MEMORANDUM AND ORDER
SAVAGE, District Judge.In this putative class action alleging race-based disparate treatment brought under 42 U.S.C. § 1981, the plaintiffs seek certification of a class of all African-Americans, union and non-union, employed by the defendant in its numerous plants throughout Pennsylvania. They allege that GPU, Inc. and its wholly-owned domestic subsidiaries discriminated against African-American employees in compensation, promotion, job assignments and training, which adversely affected their overall corporate success.
GPU contends that the plaintiffs cannot satisfy the commonality, typicality, and ade*84quacy requirements of Rule 23(a), and that the proposed class is not one of the types recognized by Rule 23(b). Fed. R. Civ. P. 23. Specifically, GPU argues that plaintiffs are attempting to pursue an “across the board” class that is now disfavored in employment discrimination cases. See General Telephone Co. of the Southwest v. Falcon, 457 U.S. 147, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982).
The plaintiffs have failed to demonstrate that the members of the proposed class were affected in the same way by the same intentional companywide discriminatory policies and practices. Instead, the plaintiffs’ proposed expansive class is composed of different classifications of employees, having different education levels, experience and tenure, who were employed in a variety of positions at the fifty different GPU-operated facilities located throughout Pennsylvania. Therefore, the plaintiffs’ motion for class certification fails for lack of commonality and typicality.
I. Background
A. GPU Organization
GPU, Inc., a holding company registered under the Public Utility Holding Company Act of 1935, was an international provider of energy-related infrastructure and services.1 It was comprised of several different entities, including the domestic utility subsidiaries implicated in this action, Metropolitan Edison Company (“Met Ed”), GPU Service (“GPU Service”), and Pennsylvania Electric Company (“Penn Electric”).2 These entities worked together under the name of GPU Energy to transmit and distribute electricity across Pennsylvania and New Jersey.3 Together these three companies employed about 2,900 employees at approximately 50 different facilities throughout Pennsylvania.4 Approximately 44 percent were nonexempt employees represented by labor unions, approximately 40 percent were exempt nonunion employees, and the remainder were both nonexempt and nonunion.5
In November 2001, GPU and FirstEnergy Corporation merged, with FirstEnergy as the surviving corporation. The domestic utility subsidiaries in this action became subsidiaries of FirstEnergy.6
B. The Plaintiffs
The original plaintiffs were five nonexempt union employees and one exempt employee. One union employee has since withdrawn as a class representative. Of the remaining five plaintiffs, one worked for Penn Electric and the others for GPU Service. All claim that discrimination affected their advancement in the company, resulting in a diminished capacity to earn more money at higher levels of responsibility. Each complains of discrimination in a different manner.
Les Jones, who began at GPU in 1981, is currently employed as a Senior Contracts Administrator, an exempt position with Fir-stEnergy.7 Jones alleges he was denied promotions that were subsequently awarded to white employees. He further contends that the salary increases and bonuses he received were not commensurate with his superior performance evaluations, as compared to bonuses and salary increases awarded to similarly situated white employees.8
Mark Burford has worked for GPU since 1984. He is an Electrical Construction Maintenance Technician, a nonexempt position governed by a collective bargaining agreement.9 He alleges that since 1996, he has been unsuccessfully requesting additional work-related training which was given to white co-workers.10
*85Glenn Montgomery was employed by Penn Electric from July 1978 until he was terminated in June 2000,11 for allegedly threatening and harassing a customer, and for using a company vehicle for personal business. Admitting the latter allegation, he contends that white employees were not disciplined for doing the same thing.12 The only named plaintiff employed by Penn Electric, he was a union member.13
John L. Greene, Jr. began at GPU in September 1993 and is currently employed as a General Utility Worker. He is a nonexempt employee and a member of a collective bargaining unit.14 He alleges that he was denied the opportunity to work as an electrician despite being qualified to do so; and, that if he had received that promotion, he would have earned additional compensation.15
Jason Tribue began working at GPU in April 1992 as a union employee covered by a collective bargaining agreement.16 He unsuccessfully sought promotion to Customer Service Representative three times.17 He alleges that GPU invoked a time-on-the-job prerequisite against him while ignoring it for similarly situated white employees.18 He was promoted to Repairman First Class, a nonexempt union position, in 2001, and resigned from FirstEnergy in 2002.19
C. Nature of Plaintiffs’ Claims
Plaintiffs claim to be pursuing a “pay” or “compensation and compensation-related practices” discrimination class.20 They assert that African-Americans were discriminated against in the areas of compensation, promotion, training, job assignments, and performance evaluations, affecting their overall corporate success.21 African-American employees, they allege, ultimately received less compensation than their white counterparts because they “were promoted less often, received lower performance evaluations, and/or were given lower salary grades (or lower base salaries within the same job grade).”22 They rely on statistical evidence which they contend demonstrates that GPU engaged in a pattern and practice of promoting African-American employees less often, had a standard operating procedure of paying African-Americans less compensation than white employees, failed to establish policies and practices that would ensure the implementation of GPU’s anti-discrimination policies, and failed to adequately monitor employment decisions.23
Plaintiffs claim that GPU decision making was centralized for compensation decisions and decentralized for promotions.24 They contend that even though GPU’s compensation decision making was centralized, the Human Resources Department (“HR”) did not adequately ensure that minorities were not being discriminated against in setting salaries.25 Plaintiffs allege that the decentralization of promotion decisions gave individual managers too much discretion.26 Promotion decisions, plaintiffs allege, were left to individual managers’ discretion which led to discrimination against minorities, despite HR’s oversight of the process and final selections. Plaintiffs claim that GPU failed to ensure that performance evaluations were not tainted by discrimination because it did not conduct separate statistical analyses of the eval*86uations to ensure that employment decisions based upon the performance evaluations were not tainted by racial discrimination. On top of the decision-making matrix, according to the plaintiffs, there was an overall lack of monitoring of all employment decisions that belied GPU’s stated commitment to diversity. In sum, plaintiffs maintain that GPU failed to adequately monitor all employment decisions to prevent discrimination against African-Americans.
II. Proposed Class Definition
A. Varying Class Definitions
The proposed class definition has changed several times during the course of the litigation. The complaint originally defined the class as “all current and former African-American employees of GPU who have been subject to one or more aspects of the systemic racial discrimination and harassment described in this Complaint.”27 Plaintiffs have since dropped their harassment claim and are now only pursuing compensation and promotion claims.28 In their opening memorandum in support of class certification, plaintiffs sought the following definition: “all Black employees of defendant” from January 1,1995 to December 31, 2001.29 Later in the same memorandum, the plaintiffs offered another definition that excluded nonexempt, that is, union employees, from the class and included only “all current and former Black exempt employees of defendant ... and all Blacks who sought such positions during that time period.”30
Plaintiffs’ Reply Memorandum suggested a fourth definition: “all Black employees of defendant ... at any time during the period January 1, 1995 to December 31, 2001 ... focus[ing] only on defendant’s discriminatory compensation and compensation-related practices (including defendant’s exempt promotion practices).”31 At oral argument, the plaintiffs returned to their original position that the proposed class is “all African American employees of Met Ed, Penn Electric and GPU, during the period January 1, 1995 through December 31, 2001.” 32 Given that the last definition is the same as the one originally proposed, it is the one I shall consider.33
The plaintiffs mix several employment areas where discrimination allegedly took place, denominating two as primary and the others as secondary. The two areas of primary “focus” are compensation and promotion, with other areas like “hiring, discipline, [and] training” having “some effect on Black employee compensation.”34 The plaintiffs argue that “inasmuch as compensation is an accurate measure of corporate success, it would be a misnomer to refer to this case as anything other than a pay discrimination class action.”35
B. Class Period and the Statute of Limitations Under § 1981
Some of the plaintiffs’ claims in this case are governed by a two-year statute of limitations, while others are controlled by a four-year statute of limitations. Which time period controls each claim depends upon whether the claim became actionable under § 1981— before or after December 1,1990.
In Patterson v. McLean Credit Union, the Supreme Court held that § 1981 provided a remedy only for discriminatory conduct occurring at the initial formation stage of the contract and conduct impairing the right to enforce contractual obligations through legal process. 491 U.S. 164, 179, 109 S.Ct. 2363, 105 L.Ed.2d 132 (1989). Discriminatory acts relating to the terms and conditions of em*87ployment that occurred after the employment relationship had been established were not actionable under § 1981. Id. at 179-80, 109 S.Ct. 2363.
In direct response to Patterson’s holding limiting the extent of § 1981’s “make and enforce” language, Congress enacted the 1991 Amendments to the Civil Rights Act. It expanded the scope of actionable discriminatory conduct under § 1981 to include the “performance, modification, and termination of contracts, and the enjoyment of all benefits, privileges, terms, and conditions of the contractual relationship.” Civil Rights Act of 1991, Pub.L. No. 102-166, Title I, § 101, 105 Stat. 1071 (1991). Consequently, discrimination occurring during the employment relationship became actionable under § 1981.
In the meantime, Congress had enacted 28 U.S.C. § 1658, which established a four-year statute of limitations for all claims arising under federal statutes enacted after December 1, 1990. Prior to the passage of § 1658, the settled practice for fixing the limitations period under § 1981 was to borrow the most relevant state law statute of limitation. Jones v. R.R. Donnelley & Sons, 541 U.S. 369, 371, 124 S.Ct. 1836, 158 L.Ed.2d 645 (2004) (citing Goodman v. Lukens Steel Co., 482 U.S. 656, 660, 107 S.Ct. 2617, 96 L.Ed.2d 572 (1987), which held that Pennsylvania’s two-year personal injury statute of limitations applied to § 1981 suits). This new statute of limitations was created in response to concerns about inconsistencies associated with borrowing state time limitations for federal claims that did not contain their own prescribed time period. See David D. Siegel, Practice Commentary, The 1990 Enactment of a Uniform Statute of Limitations on Federal Claims, 28 U.S.C.A. § 1658, at 238-1 (West 1994) (“Practice Commentary”).
In passing § 1658, Congress was responding to calls for a statute of limitations that would apply to all federal claims. However, it did not make § 1658 retroactive. Practice Commentary, at 241 — 42. Hence, § 1658 provides a uniform four-year federal statute of limitations only for federal claims legislatively created after December 1,1990.
In the aftermath of the legislation expanding the scope of § 1981 and the passage of § 1658, the question then became: what statute of limitations governs the newly-available claims relating to the privileges, terms and conditions of an employment contract? More specifically, do these newly-created § 1981 claims “arise under” the 1991 Amendments (meaning the § 1658 four-year statute of limitations would govern), or do such claims “arise under” the original § 1981 (meaning that the statute of limitations would be borrowed from state law)?
The Supreme Court answered in Jones v. R. R. Donnelley & Sons, 541 U.S. 369, 382-83, 124 S.Ct. 1836, 158 L.Ed.2d 645 (2004), holding that claims relating to the privileges, conditions and terms of employment would be governed by § 1658’s four-year statute of limitations because these claims were created by an Act of Congress enacted after 1990— the 1991 Amendments to the Civil Rights Act. Consequently, a racial discrimination claim alleging discriminatory conduct actionable under the original § 1981, that is, a violation of the pre-1990 version, is not controlled by § 1658’s four-year statute of limitations. Instead, it is governed by the relevant borrowed state law statute. Boyer v. Johnson Matthey, Inc., No. 02-8382, 2005 WL 35893, at *2-*3 (E.D.Pa. Jan. 6, 2005) (Surrick, J.) (applying a two-year statute of limitations to some claims and a four-year statute of limitations to other claims, depending on when they became actionable).
Because plaintiffs’ allegation that starting salaries were set in a racially discriminatory fashion was actionable under the original § 1981, it is not affected by § 1658, and a two-year statute of limitations, borrowed from state law, applies. The plaintiffs’ claims relating to alleged discrimination in performance evaluations, bonuses, and raises is governed by § 1658’s four-year statute of limitations because such claims address the terms and conditions of employment that were not actionable until after Congress enacted the 1991 Amendments to the Civil Rights Act. Cf. Jones, 541 U.S. at 383-84, 124 S. Ct. 1836 (holding that claims for hostile work environment, wrongful termination, and failure to transfer arose under the 1991 Act because the 1991 Act amended § 1981’s language to allow recovery for racially discrimi*88natory “termination of contracts and the enjoyment of all benefits, privileges, terms, and conditions of the contractual relation”).
Plaintiffs’ promotion claims are more complicated. In Patterson, the Court held that promotion discrimination claims were actionable under the original § 1981 if “the nature of the change in position was such that it involved the opportunity to enter into a new contract with the employer.” 491 U.S. at 185, 109 S.Ct. 2363. Consequently, “[o]nly where the promotion rises to the level of an opportunity for a new and distinct relation between the employee and the employer is such a claim actionable.... ” Id. This requires an inquiry into whether the claims involve allegations that the discriminatory actions of the defendants prevented plaintiffs from entering into a new contractual relationship with them. The plaintiffs appear to argue that promotions were, or should have been, a term and condition of their continued employment, as opposed to the start of an entirely new and discrete contractual relationship with GPU. Hence, because the plaintiffs’ promotion claims are of the type that became actionable after the passage of the 1991 Amendments, § 1658’s four-year statute of limitations applies.
Although Jones v. R.R. Donnelley & Sons applies to pending cases, Fernandez v. M & L Milevoi Mgmt., Inc., 357 F.Supp.2d 644, 651 (E.D.N.Y.2005); see also Verdin v. Weeks Marine Inc., No. 03-4571, 2005 WL 357006, at *3 n. 2, 124 Fed.Appx. 92 (3d Cir. Feb. 16, 2005) (unpublished opinion), the plaintiffs have not asked that I revisit Judge Waldman’s Order36 dismissing three class representatives because they had not alleged that any discriminatory action had been taken against them within two years preceding the filing of this action. In the end, however, determining the statute of limitations is a moot point in light of the ultimate determination of the certification motion. Nevertheless, the statute of limitations is a consideration in evaluating the proffered statistics discussed later in this opinion.
III. Class Certification Requirements
Before a putative class action can be certified, all four requirements of Federal Rule of Civil Procedure 23(a) must be satisfied and the proposed class action must be one of the types recognized by Rule 23(b). Baby Neal v. Casey, 43 F.3d 48, 55-56 (3d Cir.1994). The purpose of these requirements is to ensure that the action is pursued on behalf and in the interests of the actual class. Newton v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 259 F.3d 154, 182 n. 27 (3d Cir.2001).
The Rule 23(a) requirements are numerosity, commonality, typicality, and adequacy of representation. They require that plaintiffs demonstrate that: (1) the size of the class is so numerous that joinder of all members is impracticable; (2) there are questions of law and fact common to the class; (3) the claims or defenses of the plaintiffs are typical of the class; and, (4) the representatives will fairly and adequately protect the interests of the class. Fed. R. Crv. P. 23(a)(l)-(4).
Once they have satisfied the 23(a) prerequisites, the plaintiffs must show that the action is one of the types described in subsection (b) of Rule 23. Plaintiffs have moved for certification under (b)(2) or (b)(3). Rule 23(b)(2) certification is appropriate when the defendant “has acted or refused to act on grounds generally applicable to the class, thereby making appropriate final injunctive relief or corresponding declaratory relief with respect to the class as a whole.” Fed. R. Civ. P. 23(b)(2). For certification under (b)(3), common questions of law or fact must predominate over questions affecting only individual members so that a “class action is superior to other available methods for the fair and efficient adjudication of the controversy.” Fed. R. Civ. P. 23(b)(3).
The burden is on the plaintiffs to demonstrate that a class should be certified. See, e.g., Johnston v. HBO Film Mgmt., Inc., 265 F.3d 178, 183 (3d Cir.2001); Baby Neal, 43 at 55; Davis v. Romney, 490 F.2d 1360, 1366 (3d Cir.1974). As a general rule, the court does not explore the merits of the plaintiffs’ claims on a motion for class certification. See Eisen v. Carlisle & Jacquelin, 417 U.S. *89156, 177-78, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974). However, the Third Circuit has directed that “[a] class certification decision requires a thorough examination of the factual and legal allegations,” which necessitates “a preliminary inquiry into the merits ... to determine whether the alleged claims can be properly resolved as a class action.” Newton, 259 F.3d at 167-68; see also Falcon, 457 U.S. at 160, 102 S.Ct. 2364 (“[S]ometimes it may be necessary for the court to probe behind the pleadings before coming to rest on the certification question”).
IV. Rule 23(a) and Employment Discrimination
Unlike Title VII which permits a claim based on a theory of discriminatory impact, § 1981 provides a cause of action only for purposeful disparate treatment. Gen. Building Contractors Ass’n, Inc. v. Pennsylvania, 458 U.S. 375, 388-91, 102 S.Ct. 3141, 73 L.Ed.2d 835 (1982); Pryor v. Nat’l. Collegiate Athletic Ass’n, 288 F.3d 548, 562 (3d Cir.2002); Cooper v. Southern Co., 260 F.Supp.2d 1295, 1299 & n. 2 (N.D.Ga. 2003), aff'd 390 F.3d 695 (11th Cir.2004).37 To establish § 1981 liability for intentional racial discrimination, plaintiffs must show that: (1) they belong to a racial minority; (2) the discrimination involved the right to make and enforce contracts; and, (3) the defendants had an intent to discriminate on the basis of race. Brown v. Philip Morris Inc., 250 F.3d 789, 797 (3d Cir.2001). The core of a § 1981 action is intentional discrimination.
Prior to 1982, the broad allegation that an employer discriminated against its employees on the basis of a protected characteristic alone — such as race — was sufficient to meet the commonality criterion of Rule 23(a). Miller v. Hygrade Food Prods. Corp., 89 F.Supp.2d 643, 648 (E.D.Pa.2000) (Reed, J.). In these “across-the-board” suits, the representative plaintiff only had to allege that the defendant had a companywide policy of employment discrimination and that the employees shared a protected characteristic. Id. at 648-50 (tracing the history of across-the-board class actions). In essence, before 1982, once a class plaintiff alleged racial discrimination, commonality was presumed.
With its decision in General Telephone Co. of the Southwest v. Falcon, 457 U.S. 147, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982), the Supreme Court raised the bar for certifying employment discrimination class actions by articulating a more stringent commonality standard. Commonality would no longer be presumed. Id. at 160, 102 S.Ct. 2364. Instead, an employment discrimination class could be certified only if the plaintiff actually demonstrated that there was something about the specific pattern or policy of discrimination that affected all class members in a common way. Id.; Goodman v. Lukens Steel Co., 777 F.2d 113, 122 (3d Cir.1985). The Supreme Court stated that “[t]he mere fact that an aggrieved private plaintiff is a member of an identifiable class of persons of the same race or national origin is insufficient to establish his standing to litigate on their behalf all possible claims of discrimination against a common employer.” Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. 2364.
In finding that the plaintiffs had not demonstrated commonality and typicality, the Falcon Court observed that there is a “wide gap” between an employee’s claim of race discrimination and the existence of an entire class of persons who suffered the same discrimination at the hands of a companywide policy. Falcon, 457 U.S. at 157, 102 S.Ct. 2364. In order to “bridge the Falcon gap,” there must exist a class of persons affected in the same way by a company policy, pattern, or practice of discrimination. Reap, 199 F.R.D. at 544; see also Falcon, 457 U.S. *90at 158, 102 S.Ct. 2364. “Thus, it is the way in which the discrimination manifests itself that may present common questions of fact and law, not mere membership in an identifiable class of race or national origin.” Miller, 89 F.Supp.2d at 650 (emphasis added).
Analysis
Conceding numerosity, GPU contends that the plaintiffs cannot satisfy the commonality, typicality, and adequacy prerequisites of Rule 23(a). Nor can they demonstrate, the defendant argues, that the putative class action qualifies as either a Rule 23(b)(2) or 23(b)(3) one as contended by the plaintiffs.
The numerosity and commonality inquiries focus on the sufficiency and cohesiveness of the proposed class. The typicality and adequacy of representation prerequisites address the suitability of the named class representatives and their counsel. Barabin v. Aramark Corp., 210 F.R.D. 152, 157 (E.D.Pa. 2002) , aff'd without published opinion, No. 02-8057, 2003 WL 355417 (3d Cir. Jan. 24, 2003) . Failure of the plaintiffs to demonstrate any one of these elements will defeat certification.
A. Commonality and Typicality
Commonality requires that the named plaintiffs share a question of fact or law with the prospective class members. So long as “the named plaintiffs share at least one question of fact or law with the grievances of the prospective class,” the existence of individual facts and circumstances will not defeat commonality. Baby Neal, 43 F.3d at 56. However, after Falcon, if membership in a protected group is the only allegation forming the common nucleus of the class, the commonality requirement is not satisfied. Webb v. Merck & Co., 206 F.R.D. 399, 404 (E.D.Pa.2002) (Weiner, J.); Rowe v. Philadelphia Coca-Cola Bottling Co., No. 01-6965, 2003 WL 22594252, at *8 (E.D.Pa. Sept. 30, 2003) (Buckwalter, J.).
Factors to be analyzed in determining whether commonality exists in a discrimination case are: (1) the nature of the alleged discriminatory practice; (2) whether the practice’s effects are actually felt on a class-wide basis; (3) whether the challenged practice is uniform or diverse, considering such factors as the size of the company, the number of facilities or locations involved, extent of diversity of occupations and employment conditions, degree of geographic diversity and local autonomy; (4) the uniformity or diversity of the class members; (5) the nature of the company’s organizational structure as it relates to the centralized or decentralized supervision process of the challenged employment practice and policies; and, (6) the length of time covered by the plaintiffs’ claims and whether the challenged conditions prevailed during the entire period. See, e.g., Stastny v. Southern Bell Tel. & Tel. Co., 628 F.2d 267, 277 (4th Cir.1980); Elkins v. Am. Showa Inc., 219 F.R.D. 414, 418-19 (S.D.Ohio 2002); Bacon v. Honda of Am. Mfg., Inc., 205 F.R.D. 466, 477 (S.D.Ohio 2001), aff'd 370 F.3d 565 (6th Cir.2004); Harriss v. Pan Am. World Airways, Inc., 74 F.R.D. 24, 41 (N.D.Cal.1977).
1. Union Employees Cannot Be Included
After initially concentrating on exempt employees, the plaintiffs now insist that they are pursuing certification on behalf of both exempt and nonexempt employees.38 Yet, the record includes no documentation or analysis regarding GPU decision making affecting nonexempt employees.39 The focus of the plaintiffs’ evidence is exempt employees.
Four of the five remaining class representatives were nonexempt employees40 who were covered by different collective bargaining agreements. Plaintiffs concede that different unions governed different groups of employees at GPU, and the various union contracts are not part of the record.41 Significantly, the plaintiffs’ statistical analysis failed to control for the effects of the collective bargaining agreements on pay scales, promotions, seniority or other workplace *91matters generally affected by labor contracts.
Because the record is barren regarding the various labor contracts covering nonexempt employees, it is not possible to determine how these contracts impacted the decision-making process. Nor can a statistical analysis of compensation that includes both exempt and nonexempt employees without taking into consideration the implications of the collective bargaining agreements be reliable.
2. Nonexempt Nonunion Employees Cannot Be Included
There were a small number of GPU employees who were neither exempt nor covered by a collective bargaining agreement. No proposed class representative falls within this third employee category. Nor has any information about GPU decision making governing this type of employee been provided. Without that evidence, neither a class of nonexempt nonunion employees nor one including them can be certified.
3. Absence of Performance Evaluation Evidence
The plaintiffs did not offer any analysis of performance evaluations. If, as plaintiffs argue, performance evaluations affected each type of GPU compensation decision making, a statistical analysis of compensation or promotions would expose the “effects” of pervasive discrimination in conducting performance evaluations. Stated differently, the alleged failure of HR to separately analyze performance evaluation statistics was unnecessary because HR was analyzing compensation and promotion decisions, the product of the performance evaluations.
The only person alleging performance evaluation discrimination, Alexis, is no longer a class representative. Jones, in addition to promotion discrimination claims, also contends that his merit increases and bonus awards were not commensurate with his performance evaluations. He successfully had a vice president change his performance level to “meets expectations” from a below expectations performance review after he brought it to management’s attention.42 The following year, the same vice president also raised Jones’ incentive compensation award above the amount that Jones’ manager had recommended.43
k. Promotion
The plaintiffs have presented no statistical analysis of promotions at GPU during the relevant time period. Rather, they submitted Table 5, prepared by their expert Dr. Mark Killingsworth, which “summarizes the racial differences in exempt status.”44 This is not an analysis of promotion decision making. Studying the likelihood that a racial group will be employed in an exempt position is not probative of GPU’s actual promotion decisions from 1999 to 2001. Dr. Killings-worth’s Table 5 glosses over other reasons why proportionally more African-Americans may be employed as nonexempt employees rather than exempt — for example, locale, education or length of time at the company. While Dr. Killingsworth’s Table 5 shows that white employees are more likely to hold exempt positions (and presumably more likely to be paid more) than are African-American employees, it does not take into consideration any of the possible reasons for that difference, let alone demonstrate that the difference was likely caused by racially discriminatory promotion decision making.
The only statistical analysis of GPU’s actual promotion decisions was performed by defendant’s expert, Dr. Bernard R. Siskin. It posits that there was no discriminatory effect or intent in promotion decisions for the years 1997-2001. On the contrary, Dr. Siskin’s analysis of the statistics suggests that proportionately more African-Americans were promoted in the GPU workforce than would have been expected based on GPU’s overall African-American representation.45
*92 5. Centralized and Decentralized Decision making
That different employment decisions were made in different ways and by different persons on different levels threatens the cohesiveness of a proposed class. Plaintiffs admit that the level of involvement of Human Resources with respect to different types of decision making varied. Consequently, they would necessarily have difficulty proving that GPU’s employment decisions affected each potential class member in the same way.
The decentralized nature of the promotion decisions for exempt employees at GPU weighs against certification. Lott v. Westinghouse Savannah River Co., 200 F.R.D. 539, 554-56 (D.S.C.2000); see also Webb, 206 F.R.D. at 406; Wright v. Circuit City Stores, Inc., 201 F.R.D. 526, 542 (N.D.Ala.2001). If claims are based upon individual decisions made by local managers who had varying degrees of autonomy over compensation and promotion, a proposed disparate treatment class must fail. Vinson v. Seven Seventeen HB Philadelphia Corp., No. 00-6334, 2001 WL 1774073, at *19 (E.D.Pa. Oct. 31, 2001) (collecting cases where courts decline to certify classes because decision making was decentralized); Sperling v. Hoffmann-La Roche, Inc., 924 F.Supp. 1346, 1363 (D.N.J. 1996) (Ackerman, J.) (“[A] decision by a company to give managers the discretion to make employment decisions, and the subsequent exercise of that discretion by some managers in a discriminatory manner, is not tantamount to a decision by a company to pursue a systemic, companywide policy of intentional discrimination.... ”).
If certification of a class alleging only decentralized decision making is difficult, it is even more difficult where the proposed class is comprised of employees who were subject to both decentralized and centralized decision making. Plaintiffs who allege centralization as to one employment practice and decentralization as to another cannot show, as required by Falcon, that the class was affected by the challenged practices in the same way. Without such a showing, the requirement that class certification be based on more than mere membership in a protected class is not met.
6. Multi-Facility Class — Geographic Diversity and Diversity of Employment Conditions
The complex and multi-facility nature of defendant’s operations militates against certification. There are approximately fifty GPU locations throughout Pennsylvania potentially involved in this litigation.46 In disparate treatment cases involving multiple departments or facilities, commonality and typicality are difficult to show. Zachery v. Texaco Exploration and Production, Inc., 185 F.R.D. 230, 239 (W.D.Tex.1999).
Geographically widespread facilities make proving a pattern and practice of disparate treatment difficult. Zachery, 185 F.R.D. at 239. Employees who were from different departments, were supervised by different people, worked different shifts, and were at different levels within the company hierarchy have grievances that are not susceptible to generalized proof or defenses. Lumpkin v. E.I. DuPont de Nemours & Co., 161 F.R.D. 480, 482 (M.D.Ga.1995); see also Reyes v. Walt Disney World Co., 176 F.R.D. 654, 658 (M.D.Fla.1998). A plaintiff may represent a multi-facility class “only where centralized and uniform employment practices affect all facilities in the same way.” Reid v. Lockheed Martin Aeronautics Co., 205 F.R.D. 655, 667-68 (S.D.Ga.2001) (emphasis added) (collecting cases); see also Webb, 206 F.R.D. at 404.
7. Lack of Monitoring Does Not Demonstrate Intentional Discrimination
The plaintiffs also argue that GPU did not monitor promotion decisions for discriminatory impact. This, however, is a disparate treatment case, not a disparate impact one. There is no allegation nor even a suggestion that GPU deliberately implemented a non-monitoring policy with the intent to discriminate. Even assuming the allegations are true, “lack of monitoring” of employment decisions does not evidence discriminatory intent. Disparate treatment “implies more than intent as volition or intent as awareness *93of consequences. It implies that a decision-maker singled out a particular group for disparate treatment and selected his course of action at least in part for the purpose of causing its adverse effects on the identifiable group. A showing that the defendants were negligent will not suffice.” Nabozny v. Pod-lesny, 92 F.3d 446, 454 (7th Cir.1996) (citations and internal quotations omitted) (equal protection case).
Thus far, both nonexempt and nonexempt nonunion employees have been excluded from any proposed class, leaving only exempt employees as potential class members. Challenges to performance evaluation policies are not suitable for inclusion in a disparate treatment class under the circumstances presented in this case. Furthermore, the plaintiffs have failed to make any showing that GPU engaged in intentional discrimination in awarding promotions. Now, the analysis will focus on the claim of compensation discrimination with respect to exempt employees.
8. Plaintiffs’ Expert’s Statistical Analysis of Exempt Employee Compensation is Inadequate47
The remaining claims for potential inclusion in a class are those brought by exempt employees alleging compensation discrimination. In support of their motion, plaintiffs submitted statistical analyses performed by their expert, Dr. Killingsworth.48 Although the court does not normally look into the merits of the ease at the certification stage, the validity of the plaintiffs’ proffered statistical analysis is a proper inquiry at this point. Hopewell v. Univ. of Pittsburgh, 79 F.R.D. 689, 693-95 (W.D.Pa.1978). Because the value and usefulness of any statistical study is dependent on the validity of the method used, the analysis begins with the reliability of the expert’s methodology. Cooper v. Southern Co., 390 F.3d 695, 716-19 (11th Cir.2004). Then, the question becomes whether the statistical disparities raise a presumption of purposeful discrimination by showing that the named plaintiffs have claims in common with other class members. Id. at 716-17.
a. Omission of Variables
Statistical results are as meaningful as the factors the statistician chooses to control for or omit. Dr. Killingsworth omitted crucial variables from his compensation analysis, such as education49 and work experience, *94which affected the results of his study.50 Sheehan v. Purolator, Inc., 839 F.2d 99, 103 (2d Cir.1988). The omission of these variables means that his statistical report merely demonstrates African-American underrepresentation in the workforce. “In light of Falcon, general statistical evidence of underrepresentation in the workforce will undoubtedly not suffice to justify a single class covering different types of discrimination____” Griffin v. Dugger, 823 F.2d 1476, 1490 n. 32 (11th Cir.1987) (internal quotations and citations omitted).
The first step in performing a reliable multiple regression analysis “is to specify all of the ‘legitimate’ (i.e., nondiscriminatory) factors that are likely to significantly affect the dependent variable and which could account for disparities” in the treatment of white and African-American employees. Ot-taviani v. State Univ. of N.Y. at New Paltz, 875 F.2d 365, 367 (2d Cir.1989). By accounting for the effects of such variables, plaintiffs could have confidently pointed to unexplained remaining statistical disparities as necessarily products of discrimination. Id. Stated differently, without factoring relevant variables into the analysis, the statistician leaves open the possibility that those variables, and not racial discrimination, produced differences between African-American and white employees.
Lumping all exempt employees, Dr. Kill-ingsworth ignored critical differences in employee status, experience, job requirements, and education, thus reducing the reliability of his statistical study. By failing to rule out or even consider other potentially non-discriminatory explanations for African-American underrepresentation in the GPU exempt workforce, the statistics cannot explain that raeial discrimination is the standard operating procedure of the company — the essential requirement for a disparate treatment case. Pointing to companywide disparities without factoring into the equation different geographic areas and different jobs and different employment demographics does not necessarily lead to the conclusion that the company is responsible for those disparities and that racial discrimination was the company’s standard operating procedure. Factors outside the employer’s control, such as education, experience, and the relevant labor pools in a geographic area, could also explain why fewer African-Americans are represented in exempt positions. Consequently, failing to control for these other variables skews the results. Lott v. Westinghouse Savannah River Co., 200 F.R.D. 539, 560-61 (D.S.C. 2000); see also Tagatz v. Marquette Univ., 861 F.2d 1040, 1045 (7th Cir.1988) (finding that the expert’s failure to control for other variables made the proffered statistical proof “essentially worthless”); Webb, 206 F.R.D. at 408 n. 2 (by failing to control for job grade, the plaintiffs presented statistics that were “not instructive”).
Dr. Killingsworth criticizes Dr. Siskin for including education background in his analysis although GPU’s files did not contain information on the educational backgrounds of about half of the entire workforce. Yet, educational data was available for about 75 percent of GPU exempt employees — the only employees now under consideration.51 When Dr. Siskin used Dr. Killingsworth’s Table 3, which only showed statistically significant disparities in 1999, and factored in available educational data, the analysis revealed no significant statistical disparities in compensation for any year.52
*95b. 1.96 Standard Deviation Units
In order for statistics to rise to the level of statistical significance in a disparate impact employment discrimination case, the disparity must be equal to or greater than 1.96 standard deviation units.53 See Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977) (“Where gross statistical disparities can be shown, they alone may in a proper case constitute prima facie proof of a pattern or practice of discrimination.”) (quoting Int’l Bhd. of Teamsters v. United States, 431 U.S. 324, 339, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977)). No court has applied this 1.96 standard deviation level to a disparate treatment case.
Section 1981 claims require proof of discriminatory intent. Gen. Building Contractors Ass’n, Inc., 458 U.S. at 388-91, 102 S.Ct. 3141. Consequently, the fact that this proposed class is proceeding only on a disparate treatment theory, “while not dispositive, weighs against finding the commonality and typicality required by Rule 23.” Washington, 959 F.2d at 1570 n. 10; see also Stastny v. Southern Bell Tel. & Tel., 628 F.2d 267, 274 n. 10 (4th Cir.1980) (“As is now well recognized, the class action commonality criteria are, in general, more easily met when a disparate impact rather than a disparate treatment theory underlies a class claim.”). The defendant argues that the statistical significance standard for a disparate treatment case must be greater than 1.96 standard deviation units. Essentially, defendants argue that more than just “statistically significant” evidence must be submitted in intentional discrimination cases because disparate treatment cases require proof of discriminatory intent and cannot be shown by evidence of discriminatory impact alone.
The plaintiffs, on the other hand, offer a novel process for evaluating commonality evidence in a disparate treatment case. They contend that the first step is to decide whether the statistical evidence alone supports a certification of a hypothetical disparate impact case. If so, then additional evidence, such as affidavits submitted by proposed class representatives, should be considered to determine whether certification of this disparate treatment class is warranted.54
Because any class including nonexempt employees and promotion decisions cannot be certified, only the statistics offered in support of certifying a class of exempt employees based on compensation shall be considered. Even if the 1.96 standard for certification were applied in this disparate treatment case, the plaintiffs’ statistics are, nonetheless, insufficient.
Dr. Killingsworth’s Table 3, addressing compensation for exempt employees, barely meets the minimum standard of statistical significance in the difference between African-American and white employee compensation for the first of three years55 (2.03 in 1999, 1.49 in 2000, and 0.99 in 2001), and only for the percentage model, which purports to *96show the difference in pay between white and African-American employees as a percentage of compensation, as opposed to the difference in dollar value, that is, the difference in dollar amounts paid to African-American and white employees.56 Thus, Table 3, which presents a summary of regression analyses of black-white biweekly salary differences expressed in percentages for exempt employees reveals statistically significant disparities for 1999 only, not 2000 or 2001, and only using the percentage comparison. The other tables are not useful because they represent aggregate statistics, including all types of employees, union and nonunion, with varying education and experience levels.57
9. GPU Decision Making Is Not Entirely Subjective
Plaintiffs characterize their proposed class as a “pay” or “compensation” discrimination one. Actually, they lump together many types of discrete employment actions: hiring, setting pay rates, bonuses, promotion, and performance evaluations,58 as “compensation discrimination.” In doing so, they are asserting across-the-board claims. See Webb, 206 F.R.D. at 404 n. 1. Accordingly, to overcome their failure to satisfy the Rule 23 commonality requirement, the plaintiffs must meet Falcon’s “entirely subjective” standard.
In the absence of challenges to a specific employment practice, the plaintiffs must show that each of the company’s employment policies and practices were infused with discriminatory intent. Garcia v. Veneman, 224 F.R.D. 8, 11-13 (D.D.C.2004) (noting that Falcon’s entirely subjective standard requires “both ‘significant proof that the defendant ‘operated under a general policy of discrimination’ and ‘the discrimination manifested itself ... in the same general fashion, such as through entirely subjective decision-making processes.’”). Plaintiffs can show this by demonstrating that the “primary practices used to discriminate in the different [employee] categories are themselves similar.” Hartman v. Duffey, 19 F.3d 1459, 1471 (D.C.Cir.1994). If plaintiffs can meet this burden, the commonality requirement is relaxed and they may proceed with their broad class action. Bacon, 205 F.R.D. at 477 (noting that if entire subjectivity is met, plaintiffs would be excused from demonstrating true commonality).
The entirely subjective standard is a difficult one because the presence of some objective criteria in the process precludes a finding of “entirely” subjective. Bacon, 370 F.3d at 571-72 (noting that defendant used some objective criteria in its decision making, such as time in service, attendance, and test scores, and declining certification because decision making was not “exclusively” subjective); Griffin v. Dugger, 823 F.2d 1476, 1490 (11th Cir.1987); Appleton v. Deloitte & *97Touche L.L.P., 168 F.R.D. 221, 228-29 (M.D.Tenn.1996). Consequently, the presence of objective criteria or oversight precludes an “entirely subjective” method of class certification.
The plaintiffs have not met the Falcon “entirely subjective” threshold. Their own evidence demonstrates that GPU’s promotion and compensation decisions employ some objective criteria. While GPU’s decision-making processes incorporated some subjective elements, GPU also utilized objective standards. For competitive promotion decisions, managers compared the candidates’ qualifications with the written job requirements. HR distributed written selection guidelines and recommended questions to use in interviewing candidates. HR required managers to provide a written explanation of why certain candidates were not selected for interviews, and HR reviewed the final hiring decision to ensure that the selected candidate was qualified. Many interviews were conducted by more than one interviewer.
Performance evaluations, which factored into promotion decisions, utilized a combination of subjective and objective criteria, with the manager’s decision requiring approval of the manager’s supervisor. HR monitored its promotion policies by performing statistical analyses. This promotion structure, containing a mix of objective and subjective criteria, is not free from oversight and is not entirely subjective.
Plaintiffs acknowledge that compensation decision making was centralized. Merit increases and bonuses were fixed by formal guidelines, and starting salaries were governed by factors including the new employee’s education and experience as compared with salaries of employees already performing in that position. Hence, any subjective considerations that may have come into play were constrained by objective standards.
10. Typicality
Although they are distinct requirements under Rule 23(a), “[t]he concepts of commonality and typicality are broadly defined and tend to merge.” Baby Neal, 43 F.3d at 56. Both focus on whether a class can be practically, efficiently and fairly maintained. Barabin, 210 F.R.D. at 159.
Typicality requires a strong similarity of legal theories to ensure that the class representatives’ pursuit of their own goals will work to benefit the entire class. Barabin, 210 F.R.D. at 158-59. While typicality is not synonymous with identicality, all of the claims must arise from the same policy, practice, or course of conduct. Id. at 159; Eisen-berg v. Gagnon, 766 F.2d 770, 786 (3d Cir. 1985). If plaintiffs suffer different types of specific injuries but all of the injuries arise from the same practice, plaintiffs suffering one specific injury can represent a class suffering other injuries. Barabin, 210 F.R.D. at 159. Plaintiffs here have not demonstrated that their diverse injuries all result from the same common practice.
The differences among the named plaintiffs here demonstrate the lack of both commonality and typicality. Each works in a different job and at a different location. Each complains of a different employment practice. Some are union members covered by collective bargaining agreements and others are not.59
Only Montgomery, who was challenging disciplinary and termination decisions, worked at Penn Electric. Plaintiffs are no longer seeking to include termination and disciplinary decisions in this class.60 Hence, with Montgomery’s claims no longer in the ease, there is no Penn Electric employee as a class representative.
There are a small number of employees who are neither covered by a collective bar*98gaining agreement nor exempt. No class representative shares that status.
Where the defendant can raise unique defenses to each plaintiffs claim, typicality may not exist if the defenses could threaten to become the focus of the litigation. In re Safeguard Scientifics, 216 F.R.D. 577, 581-82 (E.D.Pa.2003) (Joyner, J.) (citing Zenith Labs., Inc. v. Carter-Wallace, Inc., 530 F.2d 508, 512 (3d Cir.1976)). In this action, the defendant’s response to each challenged job action is specific, requiring an evaluation of factors peculiar to that employee, that person’s job position and the decision-making process involved. For example, GPU argues that Greene was not promoted because he did not have the required seniority, a factor that, in his case, is subject to the union contract.61
Only one employee, Tribue, alleges promotion discrimination during the liability period.62 Three times he applied for and did not receive a promotion. He contends that he was not given the first promotion because the friend of a supervisor got it. The second time, the person promoted was a key manager’s boyfriend. The last time, a person with an advanced degree was promoted so GPU could keep that employee. By his own characterizations, the motivating factor for promoting someone else in each instance was not race. Thus, Tribue faces serious hurdles in demonstrating intent because his own testimony offers nondiscriminatory reasons for the adverse job actions.
Although originally the plaintiffs were pursuing discriminatory discipline and termination claims, they have since withdrawn them. Consequently, Montgomery does not have any complaint in common with the proposed class and is atypical of the putative class members.
Defendant points to these situations as examples of specific unique defenses that are not typical to each putative class member. The presence of these unique defenses weighs against certification.
B. Adequacy of Representation
The adequacy requirement has two parts: one focuses on the attorneys and the other on the plaintiffs themselves. A court must evaluate whether plaintiffs’ counsel are qualified and whether the named class representatives would adequately protect the interests of unnamed class members.
Counsel’s ability and incentive to represent the class vigorously must be evaluated to assure that plaintiffs’ counsel are well-qualified to conduct the instant litigation. In re Cendant Corp. Litig., 264 F.3d 201, 265 (3d Cir.2001). The burden is on the defendant to demonstrate plaintiffs’ counsel’s lack of qualifications. Thomas v. SmithKline Beecham Corp., 201 F.R.D. 386, 396 (E.D.Pa.2001) (Yohn, J.).
Defendant raises several challenges to the ability of plaintiffs’ counsel to provide competent representation in this action. It notes that plaintiffs’ complaint contains several material misstatements of fact about the allegations of the named plaintiffs. Defendant also points to problems during discovery as evidence of the inadequacy of plaintiffs’ counsel. Despite these issues, plaintiffs’ counsel is qualified to serve as class counsel, having done so in other cases in this district. See Ketchum v. Sunoco, Inc., 217 F.R.D. 354 (E.D.Pa.2003) (Green, J.). In any event, in light of the certification decision, the point is moot.
There must be no conflict of interest between the claims of the class representatives and the other members of the proposed class. In evaluating whether conflicts of interest among class members prevents certification, “[t]he adequaey-of-representation requirement ‘tends to merge’ with the commonality and typicality requirements of Rule 23(a).” Amchem Prods., Inc. v. Windsor, 521 U.S. 591, 626 n. 20, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997); see also Newton, 259 F.3d at 186 (3d Cir.2001). Therefore, given the lack commonality and typicality, the named plaintiffs could not adequately represent the interests of the proposed class.
*99Conclusion
Calling this a “compensation” discrimination case, the plaintiffs have lumped together all employment actions for union and nonunion employees with differing education and experiential backgrounds, and from different facilities in different locales. They have failed to submit evidence on GPU promotion and compensation policies for nonexempt employees; failed to account for the implications of collective bargaining agreements; and, failed to challenge specific employment practices, instead raising broad allegations of discriminatory conduct. They cannot demonstrate that GPU decision making was entirely subjective. In short, the plaintiffs have been unable to present any evidence that GPU had a purposeful companywide policy of racial discrimination that would merit certification of this proposed disparate treatment class.
ORDER
AND NOW, this 1st day of September, 2005, upon consideration of the Plaintiffs’ Motion for Partial Class Certification (Document No. 47), the defendant’s opposition, the plaintiffs’ reply, and the defendant’s sur-re-ply, and after a hearing and oral argument, it is ORDERED that the motion is DENIED.
. Joint Stip. Uncontested Facts 1 6.
. Id. 116-7.
. Pls.' Mem. of Law in Supp. of Pis.’ Mot. for Partial Class Certification at 5 (“Pis.’ Mem.”).
. Def’s Mem. in Opp. to Pis.’ Mot. for Partial Class Certification at 11 (“Def.’s Mem.").
. Id.
. Pls.' Mem. at 5.
. Joint Stip. Uncontested Facts 128.
. Compl. 139(b); Pis.’Mem. at 16.
. Joint Stip. Uncontested Facts 1116-17.
. Compl. ¶141b; Pis.' Mem. at 15.
. Joint Stip. Uncontested Facts V 3.
. Compl. 1143a; Pis.'Mem. at 15-16.
. Joint Stip. Uncontested Facts 111120, 22.
. Joint Stip. Uncontested Facts 1118.
. Pis.’ Mem. at 15.
. Joint Stip. Uncontested Facts H 23.
. Joint Stip. Uncontested Facts H 25.
. Pis.'Mem. at 16.
. Joint Stip. Uncontested Facts 111124, 26-27.
. Pis. ’ Reply Mem. of Law and App. in Supp. of Pis.' Mot. for Partial Class Certification at 3-4, 6 {‘‘Pis.’ Reply").
. Pis.’Mem. at 1; Pis.’Reply at 6.
. Pis.’ Mem. at 2.
. Pis.’Mem. at 1-2, 18-20.
. Pis.’Mem. at 5, 9.
. Pis.'Mem. at 11-12.
. Pis.' Mem. at 7-9.
. Compl. H 24.
. Pis.’Mem. at 6.
. Pis.’Mem. at 1.
. Pis. ’ Mem. at 29 (emphasis added).
. Pis.' Reply at 3-4.
. Transcript of Oral Argument (Apr. 21, 2004), at 124 ("Tr.”).
. The plaintiffs’ reply brief suggests that if the court declines to certify this class, it should certify two subclasses: "1. Black exempt employees ...; and 2. Black non-exempt employees ... during the class period.” Pis.' Reply at 4.
. Pls.’ Reply at 6.
. Id.
. See Order, Civ. A. No. 01-4950 (July 23, 2002) (Document No. 12).
. In their opening memorandum, plaintiffs sought to proceed under both disparate impact and disparate treatment theories of class liability. Pis.’ Mem. at 4. In their reply, the plaintiffs clarified that they sought certification under only a disparate treatment theory, recognizing that § 1981 does not permit recovery for disparate impact. Pis.’ Reply at 4. They stated that they wanted to introduce their disparate impact evidence, Dr. Mark Killingsworth's statistical report, in support of their disparate treatment claim, arguing that "because Plaintiff’s statistical evidence of adverse impact alone would be sufficient for certification under the disparate impact theory, the Court should consider that evidence along with deposition testimony and documentary evidence to find that Plaintiffs have established a basis for certification under the disparate treatment theory.” Pis.' Reply at 5.
. Pls.' Reply at 3-4.
. Tr. at 148-49.
. Les Jones is the only proposed class representative who is an exempt employee.
. Tr. at 120.
. Martorelli Dec. 112 (Def. 's App. Tab C); Def. ‘s Mem. at 10.
. Id. 113.
. Killingsworth First Rep. at 4.
. Siskin Comment at 13.
. Def.’s Mem. at 11; Miller Dec. 113.
. Because the record lacks evidence of GPU's nonexempt promotion and compensation practices, the statistical tables relating only to exempt employees shall be considered.
Table 1 compares average compensation of white employees to African-American employees, overall and within exempt and nonexempt positions. Table 2 presents a regression analysis of racial differences in biweeldy pay (expressed in both dollar value and percentages). Table 2 contains two "models”: Model 1 takes account of initial and most recent hire date, age, and gender; Model 2 takes account of the same factors as Model 1, while adding employee status. Tables 3 and 4 present additional regression analysis with respect to exempt employees and nonexempt employees. Table 5 summarizes the likelihood that white and African-American employees will hold exempt status. Table 6 presents the same summary, controlling for age, gender, and years in service.
. Dr. Killingsworth submitted an analysis in support of plaintiffs' motion for certification. The defendant’s expert, Dr. Siskin, commented on this initial report and offered his own analysis in support of GPU’s argument against certification, focusing on Dr. Killingsworth's omission of measurable factors, such as education and time-in-grade, and the absence of promotion statistics. Dr. Killingsworth responded to Dr. Siskin's criticisms in a second report. Dr. Siskin again commented on Dr. Killingsworth's second report.
Because the second report differs only slightly from the first and attempts to correct problems with the first report, I use the second report for this analysis. The second report states that Dr. Killingsworth was able to obtain more data on employee terminations, making it possible to identify a small number of additional persons terminated from 1996-2001. He also initially mis-coded a small number of exempt GPU officers as nonexempt; he corrected this mistake in the second report. Finally, Dr. Killingsworth responded to criticisms leveled by Dr. Siskin, explaining that including educational data for only those employees for which such data was available did not in fact create a more accurate picture of GPU's employment practices. The experts take issue with many of the decisions their counterpart made in analyzing various pieces of evidence. The Court's other rulings excluding nonexempt employees and promotion decisions render many of these disputes irrelevant.
. In his Second Report, Dr. Killingsworth did include a sub-sample with education data, but objected to the inclusion of this data because it *94was not available for all of the employees. He believed that simply deleting those employees for whom GPU did not have education information skewed the results. Killingsworth Second Rep. at 4, 7. When Dr. Külingsworth did include the education information for exempt employees, the only ones still remaining in our certification analysis, the differences in African-American and white compensation were not statistically significant. Killingsworth Second Rep. at 22 (Table 8).
. Plaintiffs complain that GPU's yearly analysis of diversity similarly did not control for education and job grade. Pis.’ Mem. at 11. Yet, their own expert’s study did not do so.
. Siskin Second Comment at 1; Killingsworth Second Rep. at 6 (noting that education data was available for 876 of 1184 exempt employees).
. Siskin Comment at 8. Dr. Killingsworth complains that Dr. Siskin simply deleted the 25% of exempt employees for which he did not have education data, thereby reducing the probative value of Dr. Siskin's data because deleting people necessarily leads to a reduced statistical signifi-*95canee of the differences in pay. Killingsworth Second Rep. at 4-7.
As noted above, even if I accepted Dr. Killings-worth's Table 3 (evaluating biweekly salary differences) and were not troubled by the failure to control for nondiscriminatory variables that could explain racial disparities, Table 3 only demonstrates disparities that rise to the level of statistical significance in one relevant year, 1999.
. Statistical significance measure the “probability that a disparity is simply due to chance, rather than any other identifiable factor.” Ottaviani v. State Univ. of N.Y. at New Paltz, 875 F.2d 365, 371 (2d Cir.1989). Standard deviation units measure statistical significance. 1.96 standard deviation units refers to the level of statistical disparity required to demonstrate legal statistical significance using a two-tailed test. See Palmer v. Shultz, 815 F.2d 84, 92-93 (D.C.Cir.1987); see also Ottaviani, 875 F.2d at 371-72 (“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.’ ... The existence of a [five percent] level of statistical significance indicates that it is fairly unlikely that an observed disparity is due to chance, and it can provide indirect support for the proposition that disparate results are intentional rather than random.”).
. Pls.’ Reply at 5.
. The plaintiffs submitted statistics spanning 1996-2001. The years relevant to our analysis are 1999-2001, as discussed above, because a two-year statute of limitations applies.
. Killingsworth Second Rep., Table 3 at 16.
. Table 1 presents a simple comparison of average biweekly salaries of African-American and white employees. It does not control for any variables. Dr. Killingsworth states that "Table 1 shows that average compensation of white employees was always substantially higher than that of black employees.” This statement, however, does not aid plaintiffs’ case. Table 1 does not take into account education, age, experience, length of GPU employment, or any number of possible non-discriminatory reasons that overall compensation of African-Americans is less than that of whites.
Similarly, Table 2 presents a regression analysis of racial differences in biweekly pay. It has two models. Model 1 takes account of most recent hire date, gender, and employee age. Model 2 also factors into the analysis the employee status (exempt v. union). Again, experience, length of employment, and education are not considered. These "summary” tables represent aggregate "lumping together" of employees and do not shed light on whether GPU intentionally discriminated against African-American employees.
I also note that the more "egregious” statistical disparities tend to appear in tables comparing African-American and white non exempt employees. See Killingsworth Second Rep. Tables 4, 9. Because there is a dearth of evidence submitted on GPU’s nonexempt employment practices, I decline to certify a class containing nonexempt and exempt employees or a class of only nonexempt employees. I also note that nonexempt employees were covered by various collective bargaining agreements. Because GPU has less control over the pay and advancement of union members than exempt employees, these Tables suggest that any racial disparities are not the fault of GPU.
. Initially, they also sought to include transfers, discipline, and termination.
. Plaintiff Melissa Alexis was employed by GPU as a Cultural Transformation Consultant from June 1999 until May 2000. She contests a performance evaluation she received in January 2000 for allegedly not supporting the work of other consultants. Pis.’Mem. at 15. Alexis originally alleged that she was unfairly given a "below expectations" performance evaluation. However, after discovery revealed that she was in fact given a "meets expectations” rating, plaintiffs withdrew her as a class representative. Joint Stip. Uncontested Facts 114.
. Compl. ¶43a; Def.'s Sur-Reply Mem. in Opp. to Pls.’ Mot. for Partial Class Certification at 18 ("Def.’s Sur-Reply").
. Def.'s Opp. at 8; Def.’s Sur-Reply at 18-19.
. While Jones includes allegations of promotion discrimination, his factual assertions are all prior to 1999 and therefore before the class period.