not just for individuals who are in the direct line of reporting to you but everybody from News who ultimately has to be approved by John Micklethwait[, ] you would see those ratings first? A. That is correct. Q. And you could either ask for changes overall to meet the curve or you can also decide that you wanted changes base[d] on a specific individual and say this is the person who should be written down to a 4 or upgraded to a 2; correct? A. In theory I can do that. Q. And you have done all those things; correct? A. Occasionally, yes. Q. What do you mean by “occasionally”? A. Ideally managers do a good job so I don’t have to go and push back and say change this or that[,] but if a manager doesn’t do a good job[,] and occasionally that happens[,] then I have a conversation and reject the ratings so they could redo them. Q. And sometimes it’s just that you and the manager have a disagreement about how to rate a particular individual, right? A. I would not go — if a manager has a clear view of the performance of a person[,] I’m not going to second[-]guess the manager. If the overall distribution looks weird or if the manager says x-person should, I want a significant raise for a certain reporter or we need to give this person a certain, a massive raise, if the manager rates that person all or afl. I would say that doesn’t make sense[,] then it’s not really a disagreement|,] it’s more of a conversation[,] so whatever we

come up with in the end that it makes sense so you have, so the pay reflects the performance rating. Q. And you also want the performance rating to reflect the performance? A. In theory that would be a good thing, yes. Id. at 72:5-73:22. Ndugga points to four instances in which Gregori rejected or modified ratings and another in which Gregori references having provided guidance on future ratings. See Mem. at 7 nn.28-29. In an e-mail exchange dated July 23, 2018, Gregori wrote to a manager, “Yours look good, largely (a bit on the generous side, but fine). One thing — I’d recommend keeping [a male employee] as all I really don’t think he’s done anything particularly impressive (or much more than previously). He’s al at the moment.” E-mail from Reto Gregori, dated July 23, 2018 (11:25 AM), annexed as Ex. 38 to Busgang Decl. (Docket # 333-38) (“July 23, 2018 E-mail”). The manager responded by explaining that the male employee’s rating had been a “bone of contention among” his supervisors, one of whom felt that it was odd to “give him all while also “prais[ing] him for improvement.” E-mail from Brad Stone, dated July 23, 2018, annexed as Ex. 38 to Busgang Decl. (Docket # 333-38). Gregori suggested that the manager “keep him [at alll. with the upward trajectory, and then upgrade him at year-end when it actually matters (and a reminder that he will never see the rating).” E-mail from Reto Gregori, dated July 23, 2018 (6:34 PM), annexed as Ex. 38 to Busgang Decl. (Docket # 333-38). In another e-mail to a manager, with the subject line “You saw I rejected ratings,” Gregori wrote, “We need a few more fs (you have only a or so of the organization rated worse than which is a little too generous). Separately, I upgraded [REDACTED] and [REDACTED] in light of your other message re $.” E-mail from Reto Gregori, dated July 23,

2018 (7:00 PM), annexed as Ex. 39 to Busgang Decl. (Docket # 333-39) (“2d July 23, 2018 E- mail’). In another e-mail to a manager, with the subject line “Ratings Rejected,” Gregori wrote, “Can you have another go and have more of a curve distribution[?] At the moment you have fs than[fs. I know everyone is brilliant, but still.’ E-mail from Reto Gregori, dated Dec. 17, 2021, annexed as Ex. 51 to Busgang Decl. (Docket # 333-51) (“December 17, 2021 E-mail”). In another e-mail to a manager, Gregori wrote: I’m sending your ratings back to you for a little tweaking. The distribution is a little skewed toward the top. Can you please have another look at the fs (you only have a Als. compared with a the last lf rating cycles)[?] I realize you have a lot of people who are rated for the first time, accounting for a large part of the a but there are also quite a lot of upgrades. E-mail from Reto Gregori, dated June 21, 2017, annexed as Ex. 43 to Busgang Decl. (Docket # 333-43) (“June 21, 2017 E-mail”). Lastly, in an e-mail dated July 21, 2020, Gregori told Micklethwait that “there’s a slight increase in upgrades, led by [manager], who wanted to signal how hard everyone worked etc etc. I told those extra-nice managers . . . we expect them to revert to a slightly more normal curve for year-end.” E-mail from Reto Gregori, dated July 21, 2020, annexed as Ex. 31 to Busgang Decl. (Docket # 333-31) (“July 21, 2020 E-mail”). Thus, while Ndugga shows that Gregori had the power to reject or modify ratings —and on three occasions exercised that power, — see June 21, 2017 E-mail; December 17, 2021 E- mail; 2d July 23, 2018 E-mail — she has not shown how often he did so. In other words, she has not pointed to evidence that would allow a finding as to the degree to which Gregori influenced ratings.

Starting in 2018, each employee was also ranked (“stacked”) within his or her peer group based on his or her overall performance rating. See Siegel Dep.: Bloomberg Excerpts at 180:3- 14; Jennings Dep.: Bloomberg Excerpts at 63:23-24. The stated purpose of peer groups was to “enable managers to better compare employees doing like jobs” and “allow [Bloomberg] to more fairly compensate employees relative to peers.” Memorandum, undated, annexed as Ex. 51 to Busgang Decl. (Docket # 53) (“Peer Groups Mem.”), at *2. Employees were assigned to peer groups based on Job Profiles and other criteria, such as experience. See Jennings Dep.: Bloomberg Excerpts at 64:10-12, 80:18-25, 90:15-24. Each group could include employees reporting to different managers and working in different Business Units. See id. at 77:10-18. Shelby Siegel, Gregori’s chief of staff, collaborated with Bloomberg’s compensation (“comp”) team and human resources department to create the peer groups. See Gregori Dep.: Ndugga Excerpts at 144:5-12. However, Gregori approved the peer groups. See Jennings Dep.: Ndugga Excerpts at 144:21-145:7. Gregori testified that “the first time that we did peer groups,” he made “about 20 peer group changes” because 20 employees “were in the wrong peer groups, so an editor or a senior editor who was in more of a more junior editor bucket or a reporter that should have been in an editor bucket or a bureau chief who was managing, something like that.” Gregori Dep.: Ndugga Excerpts at 85:9-86:23. There is no evidence that Gregori made any changes to the makeup of the peer groups at any other time. The process of ranking employees within a peer group was referred to as “stacking.” See Siegel Dep.: Bloomberg Excerpts at 180:3-7. This process was separate from, and subsequent to, the determination of an employee’s overall performance rating. See Peer Groups Mem. at *2. However, if a peer group comprised fewer than 20 employees, those employees would not be

stacked. See Deposition of Krista Sullivan, dated May 20, 2024, annexed as Ex. H to Bloom Decl. (Docket # 380-8) (“Sullivan Dep.: Bloomberg Excerpts”), at 62:13-16. Jennings testified that Gregori “had the access to move people up or down [in a stack rank] without going back into the review and changing the [underlying performance rating].” Jennings Dep.: Ndugga Excerpts at 148:2-9; accord Peer Groups Mem. at *3. Similarly, Gregori testified that “people submit their ratings for their people and then I can look at what ever stack the system creates and in rare instances if something looks out of order in theory I can change that, yes.” Gregori Dep.: Ndugga Excerpts at 60:7-11. While Ndugga argues that Gregori “shuffled the rankings . . . and directed managers to make changes to the stacks,” Mem. at 14-15, she points to no instances of him actually having done so. For her part, Jennings testified that she was not sure if Gregori “made any changes or not.” Jennings Dep.: Bloomberg Excerpts at 149:23-24. 2. Compensation Decision-Making Employee compensation at Bloomberg News was made up of a base salary and a discretionary bonus. See Jennings Dep.: Bloomberg Excerpts at 176:16-177:6. Bloomberg News used an algorithm (sometimes called “algo” within Bloomberg) to make initial compensation recommendations for employees during the relevant periods. See Sullivan Dep.: Ndugga Excerpts at 27:20-28:3. The algorithm made individualized recommendations based on a number of inputs, including: ee.

Opp. at 9; see Defendant’s Objections and Verified Answers to Plaintiff's Fourth Set of Interrogatories, dated Aug. 7, 2024, annexed as Ex. 64 to Busgang Decl. (Docket # 333-64) (“August 7, 2024, Interrogatories”), at 8-9. Additionally, the algorithm considered iT ee August 7, 2024, Interrogatories at 8; see Editorial Managers Presentation, dated Nov. 2018, annexed as Ex. 12 to Busgang Decl. (Docket # 333-12), at 3 (indicating that the algorithm considered [in when determining “imerease in comp”). After making individualized recommendations, the algorithm drew a “curve” or “plot” which displayed all of the initial compensation recommendations within a given peer group. See Jennings Dep.: Bloomberg Excerpts at 226:14-18; Editorial Managers Presentation at 3. The algorithm followed various rules in making initial compensation recommendations. See Deposition of Krista Sullivan, dated May 20, 2024, annexed as Ex. 7 to Busgang Decl. (Docket # 333-7) (“Sullivan Dep.: Ndugga Excerpts”), at 33:5-16. These rules might set a ee ee for example. Id. at 33:11-2, 14-16. Jennings explained that the rules generally operated to “distribut[e] more money to well-rated people in... peer groups that are lowly comped.” Jennings Dep.: Bloomberg Excerpts at 211:2-

ee ee Algo Rules, undated, annexed as Ex. 10 to Busgang Decl. (Docket # 333-10). Krista Sullivan, a “comp team” leader, testified that she “would discuss with Reto [Gregori] the algo rules or logic for each year.” Sullivan Dep.: Ndugga Excerpts at 97:14-19.

Sullivan added that Gregori was the only member of the Committee who would discuss what the would be with the comp team. See id. at 98:14-20. If Gregori requested to change the rules and his proposed changes were consistent with available funding, Sullivan would honor his requests. See id. at 109:21-110:6. In a given year, after the “initial rules” had been loaded into the algorithm, the comp team would “run a simulation” and show Gregori “the recommendations that were generated.” Id. at 44:24-45:4. The comp team would “sometimes make changes” before the recommendations were finalized. Id. at 51:24. “[A]t times,” these changes would be made at Gregori’s request. Id. at 52:4-6. Ndugga offers a single example of the kinds of changes Gregori requested, see Mem. at 12 n.48, 13 n.39 — an e-mail in which Gregori told Sullivan: I’m wondering whether we shouldn’t give everyone who is rated an automatic bump in 2020 bonus of a mimimum of a so it’s meaningful for the junior members of society. While mi is fine for someone in Bloomberg Opinion with a top rating, a junior markets reporter who killed it this year and also cets will still only get something like 7a more in bonus, which doesn’t seem particularly motivating ... . Some of the 2021 recommendations for middling performers seem big compared with slightly less good people. One that caught my eye is [REDACTED] in London , who is slated to get anf increase for 2021, when people rate e |female employee] in NYC, is looking at a bonus cut and a drop. Maybe we should limit increases for people rated between [fi and

E-mail from Reto Gregori, dated Jan. 4, 2021, annexed as Ex. 34 to Busgang Decl. (Docket # 333-34) (“January 4, 2021 E-mail”). Gregori also asked, “[C]an we zero out the 2021 increase for [REDACTED] on Opinion, who is already ridiculously overpaid for historical reasons[?|” Id. This question posed by Grigori is Ndugga’s only evidence that Gregori “directed changes to” initial compensation recommendations for individual employees. Mem. at 12. But Bloomberg asserts, and Ndugga

does not controvert, that “[REDACTED] on Opinion” is “not a putative class member or a male peer.” Opp. at 14. In this instance, Sullivan changed the rules in accordance with Gregori’s suggestions. See E-mail from Krista Sullivan, dated Jan. 7, 2021, annexed as Ex. 34 to Busgang Decl. (Docket # 333-34). She did not answer Gregori’s question about “[REDACTED] on Opinion.” Id. After any changes were made to either the algorithm’s rules or individual employees’ initial compensation recommendations, these recommendations would be added to Bloomberg’s online compensation system (referred to as “BOCS”). See Sullivan Dep.: Bloomberg Excerpts at 51:19-22. Managers would then adjust initial compensation recommendations at their discretion. See Jennings Dep.: Bloomberg Excerpts at 214:25-215:4; Sullivan Dep.: Bloomberg Excerpts at 26:8-13, 242:25-243:8. Jennings testified that managers should and would “tweak” the recommendations “where they feel it’s warranted.” Jennings Dep.: Bloomberg Excerpts at 244:3-4. Bloomberg points to two slide shows, both dated November 2018, which directed managers to use the recommendations as “a starting point for making decisions that are fair based on actual performance and compensation relative to peers.” Editorial Managers Presentation at 3; Team Leaders and Bureau Chiefs Presentation, dated Nov. 2018, annexed as Ex. 13 to Busgang Decl. (Docket # 333-13), at 3. When a manager decided to adjust an employee’s recommended compensation, they would enter the adjusted figure into BOCS. See Sullivan Dep.: Bloomberg Excerpts at 154:4- 157:2. BOCS would log the change and who made it. See id.

After managers reviewed initial compensation recommendations, the recommendations (as adjusted) were then submitted to more senior management at Bloomberg News. See Jennings Dep.: Bloomberg Excerpts at 290:4-9. Review of the recommendations would “go[] up through line managers to the senior executive editor.” Gregori Dep.: Bloomberg Excerpts at 94:19-20. As the recommendations advanced up the managerial chain, any manager could make adjustments before passing them along. See Jennings Dep.: Bloomberg Excerpts at 290:10-13. Ultimately, the recommendations were reviewed by Gregori — who was also able to adjust compensation, see Sullivan Dep.: Ndugga Excerpts at 169:7-12 — before Gregori and Micklethwait finalized them. See Jennings Dep.: Bloomberg Excerpts at 290:14-20; accord Gregori Dep.: Bloomberg Excerpts at 94:25-95:3. Bloomberg’s retained expert, Dr. Denise Neumann Martin, analyzed the “data resulting from the annual managerial review process” from 2018 to 2021. Martin Report § 48. Martin found that “[of] the 2,807 employee-years in the audit logs that also appear in [Ndugga’s expert’s] analysis, 2,085 or 74.3% had at least one manager change.” Id. § 48(b). In other words, approximately 74% of the time, employees’ pay differed from what the algorithm recommended. During the relevant periods, 126 managers (including Gregori) made changes to initial compensation recommendations. See id. § 48(a). Martin found that between 2018 and 2021, 2,187 changes were made to bonuses, id. at 34 fig. 8, and 3,363 changes were made to total compensation, id. at 36 fig. 9. Gregori made 286 of the changes to bonuses and 389 of the changes to total compensation. See id. § 48(e). Thus, Gregori was responsible for 13.1% of the changes made to bonuses and for 11.6% of the changes made to total compensation. Overall, Gregori made about 12% of the changes to employees’ pay during the relevant periods.

Ndugga asserts that Gregori would “direct[] lower level managers to make changes, which would not appear as attributable to Gregori in the [BOCS] audit logs.” Mem. at 16. She cites to four e-mails from Gregori by way of proof. See id. at 16 n.66; Webber Letter at 3. One e-mail is in response to a message from a manager flagging “a large discrepancy amongst some people on the TicToc [QuickTake] team.” E-mail from Rachel Parisee, dated Mar_8, 2019, annexed as Ex. 32 to Busgang Decl. (Docket # 333-32). Gregori wrote back that he had “talked with [manager] and John [Micklethwait] and they are fine to fix the discrepancies.” E-mail from Reto Gregori, dated Apr_S, 2019, annexed as Ex. 32 to Busgang Decl. (Docket # 333-32) (“April 5, 2019 E-mail”). He proposed changes to four employees’ pay. See id. In another e-mail, Gregori told Sullivan to make changes to 13 employees’ pay. See E- mail from Reto Gregori, dated July 8, 2020, annexed as Ex. 36 to Busgang Decl. (Docket # 333- 36) (“July 8, 2020 E-mail”). In another e-mail, Gregori wrote to Collins, “Looking at peer groups — [female employee] feels underpaid considering the value she brings. [BOCS] recommendation feels a bit meager.” E-mail from Reto Gregori, dated Jan. 14, 2022 (8:43 AM), annexed as Ex. 36 to Busgang Decl. (Docket # 333-55). Collins wrote back, “Got it.” E-mail from Chris Collins, dated Jan. 14, 2022 (8:49 AM), annexed as Ex. 36 to Busgang Decl. (Docket # 333-55). In a subsequent message, Collins wrote to Gregori to say “also re this, algo gives [male employee] a vs I think he asked you for i? We can make the change[.]” E-mail from Chris Collins, dated Jan. 14, 2022 (10:46 AM), annexed as Ex. 36 to Busgang Decl. (Docket # 333-55). Gregori responded, “Yeah that’s fine.” E-mail from Reto Gregori, dated Jan. 14, 2022 (3:47 PM), annexed as Ex. 36 to Busgang Decl. (Docket # 333-55).

There 1s also an e-mail responding to a manager’s explanation of her adjustments to various employees’ pay. Gregori documented further changes he made to the compensation of nine employees. See E-mail from Reto Gregori, dated Jan. 18, 2022 (11:36 AM), annexed as Ex. 56 to Busgang Decl. (Docket # 333-56) (“Jan. 18, 2022 E-mail”). The manager replied: Just went through. Looks good considering I am now deeply in the red, but for reasons I’m all in agreement with. The only other things I should mention is we no longer seem to be meeting the algo on [female employee] and possibly that’s intention since we wanted to shift $ to well-deserving women. The only other thing is [manager] was pushing to pad out [another female employee]. I resisted in the spirit of favoring algo but her argument was that she’s incredibly solid as writers go and had a good year with cover, the pay check. E-mail from Jacqueline Simmons, dated Jan. 18, 2022, annexed as Ex. 56 to Busgang Decl. (Docket # 333-56). Gregori then wrote, “May have to claw back a few of those increases. Looking pretty red overall. But will wait for everyone to submit[.]” E-mail from Reto Gregori, dated Jan. 18, 2022 (6:08 PM), annexed as Ex. 56 to Busgang Decl. (Docket # 333-56) (“2d Jan. 18, 2022 E-mail”). Further, Ndugga states that “Defendant redacted from its production entire discussions of performance evaluation and compensation decision-making for women reporters, editors, and all but a subset of producers. To avoid a dispute requiring Court intervention, Plaintiff made a compromise proposal that Defendant re-produce a random sampling of only 35% of these emails.” Mem. at 16 n.68 (emphasis in original). However, Bloomberg states that “[e]ven though the parties agreed that BLP would unredact 35% of documents with redactions concerning putative class members’ ratings or compensation . . . [o]f the documents that did involve Mr. Gregori, almost all of them were unredacted pursuant to this compromise,” save for three — “all of which Plaintiff submitted with her Motion.” Opp. at 14-15.

3. Starting Pay Starting pay for each new hire was set by managers and recruiters with input from the comp team and in dialogue with the new hire. See Jennings Dep.: Bloomberg Excerpts at 168:8- 23, 171:17-24. But Gregori “approved all” salary offers “before they went out.” Id. at 175:23- 176:2; accord Siegel Dep.: Ndugga Excerpts at 68:13-17. While Ndugga argues that “Gregori regularly made changes to starting offers,” she points to only one instance in which Gregori made a change to a salary offer: her own. Mem. at 18. Gregori suggested setting Ndugga’s salary offer “slightly lower or on par to what we pay the rotators (she didn’t make the grade).” E-mail from Reto Gregori, dated Jan. 9, 2018, annexed as Ex. 21 to Busgang Decl. (Docket # 333-21) (“January 9, 2018 E-mail”). E. Expert Reports Ndugga retained Dr. David Neumark, a labor economist, to “evaluate evidence of pay discrimination against women . . . at Bloomberg.” Declaration of David Neumark, dated July 1, 2024, annexed as Ex. 66 to Busgang Decl. at *2-*3 (Docket # 333-66), 1, 2. Neumark used “data produced by Bloomberg that contain employment history and compensation records for the employees in the proposed Classes” to “construct a dataset of annual employee records.” Neumark Report {ff 9, 10. “In addition to pay,” the data used by Neumark “include[d] an indicator for the gender of an employee, and characteristics of the individuals and their jobs.” Id. Neumark analyzed this dataset to ““compare[] compensation at Bloomberg for similarly-situated female and male employees.” Id. § 12. Neumark created a regression model to “estimate[] the female pay penalty (if there is one) once we adjust for possible differences between female and male employees that could account for a pay gap between women and men.” Id. ¥ 13. He explains:

For example, suppose that we simply compare average pay of all female and male employees at Bloomberg, and find that average pay of female employees is 10% lower. It is possible that women do different jobs, and those jobs could pay less. It is also possible that women and men are in broadly similar jobs, but the women have lower performance. In either case, our intuition would be that the 10% estimate overstates the pay gap for comparable women and men in comparable jobs, and we should hence adjust for these differences between women and men before estimating the female pay penalty. Of course, the opposite is also possible, so adding controls for the individual or job could lead to a larger or smaller estimated female pay penalty. This 1s precisely what a regression model does. A regression model “holds constant” or “controls for” these other factors. These phrases mean that, in estimating a regression model, we adjust the pay gap for differences in the jobs employees hold, and their characteristics and performance, so that we are comparing pay between women and men with similar performance, education, and experience. Id. 13-14. Accordingly, “[t]o control for differences in characteristics and jobs between men and women,” Neumark’s regression model “include[d] several variables: race, experience since hire at Bloomberg, potential experience prior to hire at Bloomberg, education degree, employment city, Job Profile, year-end performance rating score, Cost Center, and Business Unit,” along with whether someone had part-time status. Id. § 37; see id. 37(1). It did not include peer groups as a variable for several reasons, including that “the first Peer Group classification in the [dataset] appears in June 2019.” Id. § 38(a). Further, Neumark did not “identifi[y] any factor used to construct Peer Groups” that his regression model “d[id] not separately control for.” Id. Of particular relevance to the instant motion is the “Cost Center” variable. Cost Center is an “organizational category to which costs” (such as salaries) “can be charged.” Id. ¥ 30; see Jennings Dep.: Bloomberg Excerpts at 46:11-19. Neumark acknowledged that “Cost Center does not play a role in compensation guidelines” and thus that “there is no rationale to disaggregate by Cost Center in doing my analysis. However, raises and hence pay may differ by

Cost Center since different Cost Centers may have different amounts available for raises.” Neumark Report 4 30 (footnote omitted). In response to criticism from Bloomberg’s expert, Martin, “Neumark refined his analysis and produced supplemental tables (correcting for issues in some instances).” Mem. at 20 n.83; see Neumark Supp. For the proposed U.S. Class, Neumark’s supplemental regression analysis found that female employees’ total compensation was 3.1% below that of similarly situated male employees (a difference of 1.64 standard deviations). See Neumark Supp. at *4 tbl. 2. For the proposed New York Class, Neumark found that female employees’ total compensation was 4.4% below that of similarly situated male employees (a difference of 2.29 standard deviations). See id. Ndugga relies on these findings in her memorandum of law in support of class certification. See Mem. at 19-22. The “standard deviations metric” tests “whether the measured difference in pay between women and men is statistically significant and differs from a hypothetical null hypothesis of gender-neutral pay setting . .. which is what we would expect in the absence of discrimination.” Neumark Report § 20. When the measured difference in pay is farther from this null hypothesis (1.e., more standard deviations from the null hypothesis), it is less likely that the pay difference “is due to chance.” Id. Neumark’s original expert report explains, “A disparity of two standard deviations is generally sufficient to show that a result is extremely unlikely (less than a 5% probability) to be caused by chance.” Id. ¥ 21. “Labor economists and econometricians more broadly generally regard any disparity of two or more standard deviations to be ‘statistically significant.’” Id. § 22.

In his original expert report — 1.e., before he created his supplemental regression model in response to Martin’s criticism — Neumark concluded “to a reasonable degree of expert certainty that women reporters, editors, and producers at Bloomberg News were paid less than similarly-situated men during the analyzed periods.” Id. 56. F. Relevant Procedural Background Ndugga filed the operative complaint on July 10, 2024. See Compl. On December 20, 2024, Ndugga filed the instant motion. See Mot. In her memorandum of law in support of class certification, Ndugga relied heavily on Neumark’s findings, see Mem. at 19-22, particularly in order to satisfy the commonality requirement of Fed. R. Civ. P. 23(a), see id. at 33-35. On February 11, 2025, Bloomberg requested a pre-motion conference in connection with an anticipated Daubert motion to exclude Neumark’s opinions and testimony from consideration for purposes of the instant motion. See Letter from Elise M. Bloom, filed Feb. 11, 2025 (Docket # 340). The Court initially determined that it would be best to resolve the admissibility of Neumark’s opinions and testimony before the parties submitted any more arguments on class certification. See Order, dated Feb. 12, 2025 (Docket # 343). Accordingly, the Court set a deadline for the filing of Bloomberg’s Daubert motion. Id. After that motion was briefed, however, the Court changed course and determined that it would be best to decide the Daubert motion at the same time as the instant motion. See Order, entered June 9, 2025 (Docket # 374). Briefing on the instant motion therefore proceeded. On June 23, 2025, Bloomberg filed an opposition brief. See Opp. On July 14, 2025, Ndugga filed her reply brief. See Reply. The next day, Bloomberg submitted a letter “to request a pre-motion conference in connection with its anticipated motion to strike” certain material

submitted alongside Ndugga’s reply brief. See Letter from Elise M. Bloom, dated July 15, 2025 (Docket # 388), at *1. In the alternative, it “request[ed] the Court’s permission to file a sur- reply.” Id. at *3. The Court granted leave to file a sur-reply, which Bloomberg filed. See Sur- Reply. On August 18, 2025, the Court held oral argument. See Transcript of Oral Argument, filed Sept. 16, 2025 (Docket # 412). At the Court’s request, the parties filed supplemental letters. See Batten Letter; Webber Letter. Il. APPLICABLE LAW The requirements for certifying a class are set forth in Fed. R. Civ. P. 23. Rule 23(a) provides that a class may be certified where: (1) “the class is so numerous that joinder of all members is impracticable”; (2) “there are questions of law or fact common to the class”; (3) “the claims or defenses of the representative parties are typical of the claims or defenses of the class”; and (4) “the representative parties will fairly and adequately protect the interests of the class.” Fed. R. Civ. P. 23(a). Courts have also read an “implied” requirement of “ascertainability” into Rule 23. Brecher v. Republic of Argentina, 806 F.3d 22, 24 (2d Cir. 2015) (citation and internal quotation marks omitted). In addition to satisfying Rule 23(a), a class must fall into at least one of the three categories identified in Rule 23(b) to be certified. See Fed. R. Civ. P. 23(b). Ndugga contends that this case belongs to the category listed at Fed. R. Civ. P. 23(b)\(3). See Mem. at 24. Under this provision, the Court must find “that the questions of law or fact common to class members predominate over any questions affecting only individual members, and that a class action is superior to other available methods for fairly and efficiently adjudicating the controversy.” Fed. R. Civ. P. 23(b)(3).

“Certain requirements for class certification, including . .. commonality . . . can be assessed only in light of the substantive law governing a case.” Kassman v. KPMG LLP, 416 F. Supp. 3d 252, 274 (S.D.N.Y. 2018). Title VII provides, in relevant part, that an employer may not (1) “discriminate against any individual with respect to his compensation, terms, conditions, or privileges of employment, because of such individual’s . . . sex” or (2) “limit, segregate or classify employees in any way which would deprive or tend to deprive any individual of employment opportunities, or otherwise adversely affect his status as an employee, because of such individual’s ... sex[.]” 42 U.S.C. § 2000e-2(a)(1)-(2). As is discussed further, Title VII allows for claims of disparate treatment or disparate impact. E.g., Murphy v. New York City Police Dep’t, 2025 WL 3493692, at *2 (S.D.N.Y. Dec. 5, 2025).° The Second Circuit gives Rule 23 a “liberal rather than restrictive construction, and courts are to adopt a standard of flexibility.” Marisol A. v. Giuliani, 126 F.3d 372, 377 (2d Cir. 1997) (citation and internal quotation marks omitted). But “[a] party seeking class certification must affirmatively demonstrate his compliance with the Rule — that is, he must be prepared to prove that there are in fact sufficiently numerous parties, common questions of law or fact, etc.” Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338, 350 (2011) (emphasis in original). “Sometimes the issues are plain enough from the pleadings . . . and sometimes it may be necessary for the court to probe behind the pleadings before coming to rest on the certification question.” Gen. Tel. Co. of Sw. v. Falcon, 457 U.S. 147, 160 (1982). “Merits questions may be considered to the extent — but only to the extent — that they are relevant to determining whether the Rule 23

> The parties exclusively refer to case law focused on Title VII in the sections of their briefs concerning Rule 23’s commonality requirement. We thus assume that there would be no difference with respect to this issue as to claims arising under the NYSHRL.

prerequisites for class certification are satisfied.” Amgen Inc. v. Connecticut Ret. Plans & Tr. Funds, 568 U.S. 455, 466 (2013) (citations omitted). “In deciding a motion for class certification under Rule 23, the district judge must receive enough evidence, by affidavits, documents, or testimony, to be satisfied that each Rule 23 requirement has been met.” Shahriar v. Smith & Wollensky Rest. Grp., Inc., 659 F.3d 234, 251 (2d Cir_2011) (citation and internal quotation marks omitted). “[T]he preponderance of the evidence standard applies to evidence proffered to establish Rule 23’s requirements.” Teamsters Loc. 445 Freight Div. Pension Fund v. Bombardier Inc., 546 F.3d 196, 202 (2d Cir_2008). Il. DISCUSSION In her motion for class certification, Ndugga asserts: Reto Gregori has had his fingerprints on every stage of the evaluation and compensation process for BLP’s reporters, editors, and producers. The result has been lower pay for women than their male peers. Because the record establishes the requirements of Rule 23, the Court should grant Plaintiff's motion to pursue pay discrimination claims on behalf of the proposed classes. Mem. at 25. The Court therefore assesses whether Ndugga has met her burden of demonstrating that the requirements of Rule 23 are satisfied. “As is true in many Title VII class actions,” Kassman, 416 F. Supp. 3d at 274, “the crux of this case is commonality — the rule requiring a plaintiff to show that ‘there are questions of law or fact common to the class,’” Dukes, 564 ULS. at 349 (quoting Fed. R. Civ. P. 23(a)(2)). To satisfy the commonality requirement, class members’ “claims must depend upon a common contention — for example, the assertion of discriminatory bias on the part of the same supervisor.” Id. at 350. “That common contention, moreover, must be of such a nature that it is capable of classwide resolution — which means that determination of its truth or falsity will resolve an issue that is central to the validity of each one of the claims in one stroke.” Id. In the

employment discrimination context, where a class wishes to sue about many “employment decisions at once,” Rule 23(a)(2) requires the plaintiff to identify “some glue holding the alleged reasons for all those decisions together.” Id. at 352. If she does not identify this glue, then “it will be impossible to say that examination of all the class members’ claims for relief will produce a common answer to the crucial question why was I disfavored.” Id. (emphasis in original). “Rule 23 does not set forth a mere pleading standard.” Id. at 350. Accordingly, the plaintiff must “affirmatively demonstrate” at the class certification stage that some glue “in fact” exists. Id. She must make this showing by a preponderance of the evidence. See Bombardier, 546 F.3d at 202. The Court’s “rigorous analysis” of this showing “will entail some overlap with the merits of the plaintiff’s underlying claim[s].” Dukes, 564 U.S. at 351. Precisely because “the class determination generally involves considerations that are enmeshed in the factual and legal issues comprising the plaintiff’s cause[s] of action,” we pause now to consider Ndugga’s causes of action. Falcon, 457 U.S. at 160 (citation and internal quotation marks omitted). As relevant to the instant motion, she purports to represent two classes of female employees who “bring claims for pay discrimination under disparate treatment and disparate impact theories.” Mem. at 26. A disparate treatment claim requires proof that “discrimination was the company’s standard operating procedure — the regular rather than the unusual practice.” Int’| Bhd. of Teamsters v. United States, 431 U.S. 324, 336 (1977).* For a disparate treatment claim, “statistical” or “anecdotal” evidence of discrimination can “provide precisely the ‘glue’ of

* Plaintiff brings what is known as a “pattern-or-practice” disparate treatment claim on behalf of the proposed classes, see Compl. {| 167, 183, as opposed to an individual disparate treatment claim, which courts analyze under the framework of McDonnell Douglas Corp. v. Green, 411 U.S. 792 (1973), see United States v. City of New York, 717 F.3d 72, 83 (2d Cir.2013).

commonality that [Dukes] demands.” Brown v. Nucor Corp., 785 F.3d 895, 914 (4th Cir_2015) (citation omitted); see also Burgis v. New York City Dep’t of Sanitation, 798 F.3d 63, 69 (2d Cir. 2015) (collecting Second Circuit decisions which have “hinted” that “statistics alone may be sufficient” to establish disparate treatment); Kassman, 416 F. Supp. 3d at 281 (noting that disparate treatment may be established “through statistics alone” and that “anecdotal evidence relevant to show a policy of intentional discrimination . . . [but] not required”) (citations omitted). A disparate impact claim, by contrast, requires proof of (1) “a specific employment practice,” (2) “a disparity,” and (3) “a causal relationship between the two.” Chin v. Port Auth. of New York & New Jersey, 685 F.3d 135.151 (2d Cir_2Q12) (citations and internal quotation marks omitted). For a disparate impact claim, “[a] companywide policy could serve as glue [under Dukes] if it affected all class members and caused a disparate impact.” In re Countrywide Fin. Corp. Mortg. Lending Pracs. Litig., 708 F.3d 704, 707 (6th Cir_2013). A classic example of such a policy is a hiring test that disproportionately excludes members of a protected group and bears no relationship to the jobs for which it is used. See Griggs v. Duke Power Co., 401 U.S. 424, 431 (1971). The disparate impact claims here do not revolve around a hiring test, however. In Dukes, the Supreme Court held if the policy under attack case is one of “allowing discretion

... over employment matters,” then Rule 23(a)(2) demands at a minimum that a class be subject to “a common mode of exercising discretion that pervades the entire company.” Dukes, 564 U.S. at 355, 356. Thus, the mere fact that the defendant has a policy of affording managers discretion to decide employees’ pay is insufficient to satisfy the commonality requirement; rather, there must be “some other glue . . . holding these individual decisions together.” In re Countrywide Fin. Corp. Mortg. Lending Pracs. Litig., 708 F.3d at 708.

Here, as explained further below, Ndugga has not proffered sufficient evidence of discrimination to meet the commonality requirement. She also has failed to demonstrate a common mode of exercising discretion that pervades Bloomberg. A. Evidence of Discrimination The question we address on the issue of commonality is whether Ndugga has put forward sufficient evidence of discrimination to establish that “examination of all the class members’ claims for relief will produce a common answer to the crucial question why was I disfavored.” Dukes, 564 U.S. at 352 (emphasis in original). “Tn a disparate treatment case, proof of commonality for purposes of Rule 23 overlaps with [the plaintiff’s] prima facie burden to show that [the defendant] engaged in a systemwide pattern or practice of discrimination.” Kassman, 416 F. Supp. 3d at 281 (citing Dukes, 564 U.S. at 352). She “may satisfy [her] burden of showing a pattern or practice of discrimination through statistics alone” inasmuch as “anecdotal evidence” is “relevant” but “not required.” Id. (citing Bhd. of Teamsters, 431 U.S. at 339-40, and City of New York, 717 F.3d at 84). However, “the defendant might endeavor to show that the plaintiffs statistics are inaccurate, for example, infected with arithmetic errors, or lacking in statistical significance” to “rebut the inference of a discriminatory intent.” City of New York, 717 F.3d at 85. Statistics have a role to play in a disparate impact case as well. A disparate impact claim requires, inter alia, proof of “a disparity,” Chin, 685 F.3d at 151, and so the plaintiff may “make out a prima facie claim of disparate impact through statistical evidence,” Chen-Oster v. Goldman, Sachs & Co., 325 F.R.D. 55, 81 n.15 (S.D.N.Y. 2018); see also Mandala v. NTT Data, Inc., 975 F.3d 202, 209 (2d Cir. 2020) (observing that “to ultimately prove” a disparate impact claim, “plaintiffs typically rely on statistical evidence to show a disparity in outcome between

groups”). “To rebut a plaintiff’s statistics, a defendant may introduce evidence showing that ‘either no statistically significant disparity in fact exists or the challenged practice did not cause the disparity.’” Chin, 685 F.3d at 151 (quoting Robinson v. Metro-North Commuter R.R. Co., 267 F.3d 147, 161 (2d Cir. 2001). As the Second Circuit has cautioned, “[t]here is no minimum statistical threshold requiring a mandatory finding that a plaintiff has demonstrated a violation of Title VII. Courts should take a ‘case-by-case approach’ in judging the significance or substantiality of disparities, one that considers not only statistics but also all the surrounding facts and circumstances.” Waisome v. Port Auth. of New York & New Jersey, 948 F.2d 1370. 1376 (2d Cir_1991) (quoting Ottaviani v. State Univ. of New York at New Paltz, 875 F.2d 365, 372-73 (2d Cir_1989) and citing Int’] Bhd. of Teamsters, 431 U.S. at 340); see also Watson v. Fort Worth Bank & Tr., 487 US. 977, 995 n.3 (1988) (observing that “we have not suggested that any particular number of ‘standard deviations’ can determine whether a plaintiff has made out a prima facie case in the complex area of employment discrimination”). Nonetheless, a court should “look[] to whether the plaintiff can show a statistically significant disparity of two standard deviations,” even though a court may accept less-robust evidence where a “lack of statistical significance . . . reflects only [a] small sample size.” Chin, 685 F.3d at 153 (first quoting Smith v. Xerox Corp., 196 F.3d 358, 365 (2d Cir_1999); then quoting Waisome, 948 F.2d at 1379). In any case, “an equivocal statistical analysis or an analysis that ‘obfuscate[s] the principal explanatory variable’ to create the appearance of difference” will not supply the glue necessary to hold class claims together. Richardson v. City of New York, 2021 WL 1910689, at *9 (S.D.N.Y. May 12, 2021) (quoting Kassman, 416 F. Supp. 3d at 263).

Here, Ndugga asserts that “[h]er expert’s statistical analyses . . . establish commonality for Plaintiff's disparate treatment and impact claims.” Mem. at 33 (emphasis added). She writes, Dr. Neumark found that Gregori’s decision-making resulted in both classes of women being paid statistically significant amounts less than similarly situated men — with less than a 5% chance that there was no difference in pay for the New York class (2.29 standard deviations) and a 10% chance for the U.S. class (1.64 standard deviations). Id. at 34. Specifically, Neumark found that women in the New York Class “were paid 4.4% less than similarly situated men, a statistically significant difference of 2.29 standard deviations,” id. at 20, and that women in the U.S. Class “were paid 3.1% less than similarly situated men, a difference of 1.64 standard deviations,” id. at 21. In an opinion issued today, the Court rules that Neumark’s findings are admissible under Fed. R. Evid. 702 and Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). Bloomberg contends that “[e]ven 1f Dr. Neumark’s opinions are not excluded, the Court should ‘safely disregard’ them as insufficient to establish commonality, much like the expert opinions in Dukes.” Opp. at 24-25 (quoting Dukes, 564 U.S. at 354). Bloomberg argues (1) that “Neumark’s opinions, even when accepted at face value with all of their flaws, do not provide ‘substantial proof’ of sex discrimination common to the proposed U.S. Class,” id. at 25; (2) that that ““Neumark’s analyses improperly rely on the inclusion of Cost Center, an accounting category that does not affect employee compensation, as a control variable,” id. at 27; and (3) that, as concems the disparate impact claims in this case, Neumark offers no “explanation of

what specific subjective employment decision(s) (by Mr. Gregori or otherwise) caused the alleged disparities,” id. at 28. The Court next addresses these arguments.° 1. For the Proposed U.S. Class, Plaintiff’s Evidence Is Not Statistically Significant As stated above, women in the proposed U.S. Class “were paid 3.1% less than similarly situated men, a difference of 1.64 standard deviations.” Mem. at 21. Bloomberg notes that this finding “is below the level that courts (and econometricians) generally deem to be statistically significant.” Opp. at 25. Thus, Bloomberg contends that this finding “does not establish commonality.” Id. The Court agrees. “A finding of two standard deviations corresponds approximately to a one in twenty, or five percent, chance that a disparity is merely a random deviation from the norm, and most social scientists accept two standard deviations as a threshold level of ‘statistical significance.’” Ottaviani, 875 F.2d at 371 (citations omitted). However, “where statistics are based on a relatively small number of occurrences, the presence or absence of statistical significance is not a reliable indicator.” Waisome, 948 F.2d at 1379 (citations omitted). In these cases, “other indicia raising an inference of discrimination must be examined.” Id. (citations omitted). For instance, in Chin, the Second Circuit confronted statistical evidence that “compared the promotion rate for whites who were on the eligible lists to the promotion rate for Asian Americans who were on the eligible lists.” Chin, 685 F.3d at 144. Thirty-six of 256 eligible whites were promoted; zero of 12 eligible Asian Americans were. See id. The plaintiffs’ expert “calculated that the likelihood this disparity would occur due to chance was about thirteen percent.” Id. “In many (perhaps most) cases,” Chin observed, “if there is a 13-percent likelihood that a disparity resulted from

> Bloomberg launches other attacks on the sufficiency on Ndugga’s evidence of discrimination which the Court need not reach.

chance, it will not qualify as statistically significant,” but “requiring . . . 95-percent confidence would make it mathematically impossible to rely upon statistics in a case like this one, in which the relevant population included so few Asian Americans.” Id. at 153. Thus, Chin turned to “other evidence that reasonable jurors could have relied upon to find that an 87-percent likelihood that the disparity was not due to chance qualified as significant.” Id. at 153. This included evidence that the Asian-American plaintiffs “were more qualified than some of the white officers who were promoted, including comparing length of service, attendance records, and disciplinary histories.” Id. Unlike Chin, we confront a relatively large pool of data: 750 compensation records for members of the proposed U.S. Class and male peers. See Neumark Supp. at *4 tbl. 2. Faced with large datasets such as this one, courts in this Circuit generally regard disparities of fewer than two standard deviations as statistically insignificant. For instance, Boyd v. Interstate Brands Corp., 256 F.R.D. 340 (E.D.N.Y. 2009) considered statistical evidence of the “supposed disparities in the promotion rates of African-American and non-African-American . . . employees” of the defendant, premised on hundreds of records. Id. at 360; see id. at 354-55. The plaintiffs’ expert found a difference in promotion rates equivalent to 1.94 standard deviations. Id. at 360 n.24. Boyd held that the plaintiffs “cannot get past the fact that [their expert] did not find a statistically significant disparity in promotion rates.” Id. at 361. Thus, Boyd concluded that the plaintiffs’ statistics “do not demonstrate common questions of fact because they do not tend to show that being African American has had a widespread effect on employees’ promotions.” Id.°

6 Ndugga argues that “Boyd found substantial errors in plaintiffs’ data set” where “[n]o such significant shortcomings in the data are present in Dr. Neumark’s analysis.” Reply at 12 n.10.

Here, the “difference of 1.64 standard deviations” that Neumark observed with respect to the U.S. Class, Mem. at 20, is — by Neumark’s own admission — not statistically significant, see Neumark Report § 22; see also Chin, 685 F.3d at 153. Additionally, the number of records in Neumark’s analysis (750), see Neumark Supp. at *4 tbl. 2, is significantly larger than the number considered too few for “the standard deviation rule” not to apply, Waisome, 948 F.2d at 1379; see id. (disregarding this rule for a sample of 75 promotions). Ndugga attempts to distinguish Waisome, arguing that “[w]hile there were 75 promotions [under scrutiny] in Waisome, there were far more records — it is not the case that the only records that count are those of the individuals promoted; records of all individuals who could have been promoted but were not also have to be evaluated.” Reply at 13. Thus, she urges that the number of records in Neumark’s analysis (750) should be compared to the total number of people “who took the written test” (617) in Waisome. Id. But Waisome specifically referred to the “the small number of promotions” (75) when it stated that “where statistics are based on a relatively small number of occurrences, the presence or absence of statistical significance is not a reliable indicator.” Waisome, 948 F.2d at 1379 (citations omitted). What is more, in Waisome, the plaintiffs “were able to point to . . . the written test, and show — using in that aspect of the case a sufficiently large sample size — that it resulted in a statistically significant disparity.” Id. This 1s precisely what Ndugga is unable to do. Moreover, Ndugga has not offered “other indicia raising an inference of discrimination.” Id.; see Chin, 685 F.3d at 153. Instead, she offers an additional analysis conducted by Newmark which “includ[ed] years prior to the class period” that produced statistically significant results.

However, the Boyd court explicitly took the plaintiffs’ expert’s “opinions at face value and disregard[ed] defendant's criticisms” before considering his findings. Boyd, 256 F.R.D. at 361.

Reply at 14. For support, Ndugga cites to Segar v. Smith, 738 F.2d 1249 (D.C. Cir_1984), which, in her words, “accepted the statistically significant evidence from one set of promotions to bolster the less significant evidence as to other promotions.” Id. But we find no warrant to rely on data from outside the class period in evaluating whether the U.S. Class satisfies the commonality requirement. See Eagle v. Vee Pak, Inc., 343 F.R.D. 552. 573 (N.D. Ill. 2023) (“Outside-the-class period evidence alone would not suffice to show that discriminatory conduct common to all class members occurred within the class period”). And Segar provides no support for considering “additional data predating the class period.” Reply at 14. Rather, Segar stands for the now-familiar proposition that where “plaintiffs . . . rely on a sample too small to generate statistically significant evidence,” they must proffer “enough other probative evidence exists to permit an inference of discrimination.” Segar, 738 F.2d at 1283; see Waisome, 948 F.2d at 1379. Ndugga has made no such proffer. In addition to data from outside the class period, she also offers e-mails from Gregori which purportedly demonstrate that he “was aware of the systemic inequities in pay” but only “ma[de] small changes in response to specific complaints or requests.” Mem. at 22-23 (citing E-mail from Reto Gregori, dated Mar. 8, 2019, annexed as Ex. 49 to Busgang Decl. (Docket # 333-49) (“March 8, 2019 E-mail”); E-mail from Reto Gregori, dated Apr. 2019, annexed as Ex. 48 to Busgang Decl. (Docket # 333-46) (“April 1, 2019 E- mail”); April 5, 2019 E-mail; E-mail from Reto Gregori, dated July 26, 2019, annexed as Ex. 45 to Busgang Decl. (Docket # 333-45) (“July 26, 2019 E-mail”); E-mail from Reto Gregori, dated Jan. 14, 2022, annexed as Ex. 52 to Busgang Decl. (Docket # 333-52) (“January 14, 2022 E- mail”)). Ndugga is doubtlessly correct that discrimination can be established where “an employer was aware of the adverse impact a particular practice would have, yet chose to follow the practice anyway.” Reply at 21; see also Chen-Oster, 325 F.R.D. at 77 (finding that plaintiffs

“provided ‘significant proof’ of discriminatory disparate treatment” via “proof that Defendants were aware of gender disparities and gender bias at Goldman Sachs, but did not adjust their policies or practices to account for the discrimination women faced,” in addition to statistical evidence). However, the e-mails Ndugga offers do not show that Gregori was aware of systemic gender-based pay disparities or chose to follow a practice of underpaying female employees. First, none fall within the class period for the proposed U.S. Class. Second, the e-mails demonstrate, at most, three instances over two years in which Gregori identified or was made aware of discrete gender-based pay gaps. In two of these instances, Gregori took steps to address these gaps. See March 8, 2019 E-mail; April 1, 2019 E-mail; April 5, 2019 E-mail; July 26, 2019 E-mail. In the third instance, Gregori indicated that he would “look at the pay gap” and “might take a handful of dollars back from [certain male employees].” January 14, 2022 E-mail. This rather thin anecdotal evidence should be contrasted with the internal audits produced by Goldman Sachs in Chen-Oster stating that performance ratings were “colored and impacted by gender differences,” along with sworn declarations from would-be class members attesting to their experiences of discrimination. Chen-Oster, 325 F.R.D. at 77 (citation and internal quotation marks omitted). Moreover, in Kassman, the plaintiffs demonstrated “[the defendant] was aware of a pay and promotions gap” and had “not completely eradicated the gap,” but this showing “did not amount to ‘significant proof? that [the defendant] ‘operated under a general policy of discrimination’” for purposes of the plaintiffs’ disparate treatment claim. 416 F. Supp. 3d at 284 (citing Dukes, 564 U.S. at 353). Thus, even if Ndugga had made a similar showing, Kassman makes clear that this is not enough (at least with respect to her disparate treatment claim on behalf of the proposed U.S. Class).

In conclusion, the Court finds that Neumark’s statistics regarding the proposed U.S. Class “do not demonstrate common questions of fact because they do not tend to show that being” female “has had a widespread effect on” compensation for this proposed class. Boyd, 256 E.R.D. at 360. 2. For Both Proposed Classes, Plaintiff’s Statistical Evidence Is Flawed Bloomberg argues that Neumark improperly included “Cost Center, an accounting category that does not affect employee compensation, as a control variable,” undermining the probative value of his findings. Opp. at 27. The Court finds this argument persuasive. Cost Center, as Neumark acknowledged, “does not play a role in compensation guidelines, so . . . there is no rationale to disaggregate by Cost Center.” Neumark Report § 30. But he included Cost Center on the theory that “raises and hence pay may differ by Cost Center since different Cost Centers may have different amounts available for raises.” Id. (emphases added). Neumark cited to four pieces of evidence in support of this statement. See id. § 30 n.77. None justify the inclusion of Cost Center. First, Neumark cited to a memorandum addressed to John Micklethwait and dated December 12, 2017. See Memorandum, dated Dec. 12, 2017, annexed as Ex. S to Declaration of Elise M. Bloom, filed Feb. 20, 2025 (Docket # 352-17) (“2017 Compensation Guidelines Mem.”). This memorandum explains that while the “2018 Target Total Compensation guideline increase is mm for [Bloomberg] News,” it is mm for six “Business Units which . . . fall under Media cost center view.” Id. at 1. However, this reference to the “Media cost center” provides no justification for the inclusion of Cost Center. For one thing, “Business Unit” is already included as a variable in Neumark’s analysis. Neumark Report § 37(h); Memorandum of Law in Support of Defendant’s Motion to Exclude the Opinions and Testimony of David Neumark, filed

Feb. 20, 2025 (Docket # 345) (“Bloomberg Daubert Mem.”) at 14. Thus, including Cost Center does not control for any potential difference in funding available for employee compensation not already accounted for by the inclusion of Business Unit. Moreover, the 2017 memorandum only mentions the “Media cost center” and thereby does not justify controlling for all 89 cost centers. See Neumark Report at 37-38 (listing all cost centers). Second, Neumark cited to Jennings’s testimony regarding the 2017 memorandum: Q. It looks like even though these individuals had a reporting relationship up to Mr. Micklethwait, that because they were in a different cost center, that made more money available to give raises to them; is that right? A. That is how it appears in the memo, that they were getting the full funding. Q. Okay. So, even if cost center didn’t dictate sort of who your ultimate manager was or necessarily distinguish your job in other ways, it could impact the amount of funding available for salary increases and bonuses at yearend? A. So, it seems that’s the case for 2017 .... I can’t speak to any other year because I haven't seen a similar memo. Jennings Dep.: Ndugga Excerpts at 191:2-191:20. Jennings’s testimony provides no additional justification for the inclusion of Cost Center. Her testimony simply reaffirms the plain understanding of the 2017 memorandum. As just discussed, the 2017 memorandum does not justify the inclusion of Cost Center. Third, Neumark cited to another memorandum addressed to John Micklethwait, this one dated December 12, 2021. See Memorandum, dated Dec. 12, 2021, annexed as Ex. R to Declaration of Elise M. Bloom, filed Feb. 20, 2025 (Docket # 352-16) (“2021 Compensation Guidelines Mem.”). However, as Bloomberg notes, see Bloomberg Daubert Mem. at 13, this particular memorandum refers to “2022 Target Total Compensation.” 2021 Compensation Guidelines Mem at 1. Because 2022 is outside the relevant time period for either putative class, the 2021 memorandum provides no justification for the inclusion of Cost Center. Certainly, the 2021 memorandum also covers “bonuses for work performed in 2021,” as Ndugga observes.

Opposition to Defendant’s Motion to Exclude the Opinions and Testimony of David Neumark, filed Mar. 13, 2025 (Docket # 358) at 20. However, the 2021 memorandum does not provide for different amounts of funding for these bonuses across Cost Centers. See 2021 Compensation Guidelines Mem. at 1. Fourth, Neumark cited to Sullivan’s testimony regarding the 2021 memorandum. Because of the 2021 memorandum’s limitations, discussed above, Sullivan’s testimony provides no further justification for the inclusion of Cost Center as an explanatory variable. At most, Sullivan’s testimony reaffirms that “different Cost Centers may have different amounts available for raises.” Neumark Report § 30. Separately, Ndugga references yet another memorandum to John Micklethwait, this one dated December 13, 2019. See Memorandum, dated Dec. 13, 2019, annexed as Ex. 17 to Busgang Decl. (Docket # 333-17) (“2019 Compensation Guidelines Mem.”). It states that “a had been allocated “for employees who had career progressions from June through November based on preliminary projections which are currently being finalized (pertaining to Business Units that show up under News in BOCS but fall under Media cost center view).” Id. at 1. This statement does not provide support for the inclusion of Cost Center for the same reasons given above: Business Unit 1s already included as a variable in Neumark’s analysis, see Neumark Report § 37, and although the 2019 memorandum mentions the “Media cost center,” it does not refer to the other 88 cost centers Neumark controlled for. See Neumark Report at 37-38 (listing all different Cost Centers). Additionally, Neumark does not appear to have considered

the 2019 memorandum. See Neumark Report § 30 n.77.’ Therefore, the 2019 memorandum cannot support Neumark’s decision to control for a variable that appears to be irrelevant. Accordingly, Neumark was not justified in controlling for Cost Center. To the extent that Cost Center reflects budgetary constraints influencing compensation decisions, those constraints are already accounted for by the inclusion of Business Unit. See Neumark Report § 37; Bloomberg Daubert Mem. at 14. Crucially, the inclusion of this irrelevant control variable had a material effect on Neumark’s conclusions, as Martin identified: when the variable Cost Center is excluded from the analysis, any pay disparity between men and women is “no longer statistically significant at either the 5% or the 10% level.” Martin Report § 23 (emphasis added). Accordingly, the inclusion of Cost Center turns Neumark’s analysis into the sort of “analysis that ‘obfuscate[s] the principal explanatory variable’ to create the appearance of difference.” Richardson, 2021 WL 1910689, at *9 (quoting Kassman, 416 F. Supp. 3d at 283). 3. Plaintiff’s Statistics Do Not Show That Any Pay Disparity Is Traceable to Gregori Separately, Bloomberg argues, and the Court agrees, that Ndugga has not provided adequate evidence that any disparity in pay is traceable to Gregori. See Opp. at 30-31. The proposed classes’ disparate impact claims identify Gregori’s role in the compensation process as the companywide policy responsible for pay discrimination at Bloomberg News. See Mem. at 29-30. For these claims, Ndugga must do more than point to a disparity in pay. See Richardson, 2021 WL 1910689, at *7 (“Of course, in the absence of a ‘specific employment practice’ pertinent across a putative class, merely pointing to ‘an overall [class]-based disparity does not

7 Neumark also apparently did not consider similar memoranda from 2018 and 2020 which, according to Bloomberg, “appl[ied] the same guideline increases for raises across the board to all News employees without regard to . . . Cost Center.” Bloomberg Daubert Mem. at 15.

suffice.’”) (quoting Dukes, 564 U.S. at 357); see also Campbell v. Nat'l R.R. Passenger Corp., 311 F. Supp. 3d 281, 319 (D.D.C. 2018) (“Statistical evidence may, of course, be used to prove discrimination on a disparate-impact theory. Dukes did not change this standard, but rather reiterated that statistical correlation cannot substitute for a specific finding of class-action commonality.”’) (citing Dukes, 564 U.S. at 356). She must also “isolate[e] and identify[] the specific employment practices that are allegedly responsible for any observed statistical disparities.” Watson, 487 ULS. at 994; accord Loc. 3621, EMS Officers Union, DC-37, AFSCME, AFL-CIO, 2022 WL 17175798, at *11. Ndugga states that “Gregori had his hands in all aspects of compensation decision- making, from starting pay through final determinations. . . . Thus, the statistics show that Gregori’s decision-making resulted in women’s systematic underpayment.” Reply at 20. She seems to be arguing that, because Gregori was widely involved in the compensation process, any wage differences resulting from this process are necessarily traceable to him. We reject this argument. Even if we accept that “Gregori had his hands in all aspects of compensation decision-making,” this does not establish that he is responsible for any pay disparity. Gregori’s role in the compensation process may constitute a companywide policy to which all proposed class members are subject, but Ndugga still has not shown that any disparity in pay necessarily traces to his actions — as opposed to, for example, the actions of lower-level managers. “Because [Ndugga] ha[s] failed to identify a specific employment process and to draw any causal connection between that employment process and any disparities,” her disparate impact claims cannot be certified classwide. Loc. 3621, 2022 WL 17175798, at *11. Ndugga contends that “there is no way to isolate Gregori’s role and estimate what pay would have been absent him.” Webber Letter at 6; see Reply at 15-16. Title VII provides that

“if the complaining party can demonstrate to the court that the elements of a respondent’s decisionmaking process are not capable of separation for analysis, the decisionmaking process may be analyzed as one employment practice.” 42 U.S.C. § 2000e-2(k)(1)(B)(). “Whether a particular decisionmaking process is capable of separation for analysis largely turns on the details of the specific process and its implementation in a given case.” Chin, 685 F.3d at 154 (citations omitted). Ndugga has not demonstrated that the elements of Bloomberg News’ compensation process are not capable of separation for analysis. Ndugga explains that limitations in Bloomberg’s production prevented her expert from analyzing gender-based disparities in starting salaries, but she does not explain why, for example, it was impossible to “isolate[] and identify[]” disparities in performance ratings or initial compensation recommendations generated by the algorithm. Watson, 487 US. at 994 B. Evidence of a Common Mode of Exercising Discretion Even if Ndugga had put forward sufficient evidence of discrimination, she has not met her burden of showing that a common mode of exercising discretion pervades Bloomberg News. Thus, it is “impossible to say that examination of all the class members’ [disparate impact] claims for relief will produce a common answer to the crucial question why was I disfavored.” Dukes, 564 U.S. at 352 (emphasis in original). Ndugga points to Gregori’s “decision-making” as the glue that holds the proposed classes’ disparate impact claims together. Mem. at 29. “All that matters at this stage,” she argues, “is whether all class members were subject to Gregori’s discretionary decision-making. The answer is resoundingly yes. This suffices for commonality.” Id. at 30 (citations omitted). Ndugga’s argument overstates the law and her own evidence.

As Kassman explained, if “a single decisionmaker[] or a small, cohesive group vetted all of the pay . . . decisions, then the involvement of that one individual, or small group of individuals, could constitute a common practice.” Kassman, 416 F. Supp. 3d at 280 (citations and internal quotation marks omitted) (alteration in original) (emphasis added); see also Jones v. Nat’| Council of YMCAs of the United States of Am., 34 F. Supp. 3d 896, 908 (N.D. III. 2014) (“In theory, if all pay . . . decisions had to be vetted by a single decisionmaker, or a small, cohesive group, then there might be a basis to argue that it is the performance of that function. . . constitutes the common practice among employees claiming disparate impact.”). However, “mere approval or limited oversight by higher-level executives, without more, falls short of showing a sufficient common denominator.” Kassman, 416 F. Supp. 3d at 280 (citing Jones, 34 F. Supp. 3d at 908). The involvement of a single decisionmaker (or a small, cohesive group) must be “sufficiently ‘consistent,’ ‘pervasive,’ or ‘classwide’ to produce common questions and answers across the putative classes.” Richardson, 2021 WL 1910689, at *8 (citing Ellis v. Costco Wholesale Corp., 285 F.R.D. 492, 511-12 (N.D. Cal. 2012)). Richardson elaborated: If a “close-knit, centralized management team” “oversees and directs” employment decision-making for an organization, [Ellis, 285 F.R.D. 511-14], ora small group of individuals is itself responsible for selecting candidates for employment, promotion, or a raise, see, e.g., In re Johnson, 760 F.3d 66, 69-70 (D.C. Cir_2014), that involvement will define the employment outcomes of all class members. To the extent those individuals’ decision-making 1s infected with bias, all class members are liable to have suffered unequal treatment. The same cannot be said when management’s involvement is limited to approving “the recommendations of lower level direct supervisors and managers,” and there is “no evidence showing that any member of [] management changed .. . recommendations . . . with any frequency.” [Jones 34 F. Supp. 3d at 908]. A toothless “final approval” process affords ample “freedom and independence to local supervisors” and dictates the conclusion that these varied supervisors, rather than a centralized management team, are the causal force behind any unequal outcomes. Kassman, 416 F. Supp. 3d at 280-81. Id. (second alteration in original).

4]

In other words, to satisfy the commonality requirement, it is not enough that proposed class members were “subject to” Gregori’s discretionary decision-making, 1.e., that Gregori could have made decisions about their pay. Mem. at 30. Gregori must in fact have made decisions about their pay “with a regularity that would establish [him] as a common denominator” across all proposed class members. Jones, 34 F. Supp. 3d at 908; see also Simpson v. Dart, 23 F.4th 706, 711 (7th Cir_2022) (“[A] court must know how that discretion is exercised to know whether there exists ‘a common mode of exercising discretion that pervades the entire company.””) (quoting Dukes, 564 U.S. at 356) (emphasis in original). Ndugga’s reply brief concedes that this is the showing she must make. See Reply at 3 “Ultimately, what matters is if Gregori decided.”) (emphasis in original). Here, Ndugga has not shown by a preponderance of the evidence that Gregori made decisions about pay “with a regularity that would establish [him] as a common denominator” across all class members. Jones, 34 F. Supp. 3d at 908. Ndugga’s moving brief identifies five strands of proof regarding the centrality of Gregori’s role in making decisions about pay. She writes, Gregori: (1) made or required others to make changes to individuals’ performance ratings; (2) altered the rules of the algorithm that BLP used to propose bonuses and pay adjustments; (3) directed changes in the resulting proposals at the outset of the compensation review — before managers even saw the algorithm’s proposals; and then (4) using his final review authority, made changes to bonus and new target compensation figures again at the end of the compensation process. Gregori also (5) made all starting pay decisions for each individual News employee. Mem. at 5-6. The Court considers these contentions in reverse order. Ndugga contends that Gregori “made all starting pay decisions for each individual News employee.” Id. at 6; see also id. at 18, 29. Ndugga has adduced deposition testimony indicating that Gregori approved starting offers. See Jennings Dep.: Bloomberg Excerpts at 175:23-176:2;

Siegel Dep.: Ndugga Excerpts at 68:13-17. These starting offers, which included starting pay, were prepared by some “combination” of “recruiters,” “business managers,” and comp team members — not by Gregori. See Jennings Dep.: Bloomberg Excerpts at 168:8-23, 172:14-22. Ndugga points to a number of e-mails from Gregori as proof that he “made decisions on the amount to offer as starting compensation for each new hire.” Mem. at 17-18 (citing E-mail from Reto Gregori, dated Aug. 31, 2017, annexed as Ex. 30 to Busgang Decl. (Docket # 333-30); E- mail from Reto Gregori, dated Nov. 25, 2020, annexed as Ex. 35 to Busgang Decl. (Docket # 333-35); E-mail from Reto Gregori, dated Jan. 10, 2018, annexed as Ex. 40 to Busgang Decl. (Docket # 333-40); E-mail from Reto Gregori, dated Nov. 6, 2017, annexed as Ex. 41 to Busgang Decl. (Docket # 333-41); E-mail from Reto Gregori, dated Sept. 27, 2017, annexed as Ex. 42 to Busgang Decl. (Docket # 333-42); E-mail from Reto Gregori, dated Apr. 8, 2019, annexed as Ex. 47 to Busgang Decl. (Docket # 333-47); E-mail from Reto Gregori, dated Nov. 29, 2018, annexed as Ex. 50 to Busgang Decl. (Docket # 333-50)). These e-mails capture Gregori’s responses to requests to approve starting offers. In each, Gregori gives his approval, writing only “OK,” “Ok,” “Fine,” or “fine.” See id. While these e-mails indeed demonstrate that Gregori approved starting offers, they equally establish that the recommendations of recruiters, business managers, and comp team members regarding starting pay “were almost always accepted.” Jones, 34 F. Supp. 3d at 908. Ndugga also asserts that “Gregori regularly made changes to starting offers.” Mem. at 18. But she only cites to a single instance of this practice — when Gregori changed her own starting offer. See id. (citing January 9, 2018 E-mail). In the end, this evidence does not show that Gregori “changed . . . recommendations submitted to [him] for approval with any frequency, much less that [he] did so with a regularity that would establish [him] as a common denominator.” Jones, 34 F. Supp. 3d at 908.

Ndugga asserts that Gregori, “using his final review authority, made changes to bonus and new target compensation figures . . . at the end of the compensation process.” Mem. at 5-6; see also id. at 17, 29. Specifically, Gregori “made 286 changes to current bonuses” and “389 adjustments to target Total Compensation” between 2018 and 2021. Id. at 17 n.70 (quoting Martin Report §48(e)). These numbers of course cannot be considered in a vacuum. Expressed as a percentage, Gregori’s interventions represent roughly “12% of the total number of compensation changes for this time period.”® Webber Letter at 2. This situation is akin to Valerino v. Holder, 283 F.R.D. 302 (E.D. Va. 2012), where a director considered recommendations by a “Career Board,” affirming them “more than eighty percent of the time.” Id. at 317. Valerino determined that the plaintiffs failed to satisfy the commonality and typicality requirements because the claims of many plaintiffs did not “depend upon a common contention,” 1.e., the director’s decision-making. Id. at 316 (quoting Dukes, 564 U.S. at 350). So too here: because Gregori made just 12% of the total number of compensation changes, 88% of proposed class members’ claims rest not on Gregori’s intervention at the end of the compensation process but on the intervention of whoever was responsible for the other 88% of changes. Mem. at 5. Thus, their claims do not “depend upon a common contention.” Dukes, 564 US. at 350. Ndugga does not squarely address Valerino except to label it “an extreme out- of-[C]ircuit outlier.” Reply at 7.

Ndugga argues that this figure represents an “undercount.” Mem. at 16 n.68. When Gregori “agreed with the algorithm’s recommendation or a manager’s proposed change,” she explains, “his choice would not leave a paper trail.” Id. at 16. But what matters is the nature of Gregori’s choice. If Gregori ““mere[ly] approv[ed]” a recommendation or a manager’s proposed change, then he would not glue the proposed classes’ claims together. Kassman, 416 F. Supp. 3d at 280. Instead, Ndugga must show that he “vetted” each recommendation or proposed change. Id. As explained below, she has not.

Ndugga also asserts that Gregori’s “decision-making role” was not limited to “the 12% of compensation changes he personally entered.” Id. at 3. To support this assertion, Ndugga contends that Gregori “directed changes” to the recommendations generated by Bloomberg’s algorithm “before managers even saw the algorithm’s proposals.” Mem. at 5; see also id. at 12- 13, 29. However, Ndugga points only to a single, irrelevant instance of Gregori directing such a change. See id. at 12-23 (citing January 4, 2021, E-mail). The employee whose initial compensation recommendation Gregori altered is neither a putative class member nor a male peer. See Opp. at 14. Ndugga also adduces deposition testimony indicating that Gregori “at times” directed changes to initial compensation recommendations before that recommendation was reviewed by any managers. See Sullivan Dep.: Ndugga Excerpts at 51:25-52:6. Even he did “at times” make such changes, Gregori would still not glue the proposed classes’ claims together: as we have already noted, 74% of all proposed class members’ pay was changed from the pay recommendation that was generated by the algorithm, and such changes were conducted by 126 managers. See Martin Report § 48(a), (b). So to the extent that Gregori did direct changes to the algorithm’s initial recommendations, his changes were overridden 74% of the time. While the group of 126 managers included Gregori, id. J 48(b), Ndugga provides no evidence that he had any appreciable influence at this phase in relation to the other 125 managers. In the end, if Gregori “recommended pay increases and bonuses” by directing changes to the algorithm’s initial recommendations, he “merely ‘set[] up a structure’ through which employees are evaluated.” Kassman, 416 F. Supp. 3d at 278 (quoting Valerino, 283 F.R.D. at 313). This “says nothing about whether the exercise of discretion” in adopting or adjusting those recommendations was “common.” Id. (quoting Valerino, 283 F.R.D. at 313).

Ndugga next contends that Gregori “altered the rules of the algorithm that BLP used to propose bonuses and pay adjustments.” Mem. at 5; see also id. at 10-15, 29. And the algorithm, she asserts, “also glues [her] class claims together, as it shapes Gregori’s exercise of discretion by providing a starting point for compensating all putative class members.” Id. at 32. The Court cannot find that the algorithm actually “glues [her] class claims together” when, as discussed, the initial compensation recommendations it generated did not stand 74% of the time. In any case, Ndugga has not proven that Grigori had a significant impact on the algorithm. Ndugga highlights one request from Gregori to change to the size of bonuses that the algorithm would recommend for employees who received certain overall performance ratings. See id. at 12 n.48 (citing January 4, 2021 E-mail). She does not identify any other changes to the rules requested by Gregori. She also cites deposition testimony indicating that his requests to “tweak[]” the algorithm were generally honored “as long as [Bloomberg’s human resources department] felt they were consistent with . . . the funding that was approved by [Bloomberg’s] management committee.” Sullivan Dep.: Ndugga Excerpts at 109:22, 110:3-6. But the fact remains that, even in the face of Gregori’s “tweaks,” lower-level managers, exercising their own discretion, set employees’ performance ratings. These ratings interacted with the rules which Gregori, exercising his own discretion, modified. In other words, “the process identified by [Ndugga] reveals only various stages or phases, during which individual decision-makers exercised their discretion.” Moussouris v. Microsoft Corp., 2018 WL 332841, at *198 (W.D. Wash. June 25, 2018) (citing Valerino, 283 F.R.D at 313-14), aff'd, 799 F. App’x 459 (9th Cir_2019). “Identifying th[is] framework says nothing about whether the discretion itself can ‘be said to be the same or “common.”’” Id. (quoting Valerino, 283 F.R.D at 313). Cf. Richardson, 2021 WL 1910689, at *8 (observing that a central decisionmaker’s policy of capping pay increases upon

promotion at 8% “has no bearing on a supervisor’s ability to select an employee for promotion or to recommend . .. say, a 2% or an 8% increase”). While Chen-Oster (cited by Ndugga, see Mem. at 32-33) holds that a process by which employees are rated by lower-level managers according to subjective criteria can qualify as a common mode of exercising discretion, see Chen-Oster, 325 F.R.D. at 73-75, the Court finds the contrary authority cited by Bloomberg more persuasive. See Jones, 34 F. Supp. 3d at 905 (“[A] policy directing managers to evaluate employees using subjective criteria” is “not a ‘common policy’ that can support maintenance of a class action.”); Ross v. Lockheed Martin Corp., 267 F. Supp. 3d 174, 198 (D.D.C. 2017) (“The mere fact that all . . . supervisors used the same allegedly ill-defined numerical rubric to grade employee performance . . . says nothing about how individual supervisors exercised what discretion was left to them, and after [Dukes], it is clear beyond cavil that such silence isn’t enough.”); Kassman, 416 F. Supp. 3d at 277-78 (A compensation process “depending on, among other factors, their specific position and market rates for that position, current salary and performance” is “more as a framework that dictates who will make discretionary decisions rather than how they will exercise their discretion.”). Ndugga also points to a one-time change Gregori made to the makeup of peer groups, see Mem. at 14 n.53, and to Gregori’s ability to modify stack rankings (although she cites to no instances of Gregori using this ability), see id. at 15 n.60. But this evidence is not strong enough to allow a finding that Gregori’s influence over the algorithm — a program he neither coded nor completely understood, see Gregori Dep.: Ndugga Excerpts at 106:7-13 — amounted to a common mode of exercising discretion. Ndugga next contends that Gregori “made or required others to make changes to individuals’ performance ratings.” Mem. at 5; see also id. at 7-8; 29. Ndugga cites to several

examples of Gregori ordering subordinates to re-rate employees under their supervision so that their ratings conformed to Gregori’s distribution guidelines.” See id. at 7 n.28 (citing June 21, 2017 E-mail; 2d July 23, 2018 E-mail; July 21, 2020 E-mail; December 17, 2021 E-mail). Nonetheless, Gregori only “occasionally” rejected subordinates’ ratings, and there is thus a dearth of evidence that would allow a finding as to how strictly Grigori enforced the guidelines. Id. at 73:3. Ndugga also surfaces one e-mail showing Gregori personally changing two employees’ ratings and another in which he suggested that a manager change an employee’s ratings. See Mem. at 7 n.29 (citing July 23, 2018 E-mail; 2d July 23, 2018 E-mail). These isolated instances are insufficient to contradict Gregori’s testimony that “if a manager has a clear view of the performance of a person[,] I’m not going to second[-]guess.” Gregori Dep.: Bloomberg Excerpts at 73:8-10. Ultimately, even if Gregori “made or required others to make changes to individuals’ performance ratings” as often as Ndugga implies, the fact remains that these ratings were just one variable of many fed into Bloomberg’s algorithm to generate initial compensation recommendations — recommendations which were changed 74% of the time. Mem. at 5. And accepting, arguendo, that Ndugga has demonstrated a common mode of exercising discretion with respect to performance ratings through Gregori’s influence, she still must “establish a causal relationship between” this practice and the pay discrimination alleged. Chin, 685 F.3d at 151. She has not made this showing: unlike in Chen-Oster, for example, she has not proffered any evidence that female employees systematically received lower performance ratings than their male counterparts. See Chen-Oster, 325 F.R.D. at 75. Accordingly, there is no basis for the Court to conclude that Gregori’s influence over performance ratings can provide

” As Bloomberg notes, Gregori did not set these guidelines alone. See Opp. at 6 n.32.

“common answer to the crucial question why was I disfavored.” Dukes, 564 U.S. at 352 (emphasis in original). Ndugga argues generally that “Gregori had both the first and last word on class members’ compensation.” Mem. at 6; see id. at 29. Ndugga’s reply brief clarifies the thrust of this contention: “Gregori was an active reviewer of all algorithm and managerial proposals on compensation; whether he agreed with the recommendation and made no change, or disagreed and did make a change, he was evaluating the proposals and making the decision.” Reply at 3 (footnotes omitted). Ndugga’s point is that even when Gregori was not actively modifying employees’ pay, he always held the role of a reviewer. In support of this contention, she cites to deposition testimony from Gregori, Jennings, and Sullivan that she contends shows that Gregori “reviewed all compensation proposals.” Id. at 4 (citing Gregori Dep.: Ndugga Excerpts at 128:8- 129:8; Jennings Dep.: Ndugga Excerpts at 290:4-291:23; Sullivan Dep.: Ndugga Excerpts at 26:3-20). But plaintiff's assumption that Grigori’s “review” is enough to support a showing of commonality begs the fundamental question: what was the nature of Grigori’s review? To establish that Gregori’s review amounted to a common mode of exercising discretion, plaintiff must show that he alone “vetted” the pay proposals that came across his desk. Kassman, 416 F. Supp. 3d at 280. But the deposition excerpts do not show this. They show, in Gregori’s words, that “[p]eople submit to me[,] and we” — referring to himself and Micklethwait — “review[,] and that’s the final recommendation.” Gregori Dep.: Ndugga Excerpts at 192:2-3. Jennings’s testimony actually makes clear that Grigori did not have the “last word on class members’ compensation.” Mem. at 6. She testified that “[Gregori] will submit the entire department’s BOCS up, so it basically leaves his jurisdiction[,] and it goes up to the next level.” Jennings Dep.: Ndugga Excerpts at 291:10-13. Sullivan’s testimony confirms that Gregori and

Mickelthwait were the final reviewers of decisions on pay within Bloomberg News, see Sullivan Dep.: Ndugga Excerpts at 26:14-20, and that “[t]here may be further review” by Bloomberg’s overall management committee, see id. at 26:21-27:19. The documentary evidence cited by plaintiff similarly fails to show that Gregori alone vetted the pay proposals that came across his desk. One email does show Gregori changing 13 employees’ pay without prompting from anyone. See July 8, 2020 E-mail. But the remaining emails show the involvement of others. In one e-mail, Gregori changed four employee’s pay in response to concerns raised by a subordinate and in consultation with two other managers. See April 5, 2019 E-mail. Another shows that he reviewed 21 employees’ pay with Micklethwait; in this e-mail, he offered to make “last-minute post-submission changes” to the employees’ pay if his interlocutor had any objections. July 26, 2019 E-mail. And then there is an e-mail in which Gregori makes changes to nine employees’ pay in dialogue with a manager. See Jan. 18, 2022 E-mail; 2d Jan. 18, 2022 E-mail. Certainly, these e-mails reveal that Gregori’s review amounted to more than the rubberstamping in evidence in Jones, 34 F. Supp. 3d at 908, or Kassman, 416 F. Supp. 3d at 280. But they come nowhere near showing a single decisionmaker who is “vett[ing]” all pay proposals. Kassman, 416 F. Supp. 3d at 280. Furthermore, they demonstrate that Grigori’s ability to intervene, while not “toothless,” also “afford[ed] ample ‘freedom and independence to local supervisors,’” who, by most appearances, worked with — not at the sole direction of — Gregori to finalize pay proposals. Richardson, 2021 WL 1910689, at *8 (quoting Kassman, 416 F. Supp. 3d at 280-81). Ndugga has not proven that Gregori is “the causal force behind any unequal outcomes.” Id. Put simply, she has not met her burden of demonstrating that Gregori is the glue that holds together the many employment decisions that led to the outcome she seeks to challenge.

IV. CONCLUSION Because Ndugga cannot meet the commonality requirement of Rule 23, Ndugga’s motion for class certification (Docket # 329) should be denied.

PROCEDURE FOR FILING OBJECTIONS TO THIS REPORT AND RECOMMENDATION Pursuant to 28 ULS.C. § 636(b)(1) and Rule 72(b) of the Federal Rules of Civil Procedure, the parties have fourteen (14) days (including weekends and holidays) from service of this Report and Recommendation to file any objections. See also Fed. R. Civ. P. 6(a), (b), (d). A party may respond to any objections within 14 days after being served. Any objections and responses shall be filed with the Clerk of the Court. Any request for an extension of time to file objections or responses must be directed to Judge Woods. Ifa party fails to file timely objections, that party will not be permitted to raise any objections to this Report and Recommendation on appeal. See 28 U.S.C. § 636(b)(1); Fed. R. Civ. P. 72; Fed. R. Civ. P. 6(a), 6(b), 6(d); Thomas v. Arn, 474 U.S. 140 (1985); Wagner & Wagner, LLP v. Atkinson, Haskins, Nellis, Brittingham, Gladd & Carwile, P.C., 596 F.3d 84, 92 (2d Cir_2010).

Dated: March 19, 2026 New York, New York

lex_. Seddon United States Magistrate Judge