OPINION
GREGORY, Circuit Judge:This case involves allegations of racial discrimination at a steel manufacturing plant in Huger, South Carolina, owned by Nucor Corporation and Nucor Steel Berkeley (collectively, “Nucor”). We find that the district court abused its discretion and erred as a matter of law in denying class certification to the plaintiffs-appellants. We therefore vacate the order and remand the case to the district court for certification.
I.
The allegations that the appellants present in support of their racial discrimination and hostile work environment claims speak for themselves: white supervisors and employees frequently referred to black employees as “nigger,” “bologna lips,” “yard ape,” and “porch monkey.” White employees frequently referred to the black employees as “DAN,” which stood for “dumb ass nigger.” These racial epithets were broadcast over the plant-wide radio system, along with “Dixie” and “High Cotton.” Monkey noises were also broadcast over the radio system in response to the communications of black employees. The display of the Confederate flag was pervasive throughout the plant, and items containing Nucor’s logo alongside the Confederate flag were sold in the plant’s gift shop. Additionally, several e-mails that depicted black people in racially offensive ways, such as by showing them with nooses around them necks, were circulated by various employees. Once, an employee held up a noose and told a black co-worker that it was for him.
The plant is organized into six production departments: beam mill, hot mill, cold mill, melting, maintenance, and shipping. When a job opening at the plant becomes available, the position is advertised over a plant-wide posting and bidding system controlled by the central personnel department. Employees are allowed to bid on positions in any department. Although, by policy, the plant’s general manager approves all promotions and handles discrimination and harassment investigations, the record suggests that each department manager has unbridled discretion to make promotions within his department utilizing whatever objective or subjective factors he wishes. There were no black supervisors until after the institution of the Equal Employment Opportunity Commission charges that preceded this litigation. Indeed, a white supervisor testified that his department manager — who wore a Confederate flag emblem on his hardhat — told him that he would never promote a black employee to supervisor. (J.A. 1066, 1885— 86.)
The present litigation arose on August 25, 2004, when seven black employees at the plant, along with employee-plaintiffs at plants owned by Nucor in other states, brought suit under 42 U.S.C. § 1981 (2000) and Title VII of the Civil Rights Act of 1964 in the U.S. District Court for the Western District of Arkansas on behalf of themselves and approximately one-hundred other past and present black employees at the plant. At the time the litigation commenced, there were 611 employees working at Nucor’s South Carolina plant, of whom seventy-one were black. The Western District of Arkansas severed the case and transferred the claims brought by the seven plaintiffs in South Carolina to the District of South Carolina. The appellants seek a permanent injunction, back pay, compensatory and punitive damages, and attorney’s fees.
*152On May 7, 2007, the appellants filed a motion for class certification alleging:
(1) A pattern or practice of disparate treatment against African-American employees with respect to promotion opportunities at the plant; (2) Nucor’s promotion procedure, which allows white managers and supervisors to use subjective criteria to promote employees, has a disparate impact on African-American employees who apply for promotions, and (3) Nucor requires African-American employees to work in a plant-wide hostile work environment.
(J.A. 8980.) The district court denied class certification, and the would-be class plaintiffs now appeal.
II.
We review the district court’s certification decision for abuse of discretion. Doe v. Chao, 306 F.3d 170, 183 (4th Cir.2002), aff'd on other grounds, 540 U.S. 614, 124 S.Ct. 1204, 157 L.Ed.2d 1122 (2004). “[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.” Gen. Tel. Co. of the Sw. v. Falcon, 457 U.S. 147, 161, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982).
III.
Rule 23(a) of the Federal Rules of Civil Procedure provides the following:
One or more members of a class may sue or be sued as representative parties on behalf of all members only if:
(1) the class is so numerous that joinder of all members is impracticable,
(2) there are questions of law or fact common to the class,
(3) the claims or defenses of the representative parties are typical of the claims or defenses of the class, and
(4) the representative parties will fairly and adequately protect the interests of the class.
In our review of these factors, we bear in mind that
the final three requirements of Rule 23(a) “tend to merge,” with commonality and typicality “serving as guideposts for determining whether ... maintenance of a class action is economical and whether the named plaintiffs claim and the class claims are so interrelated that the interests of the class members will be fairly and adequately protected in their absence.”
Broussard v. Meineke Discount Muffler Shops, 155 F.3d 331, 337 (4th Cir.1998) (quoting Falcon, 457 U.S. at 157 n. 13, 102 S.Ct. 2364) (alteration in original). Indeed, “[certification is only concerned with the commonality (not the apparent merit) of the claims and the existence of a sufficiently numerous group of persons who may assert those claims.” Lilly v. Harris-Teeter Supermarket, 720 F.2d 326, 332-33 (4th Cir.1983).
The district court below considered each of the Rule 23(a) requirements in turn, and we shall do the same.
A.
First, regarding numerosity, the district court found that the appellants satisfied the requirement because there were ninety-four black employees who worked at the plant from 2001 through 2004. Nucor argued that because only nineteen of these employees bid on positions from 2001 to 2006, the numerosity requirement had not been met. However, the district court held that the other black employees should be counted in the class because “potential applicants are eligible to prove that they would have applied for a promotion but for the discriminatory practice.” (J.A. 8994.) *153Appellees do not challenge this finding, and we therefore presume it to be correct.
B.
Second, on commonality, the district court ruled against the appellants. The court reasoned that subjectivity in decision-making alone was insufficient to establish a disparate impact claim, discredited the statistical evidence presented in support of the disparate impact claim, and rejected the hostile work environment claim. We find that the district court improperly discounted the appellants’ direct evidence, which, alone, was sufficient to establish commonality. Moreover, the district court improperly excluded the appellants’ alternative benchmark for missing employment data and therefore erred in finding their statistical calculations to be insufficient to establish commonality. Finally, the district court erred in finding an insufficient basis for commonality on the hostile work environment claim.
1.
This Court has noted before that allegations of “a practice of disparate treatment in the exercise of unbridled discretion ... rais[es] questions of law and fact common to all [subject] black employees.” Lilly, 720 F.2d at 333. The Fifth Circuit reasoned similarly in Shipes v. Trinity Industries, 987 F.2d 311, 316 (5th Cir.1993):
The threshold requirements of commonality and typicality are not high; Rule 23(a) requires only that resolution of the common questions affect all or a substantial number of the class members. Allegations of similar discriminatory employment practices, such as the use of entirely subjective personnel processes that operate to discriminate, satisfy the commonality and typicality requirements of Rule 23(a).
(internal citation omitted). In Lilly, we further concluded that statistical evidence is unnecessary to establish commonality:
[Pjlaintiffs need not, at the time of the motion for class certification, demonstrate by statistical evidence that blacks have been terminated [or promoted] at a higher rate than have whites, or any other differential. Certification is only concerned with the commonality (not the apparent merit) of the claims and the existence of a sufficiently numerous group of persons who may assert those claims....
Lilly, 720 F.2d at 332-33.
The district court determined that the appellants’ statistics were insufficient to establish commonality. Yet, the appellants have certainly presented compelling direct evidence of discrimination, such as denials of promotions when more junior white employees were granted promotions (J.A. 1004, 1017), denial of the ability to cross-train during regular shifts like their white counterparts (J.A. 1000, 1023), and a statement by a white supervisor that he would never promote a black employee (J.A. 1885-86). This evidence alone establishes common claims of discrimination worthy of class certification. See Stastny v. S. Bell Tel. & Tel. Co., 628 F.2d 267, 278 (4th Cir.1980) (noting that class certification for a pattern-or-practice claim might be based on inferential, statistical, or direct evidence of discrimination). That the appellants also presented statistical data only strengthens their case.
The appellants have presented five statistical calculations in support of their discrimination claims,1 but we need focus *154on only one. In order to demonstrate a disparity in job promotions, the appellants compared 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. The former figure was 19.24%, and the latter figure was 7.94%. However, the appellants could not rely entirely on actual applicant data in making this calculation because Nucor has destroyed such data for promotion periods prior to 2001.2 Instead, in an effort to establish an alternative benchmark, the appellants, using change-of-status forms, identified twenty-seven positions filled during the 1999-2000 time period and extrapolated from that data using the assumption that the racial composition of the bidding pool for those jobs was the same as the weighted average of the racial composition of the bidding pools for the subsequent, post-2000 period.
Nucor provided selected promotions data for 2001 through 2006, and it argues that this data should have been used instead of the extrapolated 1999-2000 data because the appellants gathered the extrapolated data from the change of status forms, which do not indicate whether a position was open for bidding. The district court decided to exclude the appellants’ pre-2001 data and Nucor’s post-2003 data and rely solely on the data that Nucor provided for January 2001 through December 2003.3 Although the appellants argued that the 80% rule was the proper standard by which to evaluate their statistical evidence,4 the district court determined that standard deviation analysis was the proper *155method by which to do so.5 In its subsequent assessment of the statistics, the district court found that the appellants had not proven a statistically significant disparity, and it therefore ruled against them on the commonality prong of their claims. The appellants argue that the district court’s decisions were improper and negatively affected their statistics.
In another Title VII suit involving destroyed data, United States v. County of Fairfax, 629 F.2d 932 (4th Cir.1980), Fairfax County destroyed its job applications for the 1974-77 period. “As a consequence, the government extrapolated the figures for 1974-77 from the 1978 applicant pool by assuming that the applicant pool for 1974-77 contained the same proportions of blacks and women as did the 1978 applicant pool.” Id. at 940. The Court approved the use of this alternative benchmark and concluded that it was “the most salient proof of the County’s labor market.” Idf In this case, since Nucor destroyed the pre-2001 job promotions data, the appellants likewise were free to attempt to utilize an alternative benchmark in order to form their calculations, and the district court abused its discretion in ruling that only the data that Nucor provided could be used.6 7 The appellants should not be penalized because Nucor destroyed the actual pre-2001 job promotions data.8
*156The question before. the district court was not whether the appellants have definitively proven disparate treatment and a disparate impact; rather, the question was whether the basis of appellants’ discrimination claims was sufficient to support class certification. See Lilly, 720 F.2d at 332. In excluding the appellants’ data, the district court ruled, “These assumptions [regarding the 1999-2000 data] may be reasonable and the statistics based thereon may be relevant to prove discrimination at the plant. However, the necessity of the assumptions diminishes their probative value.” (J.A. 8985.) Yet, evidence need not be conclusive to be probative, and even evidence that is of relatively weak probative value may be useful in meeting the commonality requirement.
With the 1999-2000 data included, the record indicates that the appellants’ statistics would be significant at greater than two standard deviations.9 The appellants have therefore presented valid statistical evidence that independently indicates a disparate impact and disparate treatment in job promotions at Nucor, and we reiterate that an in-depth assessment of the merits of appellants’ claims at this stage would be improper. See Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 177, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974); Thorn v. Jefferson-Pilot Life Ins. Co., 445 F.3d 311, 319 (4th Cir.2006). We therefore find that the appellants’ calculations based on their alternative benchmark were adequate to establish commonality.10
In summary, because the appellants’ direct evidence alone was sufficient to demonstrate common claims of disparate treatment and disparate impact, their statistical *157data did not need to meet a two-standard-deviation threshold, and the district court erred as a matter of law in requiring them to do so. Yet, we further find that the district court abused its discretion when it excluded the appellants’ alternative calculations of the destroyed pre-2001 promotions data. With this data included, the appellants’ statistics were independently sufficient to meet the Rule 23 commonality requirement. We therefore conclude that the appellants satisfied the commonality requirement for their discrimination claims, and the district court should have found in the appellants’ favor on this portion of their motion. See Caridad, v. Metro-North Commuter R.R., 191 F.3d 283, 293 (2d Cir.1999) (“More detailed statistics might be required to sustain the Plaintiffs’ burden of persuasion, but [these statistics], in conjunction with the anecdotal evidence, satisf[y] the Class Plaintiffs’ burden of demonstrating commonality for purposes of class certification.” (internal citation omitted)).
2.
The district court also discounted the sixteen affidavits that the appellants presented in support of their pattern-or-praetice and hostile-work-environment claims because only two of the affiants were in departments other than the beam mill department. The district court found that this concentration precluded the establishment of a pattern or practice of discrimination. It reached a similar conclusion on the appellants’ hostile work environment claim: “The plant’s production departments can be classified as separate ‘environments.’ A class members [sic] claim of a hostile work environment in the hot mill will vary significantly from a class member’s claim of a hostile work environment in the beam mill.” (J.A. 8989.)
The appellants challenge the district court’s ruling by arguing that the court’s “separate environments” analysis was flawed. The district court noted that “[t]he plaintiffs have presented plant-wide racist acts potentially experienced by every African-American employee working at the plant when the acts occurred. These acts include: (1) racist e-mails, (2) display of the confederate flag, and (3) racist remarks over the plant radio.” (J.A. 8988-89.) Yet, it then proceeded to classify the departments as unique environments.
The Supreme Court has held, “When the workplace is permeated with ‘discriminatory intimidation, ridicule, and insult,’ that is ‘sufficiently severe or pervasive to alter the conditions of the victim’s employment and create an abusive working environment,’ Title VII is violated.” Harris v. Forklift Sys., Inc., 510 U.S. 17, 21, 114 S.Ct. 367, 126 L.Ed.2d 295 (1993) (quoting Meritor Sav. Bank, FSB v. Vinson, 477 U.S. 57, 65, 67, 106 S.Ct. 2399, 91 L.Ed.2d 49 (1986)). This is essentially what the appellants argue: despite the formal classification of the plant into six production departments, the racist acts had plant-wide repercussions and affected all black employees. A hostile environment determination can be made “only by looking at all the circumstances.” Harris, 510 U.S. at 23, 114 S.Ct. 367. Such circumstances include “the frequency of the discriminatory conduct; its severity; whether it is physically threatening or humiliating, or a mere offensive utterance; and whether it unreasonably interferes with an employee’s work performance.” Id.; accord Nat’l R.R. Passenger Corp. v. Morgan, 536 U.S. 101, 116, 122 S.Ct. 2061, 153 L.Ed.2d 106 (2002). The allegations made in the affidavits that the district court discredited should have factored into *158the totality-of-the-circumstanees assessment of the plant as a whole.
In Hill v. Western Electric Co., 596 F.2d 99, 102 (4th Cir.1979), the Court held:
A person who has been injured by unlawful, discriminatory promotion practices in one department of a single facility may represent others who have been injured by the same discriminatory promotion practices in other departments of the same facility. In such a case, the representatives of the class all have the same interests in being free from job discrimination, and they have suffered injury in precisely the same way in the denial of promotion.
In Hill, the Court determined that employees in one department could not be included in the class because they did all of their work off-site. Id. That is not the case here.
The appellees argue that Nucor’s management provided “clear evidence” of “the decentralized nature of Nucor’s practices.” (Appellees’ Br. 18.) However, the “clear evidence” they cite is itself contradictory on this matter. While the manager of the beam mill indicated that his department is “separate from the buildings and areas of other Nucor Steel Berkeley departments” (J.A. 7885), the melt shop manager and supervisor noted that their building “is attached to Beam Mill and Hot Mill” (J.A. 7896, 7917). Moreover, “Nucor Berkeley employees share a locker room with caster, caster maintenance, hot mill, melt shop maintenance, hot mill maintenance, as well as environmental.” (J.A. 7896, 7917.) And as noted previously, racial slurs, monkey noises, and other offensive statements were broadcast over the plant-wide radio. Thus, there is scant, if any, evidence that each of the departments is so autonomous as to justify classifying them as separate environments.
In light of the foregoing cases and facts, it was an abuse of discretion for the district court to find that the employees at the plant were separated into different environments. All of the employees worked at a single facility, in departments that were, at minimum, connected to each other, and employees shared several common areas. There is therefore sufficient evidence to indicate that all of the black employees were affected by the comments and actions of the white employees and supervisors in other departments. Thus, the affidavits of employees in one department are admissible to prove a plant-wide hostile environment that affected employees in other departments, and the plaintiffs have satisfied the commonality requirement for their hostile work environment claim. See Holsey, 743 F.2d at 216-17.
C.
The district court determined that in order for the appellants to prove typicality for their disparate treatment claim, they would have to satisfy the framework set forth in McDonnell Douglas v. Green, 411 U.S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668. However, because the court found that the plaintiffs had not demonstrated a pattern or practice of discrimination, it held that the attempts by the individual plaintiffs to demonstrate disparate treatment would not be typical of the attempts of the other class members to do so. Moreover, the district court found that because the plaintiffs had not demonstrated a disparity in black promotion rates, their disparate impact claims also would not be typical of the class. Finally, the district court found that the named plaintiffs’ hostile work environment claims were not typical of the claims of class members outside of the beam mill.
*159As noted previously, the district court abused its discretion in its rulings on the appellants’ disparate treatment, disparate impact, and hostile work environment claims; therefore, it was also an abuse of discretion for the district court to base its typicality conclusion on these rulings. Moreover, as we noted supra, commonality and typicality “tend to merge,” Broussard, 155 F.3d at 337, and the appellants have presented sufficient evidence to satisfy the threshold typicality requirement of Rule 23(a).
D.
Finally, regarding adequacy, the district court found the plaintiffs not to be adequate representatives for the disparate treatment and disparate impact claims. However, the court could “discern no conflicts of interest among the plaintiffs and class members regarding hostile work environment claims.” (J.A. 8993.) Given the above conclusions that we have reached regarding the appellants’ disparate treatment and disparate impact claims, we also find that the district comb’s assessment of the adequacy factor with regard to these claims was an abuse of discretion.
To the extent that the district court was correct that the putative class representatives have a conflict with the class in terms of competition for promotions, this conflict should not defeat class certification. Indeed, if this were true, how might a class action challenging promotion practices ever be brought — unless the EEOC deems fit to do so — when the plaintiffs seek instatement into previously denied positions? The appellees point to General Telephone Co. of the Northwest v. EEOC, 446 U.S. 318, 100 S.Ct. 1698, 64 L.Ed.2d 319 (1980), but that case is to no avail. There, the Supreme Court simply “note[d]” as an “example” that “[i]n employment discrimination litigation, conflicts might arise, for example, between employees and applicants who were denied employment and who will, if granted relief, compete with employees for fringe benefits or seniority. Under Rule 23, the same plaintiff could not represent these classes.” Id. at 331, 100 S.Ct. 1698. Setting aside the question of whether this language was meant to be controlling, it plainly does not apply to this case, which deals with promotions and not the competition for “fringe benefits or seniority” posed by new hires.
The district court can address plaintiffs’ claims for injunctive or other relief after liability and other common issues are determined. See Fed.R.Civ.P. 23(c)(5); Int’l Bhd. of Teamsters v. United States, 431 U.S. 324, 361-62, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977) (“[A]s is typical of Title VII pattern-or-practice suits, the question of individual relief does not arise until it has been proved that the employer has followed an employment policy of unlawful discrimination. The force of that proof does not dissipate at the remedial stage of the trial.”); Hill, 672 F.2d at 387 (“Bifurcation of Title VII class action proceedings for hearings on liability and damages is now commonplace.... ”). If, at the second stage of the proceeding, conflicts need to be resolved with regard to promotions, the district court can do so then. See Gunnells v. Healthplan Servs., Inc., 348 F.3d 417, 429 (4th Cir.2003) (“[T]he need for individualized proof of damages alone will not defeat class certification.”); Pettway v. Am. Cast Iron Pipe Co., 576 F.2d 1157, 1178 (5th Cir.1978). Of course, certification is conditional, and should the conflicts prove intractable at the second stage, the district court may simply decertify the *160class. See In re Sch. Asbestos Litig., 789 F.2d 996, 1011 (3d Cir.1986). But at this stage of the proceedings, we find that the appellants are adequate representatives for the disparate impact and treatment claims of the putative class.
As noted previously, the district court found the appellants to be adequate representatives for the hostile work environment claim. The appellees do not dispute this finding, and we find no abuse of discretion in the district court’s so concluding.
IV.
Our review of the district court’s assessment of the Rule 23(a) factors leads us to conclude that the court abused its discretion in denying class certification. Given our consideration above, we find that the class certification requirements of Rule 23(a) have been satisfied with regard to the appellants’ disparate impact, disparate treatment, and hostile work environment claims. We therefore vacate the district court’s denial of the appellants’ motion for class certification, and we remand the case to the district court with instructions to certify the appellants’ class action and to engage in further proceedings consistent with this opinion.
VACATED AND REMANDED WITH INSTRUCTIONS.
. First, the appellants presented a calculation of the disparity in utilization rates at Nucor by comparing the average percentage of blacks in skill positions at the plant from 2000 *154to 2004 with the percentage of blacks in the surrounding population in 2000 who were qualified for the skill positions at the plant. In the surrounding area, blacks made up 38.2% of the population qualified to hold skilled positions. However, only 13.42% of the employees at the plant from 2000 through 2004 were black. The second calculation was the same as the first, except the appellants extended the time period to 2006 to demonstrate that the disparity continued.
To demonstrate a disparity in the number of job applications by black applicants, the third calculation compared the percentage of qualified blacks in the surrounding population in 2000 with the percentage of blacks who bid on plant promotions between 2001 and December 2003. The former percentage was 38.2, and the latter was 16. The fourth calculation is discussed infra. Finally, the fifth calculation compared the percentage of qualified blacks in the surrounding community in 2000 with the estimated percentage of blacks who received promotions from December 1999 to December 2003. The former figure was 38.2%, and the latter figure was 7.9%.
. We make no judgment regarding the propriety of Nucor's destruction of the data.
. Nucor complains that the appellants themselves have created an unduly restricted data set by excluding the promotions that occurred after the filing of their suit in 2004. However, post-lawsuit promotions would not enhance the accuracy of the data on Nucor’s promotion practices. Indeed, such promotions would likely detract from their accuracy, since it is no secret that the institution of litigation often prompts companies to change their practices. These post-suit changes have only minimal weight with regard to the merits of the plaintiffs’ claims. See Holsey v. Armour & Co., 743 F.2d 199, 214 n. 5 (4th Cir.1984) ("We find no error in the fact that the district court minimized the significance of evidence of Armour's post-complaint hiring and promotion of black employees.”). The district court found in the appellants' favor on this issue, which is why it did not utilize Nucor’s post-2003 data, and we find no abuse of discretion in its doing so.
. The 80%, or four-fifths, rule "is an administrative rule of thumb used by agencies concerned with Title VII cases. It offers a definition of what is a serious difference in the passing rates for protected classes. If the selection rate for a protected class is less than 80% of the selection rate for the group selected at the highest rate, that constitutes adverse impact.” Chisholm v. U.S. Postal Serv., 665 F.2d 482, 495 n. 22 (4th Cir.1981).
. Standard deviation analysis considers the differences between expected and observed values — in this case, for example, the number of blacks promoted versus the number of blacks one would expect to be promoted based on their percentage in the relevant population (e.g., job applicants or the plant workforce). Standard deviation analysis is often viewed as superior to the 80% rule because it takes into account the natural fluctuations from the expected value that come with any random sample. In other words, for any given sample of black applicant data, an observed disparity might be the result of chance. The smaller the standard deviation, the more closely the data points are clustered around the mean (expected value), and the more likely the data points are the result of chance. "For example, if a coin were tossed ten times ... and came up heads four times, no one would think the coin was biased (0.632 standard deviations), but if this same ratio occurred for a total of 10,000 tosses, of which 4,000 were heads, the result could not be attributed to chance (20 standard deviations).” Lilly, 720 F.2d at 336 n. 17.
. The dissent does not argue with the underlying reasoning of County of Fairfax: when an employer destroys relevant employment data, the plaintiffs may utilize alternative benchmarks to make up for this lost data. Certainly, the benchmarks will not be as good as the destroyed data themselves — that would be next to impossible to achieve. Nevertheless, the plaintiffs should not be penalized by the destruction (however innocent) of such data. Any factual differences between the cases do not overcome this underlying reasoning:
First, the fact that there is only one year of missing data in this case is inapposite — the point is that a year of relevant data has been destroyed, and that destruction should not be used to weaken the appellants' case. Likewise, the absence of a statutory duty upon Nucor to maintain any data does not imply a judicial bar from allowing the appellants to approximate the destroyed data in making their claim. Finally, we certainly do not agree with the dissent's insinuation that, because this case involves promotions and not hiring, the appellants are less entitled to data (and approximations thereof) to support their claims. The dissent is correct, and should itself remember, that we are not reviewing the merits of this case, and the district court was simply incorrect to exclude all evidence of the 1999-2000 promotions.
. The district court made the following determination in excluding the appellants' pre2001 data: "Nucor provided the plaintiffs with records of promotions occurring in the plant from January 2001 to December 2003. Statistics based on actual data is [sic] more probative than statistics based on assumptions. Consequently, the Court relies on statistics resulting from an analysis of this data.” (J.A. 8985.)
. In response to appellants’ argument that the district court should have compelled Nucor to *156provide more data during discovery, Nucor argues that the appellants cannot appeal a discovery order under FRCP 23(f). We agree with the Eleventh Circuit that "jurisdiction granted by Rule 23(f) does not extend to [a] separate [discovery] order.” DeLeon-Granados v. Eller & Sons Trees, Inc., 497 F.3d 1214, 1218 n. 1 (11th Cir.2007). We therefore express no opinion at this time regarding the district court's decision not to compel Nucor to produce additional data.
. The Supreme Court has indicated that anything greater than two or three standard deviations in racial discrimination cases is suspicious. Hazelwood School Dist. v. United States, 433 U.S. 299, 308 n. 14, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977); Castaneda v. Partida, 430 U.S. 482, 496 n. 17, 97 S.Ct. 1272, 51 L.Ed.2d 498 (1977). We will presume that two standard deviations is the proper threshold for this case, but we reserve the unanswered question of whether this rule should be limited to large sample sizes. With the 1999-2000 data excluded, the appellants' promotion data demonstrated a disparity resulting in only 1.48 standard deviations — below the Supreme Court’s threshold. With the data included, however, the appellants indicate that their calculations yielded 2.54 standard deviations. For our purposes, a threshold of two standard deviations corresponds roughly to a 95% confidence level or a .05 level of significance, i.e., there is only a 5% probability that the result is due to chance. Three standard deviations would equate to a 99.7% confidence level. See generally Bruce J. Chalmer, Understanding Statistics 97-98 (1987); Wikipedia, Standard deviation, http://en.wikipedia.org/ wiki/Standard_deviation (last visited June 19, 2009).
. The dissent takes issue with our crediting the appellants' statistics. However, we emphasize that at this stage, we are dealing only with whether the appellants have presented sufficient statistical information to establish commonality. We must walk a fine line between a facial class certification assessment and an assessment on the merits, and the dissent has stepped to the other side. The dissent’s critiques might very well discredit the appellants' statistics later, upon a full review of the merits, but the information that the appellants have presented is enough to allow them to get to that point. Our crediting the statistics certainly does not portend the opening of any class certification floodgates. There are many requirements for class certification, and our decision today respects that fact.