BEAM, Circuit Judge.
Leslie Morgan; Kenneth Stacker, and Theodore Boldin, black managers at United Parcel Service (UPS), brought a civil rights action alleging individual and class claims of employment discrimination under Title VII and 42 U.S.C. § 1981. These individuals moved for class certification and the district court partially granted the motion, certifying four classes of employees (collectively, “Plaintiffs”). UPS moved for summary judgment on both the Title VII and the 42 U.S.C. § 1981 class claims or to decertify the class. Plaintiffs moved for partial summary judgment on their class claims of racially discriminatory pay. The district court denied Plaintiffs’ motion, granted summary judgment to UPS, and denied UPS’s motion to decertify the class. It held that Plaintiffs had failed to adduce sufficient evidence of a pattern or practice of class-wide discrimination. We affirm.
I. FACTUAL BACKGROUND
UPS delivers packages world-wide for individuals and businesses. UPS is divided into eleven geographic regions in the United States, each with a regional manager. A region is subdivided into approximately six districts, each with a district manager. The district managers are responsible for nearly every aspect of district operations, including picking up, sorting, and delivering packages, as well as employee development, promotion, and compensation. There are between sixty and seventy districts in the United States and each one varies in geographic size, population density, package volume, and labor climate. A district is further divided into divisions, each with a division manager. Divisions are organized along functional lines with some related to package operations and some related to supporting staff.
Two levels of managerial employees exist below the division level: supervisors and center managers. The entry level managerial position' is a supervisor who reports to the center manager. The center manager reports to the division manager, the division manager to the district manager, and the district manager to the regional manager.
UPS vests decisionmaking authority in district managers to promote employees to division manager, center manager, and supervisor. District managers hold “People’s Meetings” approximately twice a year, where they discuss the upward mobility of management personnel. At these meetings, information is presented on supervisors and center managers in the district, and their performance and readiness for promotion are assessed. At some of the meetings, a color photograph of the individual is displayed. The meeting produces a list of the employees who the district managers believe are ready for promotion to managerial positions. Open positions within the company are not posted. The district manager may consult the list compiled at the meetings to determine who will receive a promotion.
II. PROCEDURAL BACKGROUND
Plaintiffs are all UPS center managers and they allege that UPS racially discriminated against them in terms of upward mobility, working conditions, and pay. With regard to the upward-mobility claim, Plaintiffs argue that the subjective selection process for managerial positions at UPS limits the promotion of blacks, which, in turn, inhibits their overall upward mobility and causes them to peak in then-careers at or below the center-manager level.
[462]*462The district court bifurcated the trial into a liability/injunctive phase, followed by a remedial/damages phase. Four classes were certified under Federal Rule of Civil Procedure 23(b)(2) for the first phase. The court said it would consider certifying the damages phase as a Rule 23(b)(3) class action if liability was established.
The Title VII denial-of-upward-mobility class was defined as:
[Ajll black salaried full-time employees of UPS nationwide employed as center managers in Operations (Package, Hub, Feeder, Air) or Human Resources at any time between December 20, 1991 and the date of judgment, and who woi'ked as a supervisory or managerial employee of UPS for at least five years without being promoted above the center manager level.
The Title VII unequal-working-conditions and unequal-pay class was defined as:
[A]ll black salaried full-time employees of UPS nationwide employed as center managers in Operations (Package, Hub, Feeder, Air) or Human Resources at any time between December 20, 1991 and the date of judgment.
The 42 U.S.C. § 1981 denial-of-upward-mobility class was defined as:
[A]ll black salaried full-time employees of UPS nationwide employed as center managers in Operations (Package, Hub, Feeder, Air) or Human Resources at any time on or after June 17, 1989 and the date of judgment, and who worked as a supervisory or managerial employee of UPS for at least eight years without being promoted above the center manager level.
The 42 U.S.C. § 1981 unequal-working-conditions and unequal-pay class was defined as:
[A]ll black salaried full-time employees of UPS nationwide employed as center managers in Operations (Package, Hub, Feeder, Air) or Human Resources at any time on or after November 21, 1991 and the date of the judgment.
The district court granted a motion by the EEOC to intervene, but the EEOC is not involved in these appeals. At the time of the original class certification, Charles Cartwright was a member of at least two certified classes. After modification of the certified classes, the district court ruled that he no longer qualified as a member of any of the four classes and thus vacated the order allowing him to intervene. Frank Jackson was also denied intervenor status. Both Cartwright and Jackson appeal those decisions. Brian Needham, Francis Truitt, and Bedell Finley also sought, but were denied, intervenor status; however, they are not parties to this appeal.1
Discovery proceedings produced reports, statistical analyses, and models of UPS’s employment data and practices prepared by experts for all parties. Plaintiffs presented two experts who conducted statistical analyses: Dr. Weiner and Dr. Staple-ton. And the Defendant offered statistical analyses conducted by its expert, Dr. Evans. At the close of discovery, UPS moved for summary judgment on all class claims or to decertify the classes. Plaintiffs moved for partial summary judgment on their class claim of discriminatory pay. UPS also moved to bar the testimony of two of Plaintiffs’ experts as inadmissible under Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharm., Inc., 509 [463]*463U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469(1993), because of faulty methodology and data. The district court granted UPS summary judgment on all class claims and denied Plaintiffs’ summary judgment motion. The court also denied as moot UPS’s motions to decertify the class and to bar Plaintiffs’ expert testimony.
III. DISCUSSION
Plaintiffs claim the district court erred in granting UPS’s summary judgment motion on each of the three class claims: (1) denial of overall upward mobility, (2) working conditions, and (3) pay.
We review de novo a grant of summary judgment to determine whether a claim is factually supported. Jackson v. Ark. Dep’t of Educ., 272 F.3d 1020, 1025 (8th Cir.2001). Summary judgment “shall be rendered forthwith if the pleadings, depositions, answers to interrogatories, and admissions on file, together with the affidavits, if any, show that there is no genuine issue as to any material fact and that the moving party is entitled to a judgment as a matter of law.” Fed.R.Civ.P. 56(c). We must view all facts in the light most favorable to the nonmoving party and give the nonmoving party the benefit of all reasonable inferences that can be drawn from those facts. Widoe v. Dist. #111 Otoe County Sck, 147 F.3d 726, 728 (8th Cir.1998). However, “[t]he mere existence of a scintilla of evidence in support of the plaintiffs position will be insufficient; there must be evidence on which the jury could reasonably find for the plaintiff.” Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 252, 106 S.Ct. 2505, 91 L.Ed.2d 202 (1986). Therefore, in a motion for summary judgment, “[t]he judge’s inquiry unavoidably asks whether reasonable jurors could find by a preponderance of the evidence that the plaintiff is entitled to a verdict.” Id.
Plaintiffs’ case centers on discriminatory treatment in the form of unequal working conditions, promotions, and pay. More specifically, this class action alleges a pattern or practice of discrimination with regard to each of these three aspects of employment. In a pattern-or-practice class action, the class must prove that the defendant “ ‘regularly and purposefully’ treated members of the protected group less favorably and that unlawful discrimination was the employer’s ‘regular procedure or policy.’ ” EEOC v. McDonnell Douglas Corp., 191 F.3d 948, 951 (8th Cir.1999) (quoting Int’l Bhd. of Teamsters v. United States, 431 U.S. 324, 335 & 360, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977)). Stated differently, the class must prove that the “ ‘discrimination was the company’s standard operating procedure-the regular rather than the unusual practice.” ’ Craik v. Minn. State Univ. Bd., 731 F.2d 465, 470 (8th Cir.1984) (quoting Teamsters, 431 U.S. at 336, 97 S.Ct. 1843). Upon an employer’s summary judgment motion, the plaintiff class must adduce some evidence from which this conclusion can be drawn. Proving discrimination that is “ ‘isolated or sporadic’ ” is insufficient. McDonnell Douglas, 191 F.3d at 951 (quoting Cooper v. Federal Reserve Bank, 467 U.S. 867, 876, 104 S.Ct. 2794, 81 L.Ed.2d 718 (1984)).
Typically in a case like this one, plaintiffs will offer statistical evidence of disparities between protected and unprotected employees who are otherwise similarly situated. Craik, 731 F.2d at 470. In defense, employers attempt to show that the plaintiffs’ “ ‘proof is either inaccurate or insignificant.’ ” Id. (quoting Teamsters, 431 U.S. at 360, 97 S.Ct. 1843). To be legally sufficient, the plaintiffs’ statistical evidence “must show a disparity of treatment, eliminate the most common nondiscriminatory explanations of the disparity, and thus permit the inference that, absent [464]*464other explanation, the disparity more likely than not resulted from illegal discrimination.” Hervey v. Little Rock, 787 F.2d 1223, 1228 (8th Cir.1986) (quotation omitted). The bottom-line question in this case is whether Plaintiffs produced more than a scintilla of evidence showing UPS engaged in a pattern or practice of discrimination with regard to the class members. The typical McDonnell Douglas burden-shifting framework, while not irrelevant, is not tremendously helpful in this context. Craik, 731 F.2d at 470-71 & n. 7.
A. Upward Mobility
Plaintiffs argue that the district court erred in granting summary judgment to UPS on the denial-of-upward-mobility claims. Plaintiffs contend they proved discrimination in upward mobility under both disparate-impact and disparate-treatment models. Like the district court, we conclude Plaintiffs have failed to establish a prima facie case because their evidence of racial disparities was insufficient, giving rise to no inference of discrimination.
Initially, Plaintiffs offered a Wilcoxon test performed by Dr. Weiner. Plaintiffs argue this test showed that it took longer for black center managers to be promoted to their first division-level job than it did for white center managers. However, UPS’s experts pointed out, and Dr. Weiner admitted, that she did not perform the correct Wilcoxon test. When Dr. Weiner used the correct test, she determined that there was no racial disparity in the pattern of these promotions.
Next, Plaintiffs offered evidence that in thirty-five of UPS’s districts, there were no blacks promoted to division manager between 1989 and 1998. The district court pointed out two problems with this analysis. First, this tally did not take into account whether there were any qualified black employees available for promotion in those districts. UPS’s expert, Dr. Evans, performed an analysis taking the availability of qualified black employees into account and determined that the number of districts in which no blacks were promoted was actually lower than would be expected. Second, even a successful showing of discrimination in some districts does not prove nationwide discrimination. In fact, as the district court pointed out, proof of discrimination in some districts and not others tends to defeat the argument that discrimination was UPS’s nationwide standard operating procedure.
In addition, Plaintiffs presented charts comparing the representation of blacks at the division-manager level to the percentage of all black officials and managers, operatives and laborers, blue collar workers, and the total UPS workforce. The district court properly found that a racial disparity amongst these populations would not give rise to an inference of discrimination. “The ‘proper comparison [is] between the racial composition of [the at-issue jobs] and the racial composition of the qualified ... population in the relevant labor market.’ ” Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 650, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989) (quoting Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 308, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977)) (alterations in original). “[Statistics based on an applicant pool containing individuals lacking minimal qualifications for the job would be of little probative value.” Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 997, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988). The upward-mobility classes consist of center managers who have not been promoted to division manager. Thus, the probative inquiry involves a comparison between the percentage of division managers who are black and the percentage of qualified employees who are black in the population [465]*465from which division managers are chosen. As the district court held, “plaintiffs reliance on a bottom line racial imbalance in the workforce is insufficient to establish that blacks are less likely to be promoted.” Morgan v. UPS, No. 4:94-CV-1184, Mem. & Order at 12 (E.D. Mo. June 26, 2000).
Finally, Dr. Weiner presented a cohort analysis suggesting it takes black employees significantly longer to be promoted from their first full-time supervisor position to division manager than it takes similarly situated white employees. However, Dr. Weiner admitted that the data she looked at “is not appropriate for examining the time to promotion from center manager to division level manager,” id. at 13, and center managers are the class of employees at issue. UPS’s expert, Dr. Evans, performed an analysis on the correct set of promotions and concluded that it actually takes whites longer than blacks to get promoted from center manager to division manager.
Plaintiffs’ proof is insufficient under both the disparate-treatment and disparate-impact models. To avoid summary judgment under the disparate-treatment model through the use of statistics, Plaintiffs must show that “[w]hen the proper labor market is considered, the statistical evidence is sufficient to raise an inference of discrimination.” Paxton v. Union Natl, Bank, 688 F.2d 552, 564 (8th Cir.1982). As discussed above, Dr. Weiner’s correct Wilcoxon test showed no racial disparity, and her other analyses do not take into account the proper labor market — available qualified employees. Thus, the claim fails under the disparate-treatment model.
And, to the extent Plaintiffs raise a disparate-impact claim,2 they have failed to “show relevant statistical disparities permitting an inference of racial imbalance in defendant’s workforce.” Emanuel v. Marsh, 897 F.2d 1435, 1439 (8th Cir.1990). The relevant inquiry was whether promotions from center to division manager were racially discriminatory, taking into account the proper pool of available qualified employees. The evidence presented of overall racial imbalances and even more general trends in upward mobility is insufficient. Plaintiffs did not adduce sufficient evidence to avoid summary judgment.
B. Working Conditions
The district court ruled that Plaintiffs presented insufficient evidence to prove a class-based claim of unequal working conditions. Plaintiffs argue that the district court erred because (1) UPS did not move for summary judgment on working conditions, and (2) Plaintiffs presented sufficient evidence to survive summary judgment.
UPS’s motion for summary judgment was sufficient. UPS moved for summary judgment “on Class Plaintiffs’ Amended and Supplemental Complaint,” which encompassed claims of unequal working conditions. UPS also specifically mentioned working conditions in its motion and asked the court to strike those class claims as unsupported. Plaintiffs were aware that the working-conditions claim [466]*466was at issue because in their response to the summary judgment motion they discussed working-conditions evidence and cited the record on the issue. We therefore agree with the district court that UPS moved for summary judgment on this claim.
We also agree that Plaintiffs’ proof on the working conditions was insufficient to survive summary judgment. While Plaintiffs correctly note that statistical evidence is not required to prove pattern-or-practice discrimination, Catlett v. Mo. Highway & Transp. Comm’n, 828 F.2d 1260, 1265 (8th Cir.1987), the anecdotal evidence here is not enough to support their claims. Plaintiffs’ evidence consists of statements by individual class members made in response to interrogatories, unsigned by the class members. The district court did not err in finding this evidence insufficient. The grant of summary judgment for UPS on the working-conditions claim is therefore affirmed.
C. Pay
Plaintiffs argue that UPS was not entitled to summary judgment on the discriminatory pay claim. They contend that they presented sufficient evidence of a pattern or practice of discrimination with regard to pay in the form of two multiple regression analyses. They say that those analyses give rise to a reasonable inference of discrimination because they show UPS paid white center managers more than black center managers after controlling for many legitimate reasons for such a pay disparity.
A multiple regression analysis attempts to reveal relationships between explanatory variables and a dependent variable. Daniel L. Rubinfeld, Reference Guide on Multiple Regression, in Federal Judicial Center, Reference Manual on Scientific Evidence 181 (2d ed.2000). Explanatory variables are the expected influences on the dependent variable. Id. In this case, pay is the dependent variable. The explanatory variables are what bring race into the picture. In effect, the regression controls for the explanatory variables— those factors that one would expect to influence pay — and then compares the wages of white and black employees. “At best, the regression equation used to assess discrimination in compensation levels can be viewed as a good representation, based on available information, of what factors seem to be related to compensation levels, on average.” Ramona L. Paetzold & Steven L. Willborn, The Statistics of Discrimination: Using Statistical Evidence in Discrimination Cases § 6.01, p.3 (2002). Thus, the selection of explanatory variables is quite important. However, even the best regression equation cannot directly show discrimination because it cannot prove causation. The most it can show is a correlation that can give rise to an inference of discrimination. Id. § 2.05; Rubinfeld, ante, at 183-85. Whether such an inference is reasonable is the legal question we address.
All three experts performed regression analyses, and all agreed that this form of statistical analysis was proper. But the experts came to different conclusions because each of them included different explanatory variables. Plaintiffs’ first expert, Dr. Weiner, found a statistically significant pay disparity of between $1,275 and $2,050 per year. She included two years of measured performance,3 pay region, and time as a supervisor as [467]*467explanatory variables. Plaintiffs’ other expert, Dr. Stapleton, found a statistically significant pay disparity of between $562 and $852 per year. He included two years of measured performance, geographic district, tenure with UPS, time as a center manager, and whether the employee was currently in Human Resources, Air Operations, or Package Operations.
UPS expert Dr. Evans’ preferred method found statistically significant pay differentials in each year except 1995, for which he found no statistically significant difference in pay. This analysis was achieved using as variables “union tenure, part-time supervisor tenure, full-time supervisor tenure, center manager tenure, an indicator for having previously been a division manager and others for having been in air operations or having been a Package Operations Supervisor, indicators for the various districts, and current and past performance.” Dr. Evans concluded that his study demonstrated there was no statistically significant difference in pay between black and white center managers during the class period. He reasoned that 1995 was the only year for which he could do a proper analysis, because it was the only year for which there was sufficient measured performance data. In other years, he said, the analysis incorrectly indicated a disparity because insufficient measured performance data existed. Dr. Evans admitted that when no performance data was included, a statistically significant pay disparity existed.
Because the only evidence that substantiates Plaintiffs’ claim is their regression analyses, and because this sort of evidence is subject to Daubert, two issues arise: (1) admissibility and (2) the propriety of summary judgment. If the analyses were not admissible, then summary judgment was appropriate. If the analyses were admissible, summary judgment may or may not be appropriate, depending on whether they create a reasonable inference of a pattern or practice of disparate pay.
. It is not clear whether the district court deemed Plaintiffs’ regressions inadmissible or concluded that they did not create a genuine issue of fact. The district court found that “the variables are properly included in Dr. Evans’ analysis.” Morgan, Mem. & Order at 17. And it concluded, “defendants have shown that any pay disparity between black and white center managers is caused by factors other than race.” Id. at 18.
We defer to admissibility determinations under an abuse-of-discretion standard of review, even at summary judgment. Gen. Elec. & Co. v. Joiner, 522 U.S. 186, 143, 118 S.Ct. 512, 189 L.Ed.2d 508 (1997). The district court did not mention Daubert, but its analysis plainly hinges on problems with Plaintiffs’ analyses that are arguably methodological — what explanatory variables are proper. Daubert, 509 U.S. at 592, 113 S.Ct. 2786 (stating that admissibility determinations “entail[ ] a preliminary assessment of whether the reasoning or methodology underlying the testimony is scientifically valid and of whether that reasoning or methodology properly can be applied to the facts in issue”); accord Eckelkamp v. Beste, 315 F.3d 863 (8th Cir.2002) (applying deferential review to evidentiary determination plainly premised on the methodological underpinnings of an expert’s opinion). But because the district court denied UPS’s motions relating to Daubert and Rule 702 as moot,4 we are not inclined to accord the [468]*468trial court’s decision that degree of deference. Moreover, the question of what explanatory variables should be included in a particular regression normally “affect[s] the analysis’ probativeness, not its admissibility.”5 Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986).
While the omission of variables from a regression analysis may render the analysis less probative than it otherwise might be, it can hardly be said, absent some other infirmity, that an analysis which accounts for the major factors must be considered unacceptable as evidence of discrimination.
[I]t is clear that a regression analysis that includes less than all measurable variables may serve to prove a plaintiffs case.
Id. at 400, 106 S.Ct. 3000 (quotation omitted).
Admissible regressions, however, do not necessarily mean summary judgment is inappropriate. See Daubert, 509 U.S. at 596, 113 S.Ct. 2786 (“[I]n the event the trial court concludes that the scintilla of evidence presented supporting a position is insufficient to allow a reasonable juror to conclude that the position more likely than not is true, the court remains free to ... grant summary judgment .... ”). While a trial court may not weigh the evidence or consider credibility at summary judgment, Anderson, 477 U.S. at 255, 106 S.Ct. 2505, it can use the scale to a limited degree; it must ensure that the nonmoving party has at least a scintilla of evidence in support of its position. In so doing, “the court must draw all reasonable inferences in favor of the nonmoving party.” Reeves v. Sanderson Plumbing Prods., Inc., 530 U.S. 133, 150, 120 S.Ct. 2097, 147 L.Ed.2d 105 (2000); accord Carter v. Chrysler Corp., 173 F.3d 693, 701 (8th Cir.1999).
“[G]ross statistical disparities ... alone may in a proper case constitute prima facie proof of a pattern or practice of discrimination.” Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977). And the usefulness of statistics “depends on all of the surrounding facts and circumstances.” Teamsters, 431 U.S. at 340, 97 S.Ct. 1843. Here, the regression analyses are all the evidence of discriminatory pay that Plaintiffs have, so those analyses must show a gross statistical disparity and this must be a proper case — a case in which the gross disparity can give rise to a reasonable inference that paying blacks less because they are black is UPS’s standard operating procedure.
A reasonable inference of such discrimination does not arise in this case. First, Plaintiffs’ regressions fail to take into account past pay, performance, or both. Second, when evaluated in the context of the undisputed facts, the regressions fail to create a triable case of pattern-or-practice discrimination.
Evaluating the propriety of explanatory variables is a somewhat comparative exercise. When a deféndant attacks a plaintiffs regression, he must typically do more than point out the flaws in his opponent’s analysis. Instead, the defendant must show that the omission had an impact on [469]*469the result. See Catlett, 828 F.2d at 1266.6 But, “there may be a few instances in which the relevance of a factor ... is so obvious that the defendants, by merely pointing out its omission, can defeat the inference of discrimination created by the plaintiffs’ statistics.” Palmer v. Shultz, 815 F.2d 84, 101 (D.C.Cir.1987).
Selecting proper explanatory variables is a function of the particular employer’s compensation determinations. Under UPS’s pay scheme, center-manager pay has two components: base pay and salary increases. A center manager’s base pay is determined when he becomes a center manager, subject to later salary increases based on performance. An individual’s base pay remains constant during his tenure as a center manager. Center-manager base pay is calculated in part by using the individual’s prior pay, generally as a supervisor. Supervisor pay has the same two elements — base pay at the time of promotion and salary increases based on performance during one’s tenure as a supervisor. Supervisor base pay is calculated using that individual’s prior hourly wage. Individuals in certain operations (e.g., air operations) are paid more as a part of their job; thus, their hourly rate, and supervisor and center-manager base pay are higher. Union membership could have a similar effect.
The discrimination alleged in this claim is that black center managers, between 1991 and the time of judgment, were paid less than similarly situated white center managers because they were black. All experts agree that their regressions revealed no racial disparity in terms of center-manager pay raises for the class period. So in order for the pay of black class members to be lower than that of white center managers, the discrimination had to occur with regard to center-manager base pay. Because the base pay is set according to past pay, variable(s) representing that factor are important.
As UPS points out, both Dr. Weiner and Dr. Stapleton excluded past pay from their regressions. Dr. Stapleton admitted that if past pay was included in the analysis, then no statistically significant disparity between white and black pay existed.7 Plaintiffs argue that the exclusion was justified under Bazemore v. Friday, 478 U.S. 385, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986). We disagree.
Bazemore’s relevance involves the somewhat related questions of what role past pay has in a pay discrimination case and when an explanatory variable is properly omitted from a regression analysis. Plaintiffs correctly note that Bazemore means that past discriminatory pay cannot be a legitimate excuse for a later pay [470]*470disparity. 478 U.S. at 394-96, 106 S.Ct. 3000. 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. See Paetzold & Willborn, ante § 6.13 (discussing the problem of tainted variables). Thus, a regression could be probative of discrimination without taking into account past pay if that past pay was set according to race. This is because past discriminatory pay makes current pay disparities actionable8 under Bazemore, and because the variable can be omitted from a regression if tainted by the discrimination at issue.
In Bazemore, both the district court and the court of appeals found that past pay was set according to race. 478 U.S. at 390-91, 394, 401-02, 106 S.Ct. 3000. Thus, the salary figures used in the plaintiffs’ regression in that case were proper even though past discriminatory pay accounted for some of the later disparity. In other words, Bazemore’s regressions did not need to account for past pay because there was evidence of past discrimination. Here, Plaintiffs would be excused from accounting for past pay if they could show past discrimination. We see no such evidence.
Plaintiffs appear to justify the omission of past pay by pointing to the class-period disparity that their experts found, inferring discrimination, and claiming that a reasonable inference is that past pay disparities were also race based. This argument is circular. The correlation between race and center-manager base pay from 1991 on exists only if past pay is omitted from the regression. Past pay can only be omitted if it is somehow linked to race. Plaintiffs’ only evidence of discrimination in past pay is the apparent correlation between race and center-manager base pay during the class period. But that correlation is what Plaintiffs have evidence of only by omitting past pay. They have no evidence, statistical or otherwise, that past pay disparities were racially discriminatory. This sort of bootstrapping cannot create an inference of discrimination with regard to either class-period base pay or past pay.
The variables Plaintiffs did use also do not account for the role of past pay. For example, because performance affects past pay, performance variables arguably could be proxies for past pay.9 Dr. Stapleton appreciated this aspect of the case but refused to include “lagged” performance measures precisely because “it’s really just explaining current pay experience by past pay experience.” And he offered this past-pay rationale as his reason for omitting most other experiential factors that affect pay at UPS. As explained above, this observation could be meaningful if there had been some evidence that past pay experi[471]*471ence was tainted by discrimination. There was none here and the assumption that past pay should be omitted, as well as most factors related to it, was unjustified.
There are a variety of other concerns that reduce the import of Plaintiffs’ regressions. The performance data, which was included in all experts’ regressions to some extent, poses a problem. Performance data from performance evaluations that UPS conducted existed for 1991, 1992, 1993, 1994, and 1995.10 Performance was certainly relevant to the employer’s pay decisions, but it appears to us that performance could only affect center-manager base pay indirectly' — by influencing past supervisor pay through past supervisor salary increases, past hourly rates, or both. So Plaintiffs’ regression should have utilized past performance as far back as was available, rather than just the two years they did include.
UPS offered its own regression indicating the omission had an impact. Dr. Evans conducted a regression for 1995, included all available performance data, and concluded that no statistically significant pay disparity remained — i.e., the inclusion of performance eliminated the significance of race. We think this was sufficient to show the adverse impact of omitting available performance data and showed that other performance, even though not measured by evaluations, was important. At the very least, the fact that no correlation existed in 1995 severely damages the inference that UPS had a nationwide standard operating procedure of paying blacks less than whites.
We also have no indication of which employees’ performance data was included in the regressions. Given the nature of the claim — that center managers’ base pay was lower for blacks than whites — and the faet that center-manager base pay is set only at the time of promotion, performance prior to the time of promotion is all that is relevant. Thus, for class members that were promoted to center managers before 1991, there are no performance evaluations from which to draw data. And for individuals that were promoted after 1991, until 1995, the only relevant performance data comes from pre-promotion evaluations. Post-promotion performance could simply have no affect on base pay. And such performance seems too attenuated an indicator of pre-promotion performance to be included.
The nature of UPS’s management system also makes a claim of nationwide pattern-or-practice discrimination difficult. Pay was set according to a decentralized scheme in which district managers were given the discretion to set pay. With approximately seventy districts in the United States, each with their own manager, the statistical evidence would have to be quite strong to raise, by itself, an inference of nationwide discrimination.
One of the most important flaws in Plaintiffs’ case is that they adduced no individual testimony regarding intentional discrimination. As mentioned above, Plaintiffs’ purported anecdotal evidence was insufficient for the working-conditions claim, and we see none with regard to pay. Although such evidence is not required, the failure to adduce it “ ‘reinforces the doubt arising from the questions about validity of the statistical evidence.’ ” EEOC v. Sears. Roebuck & Co., 839 F.2d 302, 311 (7th Cir.1988) (quoting Griffin v. Board of Regents, 795 F.2d 1281, 1292 (7th Cir.1986)). Such evidence would bring the “cold numbers convincingly to life.” Teamsters, 431 U.S. at 339, 97 S.Ct. 1843.
[472]*472Finally, as explained above, there was no apparent discrimination with regard to pay raises, promotions, or working conditions. Statistical evidence like that adduced here, at best, can only show a correlation between race and the dependent variable that gives rise to a further inference of pattern-or-praetice discrimination in a proper case. Here the flaws are too many and the probative value too slight for Plaintiffs’ regressions to carry the day by a preponderance of the evidence in the context of this case. It would be manifestly unreasonable to infer from Plaintiffs’ regression analyses that UPS set center managers’ base pay lower for blacks as a matter of practice all across the country during the period in question. Accordingly, summary judgment was appropriate.
D. Remaining Issues
We affirm the two remaining issues on appeal. First, the district court did not err in denying the motions to intervene brought by Frank Jackson and Charles Cartwright. Second, with regard to UPS’s cross-appeal, the district court properly denied as moot the motion to decertify the classes.
IV. CONCLUSION
We affirm.