concurring in part and dissenting in part:
I.
I concur in the majority opinion as to Section 111(A) (the district court’s determination on the numerosity factor), that portion of Section 111(B)(2) reversing the district court’s judgment as to the hostile work environment claim concerning the commonality factor,1 that portion of Section III(C) as to the typicality factor for the hostile work environment claim only, and Section III(D) reversing the district court’s judgment as to the adequacy of representation.2 However, because I believe the majority opinion fails to adhere to the standard of review and usurps the role of the district court, I respectfully dissent as to sections 111(B)(1), (2), and III(C) regarding the commonality and typicality factors for the disparate treatment and disparate impact claims. For the following reasons, I would affirm the district court’s judgment not to grant class certification as to those claims.
II.
“We review the district court’s certification decision for abuse of discretion.” *161Gregory v. Finova Capital Corp., 442 F.3d 188, 190 (4th Cir.2006) (citing McClain v. S.C. Nat. Bank, 105 F.3d 898, 902 (4th Cir.1997)). It is the party seeking class certification who bears the burden of proving the requirements of Rule 23. Lienhart v. Dryvit Sys., Inc., 255 F.3d 138, 146 (4th Cir.2001); Int’l Woodworkers of Am. v. Chesapeake Bay Plywood Corp., 659 F.2d 1259, 1267 (4th Cir.1981). “A court has broad discretion in deciding whether to allow the maintenance of a class action.” Roman v. ESB, Inc., 550 F.2d 1343, 1348 (4th Cir.1976).
A district court has abused its discretion if its decision “is guided by erroneous legal principles” or “rests upon a clearly erroneous factual finding.” Westberry v. Gislaved Gummi AB, 178 F.3d 257, 261 (4th Cir.1999). We do not ask whether we would have come to the same conclusion as the district court if we were examining the matter de novo. Fellheimer, Eichen & Braverman, P.C. v. Charter Techs., 57 F.3d 1215, 1223 (3d Cir.1995). Rather, after reviewing the record and the reasons the district court offered for its decision, we reverse for abuse of discretion if we form “a definite and firm conviction that the court below committed a clear error of judgment in the conclusion it reached upon a weighing of the relevant factors.” Westberry, 178 F.3d at 261.
Morris v. Wachovia Sec., Inc., 448 F.3d 268, 277 (4th Cir.2006); see also Thorn v. Jefferson-Pilot Life Ins. Co., 445 F.3d 311, 317 (4th Cir.2006) (“A district court per se abuses its discretion when it makes an error of law or clearly errs in its factual findings.”).
In the class certification context of Rule 23, our decisions have specifically acknowledged that there is no abuse of discretion, as a matter of law, when a reasonable jurist could have made the decision at issue based on the evidence in the record.
[A] great deal of deference must be afforded to a District Court’s decision to deny certification of a plaintiff class for a class action. Indeed, even if reasonable persons can disagree as to whether the class certification was proper, the District Court’s decision certainly does not amount to an abuse of discretion.
Simmons v. Poe, 47 F.3d 1370, 1381-82 (4th Cir.1995). “[W]e cannot hold that the district court’s refusal to expand the class ... was an abuse of discretion, though reasons clearly existed for taking the other course, and taking it surely would not either have abused discretion.” Lewis v. Bloomsburg Mills, Inc., 773 F.2d 561, 564 (4th Cir.1985).
On appeal from an order granting or denying certification, the appellate court does not reweigh the evidence de novo, but is to accord the district court’s decision “broad discretion.” See Lienhart, 255 F.3d at 146 (Within the framework of Rule 23 “[a] district court has broad discretion in deciding whether to certify a class.” (internal quotations omitted)).
At its core, the district court determined the appellants could not meet either the commonality or typicality factors for certification under Rule 23(a) because they failed to meet their burden of proof as to the direct or statistical evidence of discriminatory promotions. The record in this case reflects the district court’s judgment was neither “guided by erroneous legal principles” nor grounded “upon a clearly erroneous factual finding.” Thus, when the full context of the appellants’ proffered evidence is examined and the standard of review is observed, the district court cannot be said to have abused its discretion in denying certification because a reasonable jurist could have reached that decision on this record.
*162While the majority opinion is correct that a certification determination is not a judgment on the merits of the underlying claims, neither is it a pleadings-based determination that can be verified by any evidence, no matter how deficient. Certainly since the United States Supreme Court’s decision in General Telephone Co. v. Falcon, 457 U.S. 147, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982), district courts have been required to undertake a “rigorous analysis” not just of the plaintiffs’ claims as pled, but of the evidence to support those claims in order to make an appropriate judgment on Rule 23 certification. 457 U.S. at 161, 102 S.Ct. 2364 (“[A] Title VII class action, like any other class action, may only be certified if the trial court is satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.”). “[Sjometimes it may be necessary for the court to probe behind the pleadings before coming to rest on the certification question.” Id. at 160, 102 S.Ct. 2364.
In the case at bar, the district court followed its duty and performed that rigorous analysis, although with a result that the majority may not have reached if they were sitting as the trier of fact. This Court has been clear in explaining that an objection to examining the merits of a class certification claim, as reflected by the majority opinion’s citation to Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974), supra at 156-57, was rendered invalid after Falcon. The district court’s role, and ours, as explained in Gariety v. Grant Thornton, LLP, 368 F.3d 356 (4th Cir.2004), is telling:
If it were appropriate for a court simply to accept the allegations of a complaint at face value in making class action findings, every complaint asserting the requirements of Rule 23(a) and (b) would automatically lead to a certification order, frustrating the district court’s responsibilities for taking a “close look” at relevant matters, Amchem, 521 U.S. at 615, 117 S.Ct. 2231, 138 L.Ed.2d 689, for conducting a “rigorous analysis” of such matters, Falcon, 457 U.S. at 161, 102 S.Ct. 2364, and for making “findings” that the requirements of Rule 23 have been satisfied, see Fed.R.Civ.P. 23(b)(3). Moreover, if courts could only consider the pleadings, then “parties would have wide latitude to inject frivolous issues to bolster or undermine a finding of predominance.” Robert G. Bone & David S. Evans, Class Certification and the Substantive Merits, 51 Duke L.J. 1251, 1269 (2002).
When Rule 23(c), which originally required certification orders to be made “as soon as practicable after commencement of [the] action,” was amended in 2003 to require the court to determine class certifications “at an early practicable time,” the Advisory Committee on Civil Rules explained the preexisting and longstanding practice that prompted the change:
Time may be needed to gather information necessary to make the certification decision. Although an evaluation of the probable outcome on the merits is not properly part of the certification decision, discovery in aid of the certification decision often includes information required to identify the nature of the issues that actually will be presented at trial. In this sense, it is appropriate to conduct controlled discovery into the “merits, ” limited to those aspects relevant to making the certification decision on an informed basis.
Fed.R.Civ.P. 23 advisory committee’s note to 2003 amendments (emphasis added).
*163The Eisen decision, upon which the district court relied, does not require a court to accept plaintiffs’ pleadings when assessing whether a class should be certified. In Eisen, the Supreme Court held that the district court’s preliminary hearing on the merits of the case — concluding that the plaintiff was “more than likely” to prevail — was inappropriate for the purpose of determining whether a class action could be maintained. 417 U.S. at 177-78, 94 S.Ct. 2140, 40 L.Ed.2d 732. Eisen simply restricts a court from expanding the Rule 23 certification analysis to include consideration of whether the proposed class is likely to prevail ultimately on the merits. See Castano, 84 F.3d at 744; 5 Moore’s Federal Practice ¶ 23.84[2][a] (3d ed.2003). As the Supreme Court itself stated in a post-Eisen case, “sometimes it may be necessary for the [district] court to probe behind the pleadings before coming to rest on the certification question.” Falcon, 457 U.S. at 160, 102 S.Ct. 2364; see also Coopers & Lybrand v. Livesay, 437 U.S. 463, 469, 98 S.Ct. 2454, 57 L.Ed.2d 351 (1978) (“[T]he class determination generally involves considerations that are enmeshed in the factual and legal issues comprising the plaintiffs cause of action” (internal quotation marks omitted)).
Thus, while an evaluation of the merits to determine the strength of plaintiffs’ case is not part of a Rule 23 analysis, the factors spelled out in Rule 23 must be addressed through findings, even if they overlap with issues on the merits. Eisen’s prohibition against assessing plaintiffs’ likelihood of success on the merits as part of a Rule 23 certification does not mean that consideration of facts necessary to a Rule 23 determination is foreclosed merely because they are required to be proved as part of the merits. The analysis under Rule 23 must focus on the requirements of the rule, and if findings made in connection with those requirements overlap findings that will have to be made on the merit s, such overlap is only coincidental.
368 F.3d at 365-66.
The majority’s citation to Thorn v. Jefferson-Pilot Life Ins. Co., 445 F.3d 311 (4th Cir.2006), is also puzzling, as this post-Gariety decision acknowledged the type of inquiry undertaken by the district court in the case at bar:
At the class certification phase, the district court must take a “close look” at the facts relevant to the certification question and, if necessary, make specific findings on the propriety of certification. Gariety, 368 F.3d at 365 (internal quotations omitted). Such findings can be necessary even if the issues tend to overlap into the merits of the underlying case. Falcon, 457 U.S. at 160, 102 S.Ct. 2364.
Thorn, 445 F.3d at 319.
Thus, the district court’s examination and evaluation of the plaintiffs’ evidence to support the class certification claim was not only appropriate, but required under Falcon and Gariety. That the majority would have reached a different conclusion, albeit lacking all the advantages of the district court’s first-hand knowledge of the case, establishes nothing because the standard of review goes to what a reasonable jurist could have found, which may very well lead to differing views on the same set of facts. See Morris, 448 F.3d at 277 (“We do not ask whether we would have come to the same conclusion as the district court if we were examining the matter de novo.”) (citing Fellheimer, Eichen & Braverman, P.C. v. Charter Techs., 57 F.3d 1215, 1223 (3rd Cir.1995)).
*164In my view, the majority opinion steps beyond the standard of review and improperly reweighs the evidence so as to achieve a certification that was properly within the district court’s discretion to deny. The majority appears to sanction a rule of mandated class certification even where a reasonable jurist could properly find the named plaintiffs’ evidentiary basis was insufficient. Such a result is not consonant with the appellate standard of review.
III.
The majority opinion initially rests on the conclusion that the appellants’ direct evidence alone was sufficient to require class certification of the claims for disparate treatment and disparate impact. Supra at 152-53. I agree with the majority opinion that direct evidence could be sufficient, in a proper case, to support a certification decision without corollary statistical proof, although that is not a common practice in at least the disparate impact context. Nonetheless, the majority’s conclusion — that the direct evidence alone was sufficient to merit certification — seems an odd decisional basis because that argument was not made by the appellants in the district court. Neither the appellants’ Statement of Issues, briefs or oral argument raised, as a point of error, the failure of the district court to grant certification on the basis of direct evidence alone.
The appellants’ argument below was that the direct evidence “bolstered” their main evidentiary claim, which was their statistical analysis. See, e.g., J.A. 965 (“The plaintiffs have also bolstered the statistical evidence of a pattern or practice of subjective bias with both direct and anecdotal evidence of actual racial bias by the promotion decision-makers ....”) (emphasis added); J.A. 8473 (“The commonality and typicality of plaintiffs’ pattern-or-practice claim is further bolstered by the subjective nature of the interview and selection stage of Nucor’s promotion process.”) (emphasis added); Br. of Appellant at 37 (“Plaintiffs did not rely on the mere existence of a subjective promotion procedure, but presented a combination of statistical, anecdotal and direct evidence showing that such procedure was racially discriminatory throughout the plant.”). Nonetheless, as noted below, when the appellants’ limited direct evidence is examined, a reasonable jurist could find that evidence failed to meet the appellants’ burden of proof as to either commonality or typicality for the certification of a class of all African-American plant employees.
Similarly, the majority opinion appears to reweigh the evidence in order to sustain the appellants’ statistical model. As noted below, the district court was within its proper exercise of discretion to accord the statistical evidence little or no weight in the class certification decision. A reasonable jurist, on this record, could find the appellants’ statistical evidence too speculative and lacking a proper foundation so as to be without evidentiary value in meeting the appellants’ burden of proof.
A. The Direct Evidence
The appellants did not argue to the district court that their “direct evidence” was alone sufficient to support a finding of commonality or typicality for class certification under Rule 23. They presented direct evidence to “bolster” their statistical evidence, not as standalone proof for Rule 23(a) purposes. However, the appellants’ choice of methodology is not the dispositive issue on appeal. What the standard of review should examine in this case is whether the district court, based on the direct evidence alone, could have reasonably found that evidence insufficient to meet the appellants’ burden of proof on typicality and commonality for the pro*165posed class. That a reasonable jurist could have also found the same evidence sufficient for certification is irrelevant to the abuse of discretion standard where the evidence also could support the opposite finding. It appears plain, on this record, that a reasonable jurist could have found, as the district court did, that the appellants’ direct evidence (though considered as only supplemental evidence) was simply inadequate to meet their burden of proof.
Under Rule 23(a)(2) and (3), the appellants were required to prove there were “questions of law or fact common to the class” and “claims ... of the representative parties are typical ... of the class.” Fed.R.Civ.P. 23. The purported class, all former and current African-American employees of the Nucor plant during the relevant time period, worked in all the plant’s departments, which include at least the beam mill, hot mill, melt shop, cold mill and shipping departments. Thus, the appellants’ claims must have common questions among the employees in all these departments and be typical of those claims. The appellants’ direct evidence, however, could reasonably be determined as failing to meet the burden of proof for either commonality or typicality for the purported class.
The statements of the three affiants cited by the majority do claim race-based denials of promotions, the employment practice at issue in their disparate treatment and disparate impact claims. However, all three were employees only in the beam mill. While some of the appellants’ declarations allege employment discrimination in other departments as to them as beam mill employees, they fail to do so as to non-beam mill employees.
The district court noted that of all the appellants’ direct evidence, only two non-beam mill employees allege discrimination in another department. Even the statement alleging bias of a supervisor was a statement by the department manager in charge of the beam mill about employees in the beam mill.
[T]he existence of a valid individual claim does not necessarily warrant the conclusion that the individual plaintiff may successfully maintain a class action. It is equally clear that a class plaintiffs attempt to prove the existence of a companywide policy, or even a consistent practice within a given department, may fail even though discrimination against one or two individuals has been proved.
Cooper v. Fed. Reserve Bank of Richmond, 467 U.S. 867, 877-78, 104 S.Ct. 2794, 81 L.Ed.2d 718 (1984) (characterizing the holding in Falcon); see also Int’l Bd. of Teamsters v. United States, 431 U.S. 324, 336, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977) (to prove a pattern or practice violation a plaintiff must “prove more than the mere occurrence of isolated or ‘accidental’ or sporadic discriminatory acts” but must “establish by a preponderance of the evidence that racial discrimination was the company’s standard operating procedure— the regular rather than the unusual practice”).
On this record, a reasonable jurist could conclude the appellants failed to meet their burden of proof to show common issues and typical claims between themselves, as employees in the beam mill seeking promotions, and the workers in all the other departments who sought promotions. The appellants simply failed to produce that evidence.
The effect of this failure of proof, for certification purposes, is that the appellants failed to prove commonality or typicality for the non-beam mill employees they seek to represent. In that circumstance, a reasonable jurist, like the learned district court judge in this case, could find *166the appellants failed to show commonality or typicality for certification of a class consisting of all African-American plant employees. Accordingly, the district court acted within its discretion to deny class certification on the basis of the appellants’ direct evidence.3
B. Statistical Evidence
Nucor provided the appellants with actual promotion-related job posting and applicant data from January 2001 to February 2006.4 According to the appellants’ own expert, statistical analysis of this data resulted in a standard deviation of - .84, a result indicating no statistically significant disparity in the actual versus expected percentage of African-American employees successfully bidding on jobs. J.A. 5872-73. The appellants, however, objected to use of the 2001 to 2006 time period because it included job posting and bidding information occurring after suit was filed at the end of 2003.5 The district court agreed and held “that the most reliable statistics are those gathered from [actual] promotion data occurring before this action was filed”, and determined that “post-suit promotion data is entitled to minimal weight.” J.A. 8986 n. 4.
With the post-lawsuit data favorable to Nucor excluded from consideration, the appellants’ experts calculated the actual versus expected standard deviation based on the actual job posting and bidding data from 2001 to 2003 as - 1.48.6 In other words, the appellants were unable to produce statistically significant evidence of race discrimination in promotions based on actual job posting data.7 See Castaneda v. Partida, 430 U.S. 482, 496 n. 17, 97 S.Ct. 1272, 51 L.Ed.2d 498 (1977) (recognizing *167that differences between the expected value and the observed number greater than two or three standard deviations are significant); Warren v. Halstead Indus., Inc., 802 F.2d 746, 759 n. 17 (4th Cir.1986) (recognizing that a standard deviation greater than two or three excludes “chance” as the cause of under-representation).
Since the appellants could not prevail based on the actual promotions data from 2001 through 2003, they sought to create additional data that would dilute the actual data analysis and produce the needed - 2.00 or greater standard deviation. The appellants contended they could do so by comparing “the estimated percentage of blacks who sought promotions between December 1999 and December 2003 with the estimated percentage of blacks who received promotions during that period.” Supra at 154. However, their proposed data substitute (the “2000 data”) was only estimated for 2000 because the bulk of the time period (from 2001 to 2003), reflected actual job postings for which African-American employees were known to have applied.8, 9 The appellants contended to the district court that if their version of the promotion statistics for the 2000 through 2003 period (four years) was considered instead of three years (2001 to 2003), they could show the required standard deviation to make a prima facie case of discrimination. It is the appellants’ means of creating the 2000 data that validates the district court’s exercise of discretion in discounting that analysis because the 2000 data was without a valid foundation, purely speculative, and thus not entitled to probative weight. Even the appellants’ own experts agreed the 2000 data suffered from significant infirmities.
Dr. Edwin Bradley, one of the appellants’ statistical experts, stated that “[a]ll statistics involve a comparison between a benchmark of what was expected to occur and what actually happened.” J.A. 404 (emphasis added). Thus, there were two variables that the appellants’ experts, Dr. Bradley and Dr. Liesl Fox, needed to formulate as to what was expected to occur in 2000 and what actually happened: (1) the number of posted job promotions available for bidding by the existing Nucor employees during 2000, and (2) whether those job promotions were “similarly situated” as defined by the district court (i.e., job promotions for which at least one African-American employee applied). The assumptions required for both variables undermine the validity of the 2000 data.
In order to construct the first variable, the number of promotion positions available in 2000 open for employee bid, the appellants relied on twenty-seven “change-of-status” forms culled from the personnel files produced by Nucor. It is not clear from the record whether the twenty-seven change-of-status forms represent all, or only selected, jobs open for promotion in 2000 because these forms are not identified. Nucor argues that the change-of-status forms “are simply a company record *168which documents any employee’s change of status, whether the employee was promoted, demoted, received a standard pay increase, or was transferred.” Appellee’s Br. at 36.
Indeed, the change-of-status forms found in the record for 2000 fail to bolster the appellants’ claim that those forms reflect only promotion positions open to bid. For example, nine change-of-status forms appear in the record dated between December 1999 and January 2001 (including three in January 2001). J.A. 8397-99, 8406-07, 8416, 8673-74, 8701. Of these forms, one (J.A. 8399) simply reflects an increase in pay for an existing employee. Two (J.A. 8397, 8406) reflect completion of probation for an existing employee. Another (J.A. 8407), reflects a new hire and not a promotion. The remaining five reflect promotions of some type, though only one (J.A. 8673) identifies a promotion acquired through a bid process. On this record, it is difficult, if not impossible, to discern whether the 2000 data based on the nebulous change-of-status forms proves those positions were promotion positions available for employee bidding and thus relevant to the formulation of statistical evidence for the appellants’ claims. Perhaps that is one reason the district court found not that the 2000 data represented twenty-seven positions reflecting promotion openings available for bidding but that “the plaintiffs searched through personnel files provided by Nucor and located twenty-seven positions filled between December 1999 and January 2001.” J.A. 8984 (emphasis added).
But even if one assumes the twenty-seven change-of-status forms represented actual promotion openings in 2000, Nucor’s expert, Dr. Finis Welch, testified “it is not known whether these were posted positions” (i.e. available for bid by employees). J.A. 5911 n. 1. Thus, based on the record, or lack of it, to conclude that appellants established the first variable necessary for their statistical analysis to verify the 2000 data is highly dubious.
For the second variable, whether the twenty-seven positions represent positions “similarly situated,” Drs. Bradley and Fox assumed “that the racial composition of the bidding pool [applicants] for those jobs was the same as the weighted average of the racial composition of the bidding pools (applicants) for the” 2001-2003 period. Supra at 154; J.A. 8984. However, through its discovery orders the district court had defined “similarly situated” jobs to mean those on which an African-American employee actually bid. Drs. Bradley and Fox could only confirm that one of the twenty-seven “promotions” in the 2000 data involved an African-American employee actually bidding on the job.10 J.A. 5853. Dr. Fox testified in her deposition as follows:
Counsel: Okay. Now, what determine-— what did you do to determine that these [27] job selections met the Court’s definition of the same or similar?
Dr. Fox: Job title.
Counsel: Okay. How many of these job selections involved — these 27 job selections involved an African-American actually bidding on the job?
Dr. Fox: At least one.
Counsel: Okay. Other than the one, how many involved?
Dr. Fox: I don’t know.
Counsel: You don’t know. You have no way of knowing that?
Dr. Fox: No.
*169Counsel: Okay. Is not the definition of similarly situated, does that not also include a job on which an African-American actually bid?
Dr. Fox: That was the Court’s order, yes.
Counsel: Okay. So have you made any attempt in your analysis of these 27 to bring them within the Court’s definition of same or similar?
Dr. Fox: Those were destroyed. It’s not possible to do that.11
J.A. 5852-54. Instead, Dr. Bradley and Dr. Fox determined the jobs were “similarly situated” based on the similarity of the job title alone. J.A. 5854.
Dr. Welch criticized these assumptions because “it is not known whether ... any African-American employees applied” for the twenty-seven job selections.12 J.A. 5911 n. 1. Dr. Welch also believed that “Drs. Bradley and Fox may have missed two selections of African-Americans into a job that is included in the posting data,” and that there were “African Americans, including a named plaintiff, who moved into a job prior to 2001 that is a job held by employees at the time they won a position in the posting data.” J.A. 5911-12 n. 2, 3.
Drs. Bradley and Fox also recognized that
Bidding records [from 2001 to 2006] were provided only for “similarly situated” jobs for which at least one African-American bid. That means that any postings which were for “similarly situated” jobs but had no African-American bidders, and therefore all white bidders, were not included in our calculation. Thus, the African-American representation among bidders for “similarly situated” jobs provided by Nueor-Berkeley is necessarily inflated.
J.A. 583. According to Dr. Welch, this necessarily means that by applying the 2001 to 2003 weighted average of the bid pool applicants to the 2000 data, “they are overstating the expected number of African-American selections.” J.A. 5912. In Dr. Welch’s view, “there is no reason to assume that the applicant pools for each of [the twenty-seven] positions necessarily included at least one African American.” J.A. 5912 n. 4. Thus, the appellants’ support in the record for the second variable, that the twenty-seven positions used to create the 2000 data were positions for which it could be legitimately assumed at least one African-American applied, is as dubious as the first variable.
The appellants certainly were entitled to proffer valid statistical evidence for 2000 upon which to expand the standard deviation analysis. That being said, however, no precedent or principle of law requires inherently unreliable evidence to be given evidentiary weight. The appellants chose a method of statistical proof, but based it on evidence a trier of fact could determine was inherently faulty.13 Consequently, a *170reasonable jurist could have determined the appellants failed to meet their burden of proof for the Rule 23(a) factors through the proffered statistical evidence. Accordingly, the district court did not abuse its discretion by failing to give the appellants’ 2000 data, and the resulting standard deviation analysis, evidentiary weight and denying certification.14
IV.
Even if we assume that the 2000 data (and its resulting standard deviation analysis) was entitled to some evidentiary weight, the precedent cited by the majority opinion does not support the conclusion that the district court abused its discretion in refusing certification as to the disparate treatment and disparate impact claims. Our decision in United States v. County of Fairfax, 629 F.2d 932 (4th Cir.1980), does not mandate that a district court must give evidentiary weight to every datum a prospective class representative presents just to increase the strength of their statistical proof.
Moreover, there are significant distinctions between the case at bar and County of Fairfax. A distinction of particular importance is that the “missing” data in County of Fairfax involved three out of the four years from which the statistical analysis was to be made. There was no credible basis in that case by which to use the one year of actual data to foster a valid analytical sample. One year’s data was simply insufficient upon which to make a statistical analysis. Accord EEOC v. Am. Nat’l Bank, 652 F.2d 1176, 1195 (4th Cir. 1981). In contrast, the case at bar is the exact opposite, with three years of actual data. There is no precedent that requires a court in all cases to add an additional period of data, particularly data lacking in evidentiary credibility.15 Moreover, Fair-fax County, as a municipality, was under a statutory duty to maintain employment data in the 1970s, a duty which it failed to fulfill. No such duty is applicable to Nu-cor. See County of Fairfax, 629 F.2d at 937 n. 4 (“In violation of the record keeping regulations of the Revenue Sharing Act and the Crime Control Act, defendants had destroyed pre-1978 applications for employment.”).
Another distinction is that County of Fairfax involved an appeal from the district court’s judgment on the merits of the case, not a denial of class certification under Rule 23. Thus, the case at bar involves a different legal analysis and is subject to an abuse of discretion standard of review on appeal.
Further, although County of Fairfax involved claims of discrimination “in recruitment, hiring, assignments, and promotions,” 629 F.2d at 936, the sufficiency of the applicant flow data in that case involved only the hiring data, not promotions data. Unlike Wrings, which came from the surrounding community in County of Fairfax, Nucor only promotes from within its existing workforce. The data universe for hiring in County of Fairfax was thus much larger and without restriet*171ing variables that made projections on estimated data problematic. However, in the case at bar, the data universe contained restrictive variables limiting extrapolations because only current Nucor employees who were qualified and who in fact bid on a promotion could comprise that group.
County of Fairfax does not stand for the proposition that class certification claimants are entitled to a waiver of the rules of evidence for purposes of meeting their burden of proof. The district court’s judgment was not based on erroneous legal principles and was thus not an abuse of discretion.
V.
When an appellate court reviews a trial court’s determination upon an abuse of discretion standard, it must accord that court’s factual conclusions supported by the record proper deference. See United States v. Pittman, 209 F.3d 314, 316 (4th Cir.2000) (The abuse of discretion “standard of review mandates a significant measure of appellate deference to the judgment calls of trial courts.”); see also United States v. Mason, 52 F.3d 1286, 1289 (4th Cir.1995) (“Under the abuse of discretion standard, this Court may not substitute its judgment for that of the district court; rather, we must determine whether the court’s exercise of discretion, considering the law and the facts, was arbitrary or capricious.”).
This Court has recently stated that “[a]t its immovable core, the abuse of discretion standard requires a reviewing court to show enough deference to a primary decision-maker’s judgment that the court does not reverse merely because it would have come to a different result in the first instance.” Evans v. Eaton Corp. Long Term Disability Plan, 514 F.3d 315, 322 (4th Cir.2008). Based on the record in this case, a reasonable jurist could plainly find the appellants’ direct and statistical evidence failed to show commonality and typicality for the purported class sufficient to meet the certification threshold. As the district court’s judgment was neither based on a “clearly erroneous factual finding” nor “erroneous legal principles,” it should be accorded the deference due under the abuse of discretion standard of review.
In my view, the majority opinion fails to accord the district court proper deference in this case and, instead, reaches factual conclusions de novo that are not within the proper scope of review by an appellate court. The majority’s opinion could reasonably be interpreted to require class certification so long as future plaintiffs seeking class certification can produce any data set exceeding two standard deviations without regard to the unreliability of that data or the process by which it was derived. Equally troubling is the prospect that certification is required no matter how attenuated and insufficient a plaintiffs direct evidence is shown to be. Such an extension of our case law ignores the prudent judgment and proper discretion of a district court, which has seen and heard the prolific evidence first-hand, and constrains the district court to a merely ministerial act devoid of evidentiary grounding. That result is incongruent with an appellate court’s adherence to the standard of review for an abuse of discretion. I would thus affirm the district court’s judgment denying class certification on the appellants’ disparate impact and disparate treatment claims for the reasons set forth above, and respectfully dissent from the majority opinion in that regard.
. To the extent the district court's judgment was based upon a determination with regard to Sections 111(B)(2) and (C) that the appellants' hostile work environment claim did not have common questions of law or fact or was not typical of the class because the evidence showed separate, unconnected work environments, that determination is not supported by the record. While Nucor's evidence shows its operation of the plant departments was distinct and separated among the departments, that evidence is not probative as to the effects of the alleged acts on the employees’ working environment across the whole facility. However, should Nucor adduce relevant evidence on the merits showing distinct and unconnected work environments as affects the appellants and other employees, the district court may take such further action as Rule 23(c)(1)(C) permits.
. I also concur in the holdings represented by footnote 4, that the district court did not abuse its discretion in excluding post-2003 promotion data, and footnote 8, that Rule 23(f) did not permit interlocutory appeal of the discovery order.
. If the purported class was only beam mill employees, it may well have been an abuse of discretion not to certify the promotion claims (ignoring the numerosily factor). However, that was not the proposed class. The purported class in this case is all the plant's African-American employees and, for the reasons stated, the direct evidence is simply insufficient to find an abuse of discretion by the district court.
. The appellants’ case is solely concerned with Nucor’s actions on promotions within the South Carolina plant's internal workforce. Hiring or termination of employment is not at issue. Moreover, it is uncontested that promotions at the Nucor plant are only from within the existing workforce and the promotions at issue in this case are those subject to bid by employees.
. The district court’s opinion states that suit was originally filed on August 25, 2004. However, court records indicate that suit was originally filed against Nucor Steel on a company-wide basis in the Western District of Arkansas on December 8, 2003. That portion of the suit pertaining to discrimination claims at the Nucor-Berkeley plant at issue here was transferred to the District of South Carolina on August 25, 2004. J.A. 9007. The appellants then filed a Third Amended Complaint on January 28, 2005. J.A. 43.
. Our case law makes clear that standard deviation analysis is the proper method of statistical analysis. See EEOC v. Federal Reserve Bank of Richmond, 698 F.2d 633, 647 (4th Cir.1983), rev'd on other grounds, sub nom. Cooper v. Fed. Reserve Bank of Richmond, 467 U.S. 867, 104 S.Ct. 2794, 81 L.Ed.2d 718 (1984) ("[W]e have adopted the rule that the proper method for determining 'legal significance’ on the basis of statistical evidence is through the use of the standard deviation analysis ....”) (citing Moultrie v. Martin, 690 F.2d 1078 (4th Cir.1982)). The - I. 48 standard deviation resulted from data that controlled for two factors, the employees’ training and discipline. When those two factors were not controlled, the standard deviation was calculated at - 1.53, which the appellants’ expert testified was not a statistically significant difference. J.A. 5857.
. The actual data from 2001 to 2003 came from "the posting notices of the jobs, the bids that were submitted by the individual employees, the prevailing bidder, and the personnel files and everybody that bid on those jobs.” J. A. 9024.
. The majority opinion recites the additional time period for statistical information as "1999-2000.” The actual period was December 9, 1999 to January 4, 2001, not quite 13 months. J.A. 1162 n. 16-17. I refer to that December 9, 1999 to January 4, 2001 time period as simply "2000” to more accurately reflect the single year for which the appellants sought to create a statistical data substitute.
. As explained in more detail, infra, the "2000 data” consisted of 27 "change-of-status” forms for employment actions taken by Nucor from December 1999 to January 2001 and the appellants’ resulting projection of the percentages of African-American employees who applied for the employment positions represented by the change-of-status forms during this timeframe and were not promoted.
. In addition to Nucor’s assertion that the twenty-seven change-of-status forms may not have represented a "promotion,” Nucor also asserts that even if each change-of-status form represented a promotion, there is no way to know if the job was posted for bidding.
. The appellants proffered no evidence, from themselves or any other Nucor employee (past or present) that any African-American employee bid on any of the positions ostensibly represented by the 2000 data.
. Accordingly, Dr. Welch limited his analysis to "all of the job postings containing applicant information” because “[p]rior to 2001 we know neither the applicant mix regarding race nor do we know anything regarding qualifications.” J.A. 5893.
. The district court's observations that the appellants’ “assumptions may be reasonable and the statistics based thereon may be relevant” means no more than it says. J.A. 8985. The district court, in context, was merely acknowledging the theoretical possibility that the appellants could present relevant evidence — a proposition disproved by the evidence they did present.
. Insomuch as neither the appellants' direct evidence or statistical evidence met the Rule 23 burden of proof, it was not an abuse of discretion to determine the two deficient forms of evidence could not, taken together, meet the appellants' burden of proof.
. The majority opinion seems to imply some bad purpose on the part of Nucor because the 2000 data on actual bid promotions was unavailable. Appellants have never made a spoliation of evidence claim or pointed to any legal duty on the part of Nucor to have maintained the precise records found absent. While appellants were certainly entitled to seek to create substitute data for 2000, nothing entitled them to a data substitute based on insufficient evidence and invalid assumptions.