COHN, D.J., delivered the. opinion of the court, in which, COLE, J., joined. ROGERS, J. (pp. 402-03), delivered a separate dissenting opinion.
OPINION
COHN, District Judge.This is a race discrimination case under Title VII. Plaintiffs claim that facially neutral criteria used to determine promotions ' had the effect of discriminating against African-American employees in the Department of Defense’s Defense Finance and Accounting Service, Columbus Center (“DFAS-CO”), in violation of 42 U.S.C. § 2000e-16. Plaintiffs appeal the decision of the magistrate judge granting summary judgment to the Secretary and finding that five of the named plaintiffs, Curtissene Anderson, Richard Goudy, Denise Goudy, Linda Pettes, and Jean Washington lacked standing. Plaintiffs also raise an issue regarding the .. magistrate judge’s treatment of its motion for. sanctions. We agree with the magistrate judge’s decision regarding standing with the exception of one plaintiff, Denise Goudy. We,disagree with the magistrate judge’s decision to grant summary judgment in favor of the Secretary, finding sufficient evidence . in the record of a disparate impact. We also find that the magistrate judge erred in its handling of plaintiffs motion for sanctions. Accordingly, for the reasons that follow, the decision of the magistrate judge is AFFIRMED IN PART, REVERSED IN PART, and REMANDED for proceedings consistent with this opinion.
I. BACKGROUND
A. The DFAS-CO
The DFAS-CO is an independent agency within the Department of Defense, cre7 ated in 1991. Its chief function is paying the Department’s contractor and vendor bills, but it also provides accounting services for smaller Department agencies. The new office expanded: rapidly, from four to five hundred employees at its opening, to about 3,500 employees in 1995. Mahlon Boyer, Director of Human Resources at DFAS-CO, estimated that about three thousand employees were hired at the office between 1991 and 1995.
Charles Coffee (“Coffee”), the Director of DFAS-CO, describes the promotion process in the agency, as it existed in the mid-1990s, as follows. First, the manager with an opening in the department decides whether the position should be filled externally, or through a temporary or permanent promotion. Next, the manager works with a human resources specialist to develop a “recruiting plan” that details the qualifications needed for the position. If the manager chooses a permanent competitive promotion as the best option, an internal posting of the job vacancy takes place. Employees then submit applications. A human resources employee ranks’ the ap*391plications numerically, using values assigned for experience, education, awards, disciplinary actions, and performance reviews. This information is provided to human resources by the employee’s supervisor. Human resources issues a document called a promotion certificate referring the top fifteen ranked candidates to the hiring manager. The agency’s equal employment opportunity office reviews the certificates to determine whether any of the finalists are members of underrepresented groups, and puts this information on the certificate. (Underrepresentation is determined by looking at the number of members of that group in the position’s grade level.) The hiring manager, using the certificates, ranks the finalists and determines which ones she wishes to interview. The ranking is based on qualifications detailed in the recruiting plan. After choosing the successful candidate, the manager must use a “selection matrix”- to document the rationale behind the choice. The manager must give an explanation for not choosing a candidate from an underrepresented group, if such candidates are included in the finalist pool. The manager has the option of rejecting all fifteen finalists and requesting that the position be filled externally. The manager then submits her recommendation to human resources and to the agency director for approval. Plaintiffs contend that the promotion process consists of more than eighteen different steps. To be sure, the process is complex and allows for a fair amount of subjective decision making by a particular manager.
B. Concerns Over Minority Advancement at DFAS-CO
1. The Newsletter Article
The DFAS-CO publishes a weekly newsletter. In the October 20-26, 1996 edition of the newsletter, a short article appeared addressing concerns employees had voiced regarding why so few African-American employees were being hired to regular GS levels. The article was organized in a question-and-answer format. One question was, “Why so few black, Asian or Hispanic males in management above the research level?” The answer acknowledged that Grades in GS/GM 13-15, the top management grades, black males (numbering 5) represented 4.2% of positions, whereas statistical data suggested they should comprise 6.5%. The answer continued, “The Center faces the challenge of planning for increased representation of minority groups in view of workyear projections. Ideas already under consideration include rotational assignments, cross-training and use of special projects to gain skills and exposure.” One employee, plaintiff Willie Phillips, wrote a letter to managers, complaining of the DFAS-CO’s “racist policies” and also alleging that the article suggested minorities needed a special boost to obtain promotions, when in fact they were being unfairly denied promotions to which their skills and experience entitled them. Coffee responded to Phillips’ letter affirming his commitment to “achievement of a better work force balance.” ■
2. The PAT Report
On December 6, 1996, Coffee announced the establishment of a Process Action Team (“PAT”) to address minority employees’ concerns regarding advancement within the agency. The team was comprised of both DFAS-CO employees and outside consultants. The PAT Report drew from two sources of information in its analysis: it' conducted interviews with African-American employees to examine impressions of the promotion system; and it analyzed EEO statistical data for 1995 and *3921996 concerning merit promotions, performance ratings, cash awards for excellence at work, and disciplinary actions. The team drew some conclusions that validated the employees’ concerns. The report stated that “[mjinorities continue to be promoted at' lower rates than non-minorities, especially at higher grade levels.” It also noted a strong tendency of upper-level selecting officials to choose employees from their own ethnic group for promotions. In considering the distribution of awards, the report found that minorities groups, combined, received 23% of employee awards, while white employees received 77% of awards. In the area of discipline, black employees received 49% of all disciplinary actions for the two years. The report noted that non-minorities “continue to receive the highest percentage of Exceptional and High ratings,” and that, “[a]s a mandatory factor in the Merit Promotion process, these trends reflect unfavorably for minorities.” The report’s recommendations included ensuring even distribution of “detailed” positions (temporary promotions and transfers) among employees, enforcing a zero-tolerance policy on “prohibited personnel practices,” reviewing award policies to ensure consistent application, clarifying selection procedures, and disseminating to employees complaint procedures regarding promotion selections.
C. EEOC Complaint, Investigation, and Decision
On July 30, 1997, Willie Phillips, on behalf of a group of African-American employees at DFAS-CO, filed a discrimination complaint with the DFAS-CO’s equal employment opportunity office. The complaint alleged that the following practices were causing a disproportionately low number of African-Americans to receive promotions: (1) the temporary assignment (“detailing”) of white employees into higher-grade positions, which placed them at an advantage in consideration for permanent promotions; (2) ethnic preference for white candidates among white officials making promotion selections; (3) discretion vested in selecting officials to consider “specialized experience” in promotion decisions; (4) manipulation of performance reviews, employee awards, and the timing of job announcements to give preference to white candidates in the promotion process; and (5) granting promotions to white employees who did not meet the time and grade requirements for the position.
The following is also in the record, although the parties do not reference it on appeal: An EEO officer prepared a report, drawing on both agency data and the PAT Report. Phillips attached the report as an exhibit accompanying the complaint. The report found that African-Americans received about 22% of the competitive and non-competitive promotions agency-wide in 1995 and 1996, and comprised approximately the same percentage of the DFAS-CO workforce during that time period. Thus, the bottom line appeared neutral. The report suggested disparities between African-American and' white employees, however, with respect to awards and discipline. Grouping together the categories of awards recognized in the office in the fiscal year 1995, African-Americans, totaling 670 in DFAS-CO, received a total of 159 awards; thus, approximately 24% of African-Americans received awards. White employees, numbering 2,253 in the office, received 787 awards; 34.9% of whites received awards.1 In the disciplinary area, African-American employees in fiscal year *3931996 received 31, or 48.9% of the total disciplinary actions; whites received 30, or 46.87%. The number of African-American employee disciplinary violations comprised 4.5% of the African-American workforce in the office (31/684). The number of white employee violations comprised 1.33% of the white workforce in the office (30/2253).2 With respect to performance ratings, the report listed only African-American shares of the various ratings as a portion of the total, without comparative race data. African-Americans received 15.75% of exceptional performance ratings in fiscal year 1996, while comprising almost 22% of employees. The DFAS-CO EEO office referred the complaint to the EEOC.
On May 5, 1998, an administrative law judge issued a recommendation that the group’s class complaint be dismissed because their complaint was not timely filed and the “continuing violation” doctrine did not apply.
D. Federal Complaint and Initial Dismissal
On August 7, 1998, Phillips filed a complaint in the Southern District of Ohio against the defendant William S. Cohen, as Secretary of Defense. The complaint alleged that DFAS-CO maintains an unlawful discriminatory “glass ceiling” preventing the promotion of African-Americans. The complaint reiterated the five mechanisms disadvantaging minority promotion applicants set forth in the EEOC complaint: detailing; the use of white selecting officials; vesting too much discretion in hiring managers; the manipulation of awards, discipline, and performance reviews to promote more white than African-American employees; and promoting white candidates who do not meet the grade requirements for the job. Phillips sought class action status.
On July 30, 1999, the district court dismissed the complaint on the grounds that Phillips failed to file a timely administrative complaint. A panel of this Court reversed, finding that equitable tolling under the “continuing violation” doctrine applied. Phillips v. Cohen, 3 Fed.Appx. 212 (6th Cir.2001) (unpublished).
On remand, on May 2, 2002, the district court referred the case to a magistrate judge “for all further proceedings, including trial and entry of judgment.”
On October 15, 2002, Phillips, along with the other sixteen plaintiffs named above,3 filed an amended complaint naming current Secretary of Defense Donald Rumsfeld as a defendant. Plaintiffs withdrew the request for class action status and set forth allegations of discrimination under a disparate impact theory, as follows:
During the 1991-95 hiring expansion at DFAS-CO, many new hires were white women with no prior federal experience. By contrast, many of DFAS-CO’s African-American employees had transferred from other federal positions when the agency opened. A clique of white supervisors held social functions to which white employees were invited. A morale problem developed in the agency. The 1997 PAT Report showed evidence that promotion policies disparately impacted African-Americans. DFAS-CO has engaged in employment practices that treat white employees more favorably in promotion, discipline, awards, and performance appraisals.
*394E. Expert Opinions
Plaintiffs and the "Secretary each solicited the opinion of an' expert. Neither expert investigated the impact on African-Americans of the specific employment practices listed in the complaint — such as detailing, or the. use of performance reviews, awards, and disciplinary actions as a factor in promotions decisions — in the disparate impact analysis.
Plaintiffs’ expert, Ramona Paetzold (“Paetzold”), addressed two questions in her report: whether African-Americans are under-represented in the various job grades at DFAS-CO; and whether African-Americans are promoted at a lower rate than whites. To answer the first question, Paetzold relied on census data indicating that Columbus’s working population is ten percent African-American; she did not attempt to refine her benchmark representation rate in terms of work skills. She then, using a Department of Defense database of employee records, listed the percentage African-Americans comprised in each job grade at DFAS-CO as averaged over the 1992-96 time period. She concluded that African-Americans had a representation rate lower than whites’ rate only in Grade 14.4 Second, to assess whether African-Americans were promoted at a lower rate than whites, Paetzold used the Department database to add up the total number of promotion actions that took place during the entire 1992-96 time span.5 She then created a fraction in which the total number of promotions of African-Americans during that time period formed the numerator and the total number of African-American employees, counting only once employees who appeared in the records in multiple years, the denominator.6 1 She found a statistically significant difference between the “promotion rates” for African-Americans (42.23%) and for white employees (45.87%). The disparity was more marked when rates for African-American women (42.34%) and for white women (56.07%) were compared.7 Because of the unavailability of data on which employees applied for which promotions, Paetzold conducted “one ' large analysis that spanned the years”; she did not attempt to exclude non-race merit variables by determining how many African-American and how many white employees were qualified for each promotion granted.
The Secretary’s expert witness, John G. Claudy (“Claudy”), also addressed the two questions of whether African-Americans were under-represented at DFAS-CO and whether African-Americans were promoted at a lower rate. To determine representation, he compared the percentage of African-American employees in a grade level with the corresponding percentage of the African-American working population with skills corresponding to that grade level. He determined that there were *395more African-Americans in every grade level at DFAS-CO than one would expect. Addressing whether African-Americans had lower promotion rates than white employees, Claudy responded differently than Paetzold to the lack of data on which employees applied for which promotions. Whereas Paetzold ignored the merit aspects of each promotion, Claudy created “constructed pools of applicants.” Working with yearly “snapshots” of DFAS-CO’s employee data, he analyzed promotions in pairs of years; if an employee was listed at a higher grade in the second snapshot, then a promotion had taken place. For each such promotion, he created a pool of DFAS-CO employees in the same (pre-promotion) grade level and job series as the successful applicant. Using this method, he found there were greater than expected numbers of promotions of African-American employees in half of the fourteen grade levels, and fewer than expected in half. In only one grade level, 5, did a downward departure represent more than two standard deviations from the number that was expected.
F. The Secretary’s Motion for Summary Judgment
The Secretary moved for summary judgment on the grounds that Plaintiffs failed to establish a genuine issue of material fact as to disparate treatment in promotions for African-Americans and on the grounds that all of the plaintiffs lacked standing.
On August 4, 2003, the magistrate judge granted summary judgment to the Secretary. First, the magistrate judge found that four of the named plaintiffs-Curtis-sene Anderson, Richard Goudy, Linda Pettes, and Jean Washington-lacked standing because they did not show that they applied for and were denied promotions for which they were qualified. The magistrate judge further concluded that Denise Goudy lacked standing because she failed to state what promotions she was denied or that she was qualified for them. However, the magistrate judge found that the remaining plaintiffs-Willie Phillips, Yulan-der Edwards, Jackie Edwards, Jackie MeGrady, Susan McNeal, Mary Phillips, and Lerdón Woodfolk, Emma Fields, Kevin Gray, Carlissa Miles, Lida Prater, and Marica Spence-had standing.
Next, the magistrate judge concluded that plaintiffs had not made out a prima facie case of impact disparity. The magistrate judge found Claudy’s statistics to be more reliable than Paetzold’s, on the grounds that Paetzold’s failure to consider merit factors prevented her data from “eliminatfing] the most common nondiscriminatory reasons for the disparity.” The magistrate judge then concluded that Claudy’s finding of a statistically significant disparity in. promotion rates at just one grade level (5) was not significant enough to constitute a disparate impact. Further, the magistrate judge found that the PAT Report did not establish sufficient evidence of disparate impact to shift plaintiffs’ burden of proof to the Secretary.
G. The Lost/Destroyed Document Dispute and Sanctions
On December 15, 1998, the Secretary filed a motion for a protective order staying discovery pending the district court’s decision on the Secretary’s November 2, 1998 motion for summary judgment. The magistrate judge granted the protective order, “on the condition that defendant’s counsel locate' and gather all documents responsive to the discovery requests and produce documents responsive to it within eight business days of the Court issuing any decision denying the motion to dismiss/for summary. judgment.” Plaintiffs had made discovery requests encompass*396ing promotion records. Plaintiffs had also made discovery requests for all documents used in the PAT Report. Subsequently, the complaint was dismissed for failure to exhaust administrative remedies, a decision which as noted above this Court reversed. ■
On remand, it became clear in the course of discovery that DFAS-CO did not have possession of the documents used to create the PAT report, or of the files listing which employees applied for and received merit promotions during the 1992-96 period. The PAT documents were lost, DFAS-CO claimed, in the course of the agency’s interactions with a consulting firm that prepared' the PAT Report. The promotion information was destroyed after two years pursuant to an agency internal regulation.
Plaintiffs moved on April 24, 2003 for sanctions, arguing that the 1998 protective order required DFAS-CO to find and retain documents previously requested in discovery. The Secretary argued that the order to locate and keep the documents was contingent on the denial of summary judgment.
The magistrate judge granted plaintiffs’ motion for sanctions, but decided to defer until trial the decision as to the type of sanctions. On appeal, plaintiffs say that the magistrate judge abused his discretion in failing to impose sanctions when he granted the motion.
H. Plaintiffs’ Appeal
Plaintiffs timely appealed, challenging the magistrate judge’s decision on standing with respect to plaintiffs Curtissene Anderson, Richard Goudy, Denise Goudy, Linda Pettes, and Jean Washington, as well as the grant of summary judgment to the Secretary. Plaintiffs also appeal the magistrate judge’s failure to impose sanctions at the time it granted its motion for sanctions.
II. ANALYSIS
A. Standard of Review
We review de novo the district court’s grant of summary judgment. Andersoris, Inc. v. Consol, Inc., 348 F.3d 496 (6th Cir.2003). Summary judgment is proper “if the pleadings, depositions, answers to interrogatories, and admissions on file, together with affidavits, if any, show that there is no genuine issue of material fact and that the moving party is entitled to a judgment as a matter of law.” Fed.R.Civ.P. 56(c).
We review for abuse of discretion the district court’s decision on whether and how to sanction the Secretary for discovery violations. See Nat’l Hockey League v. Met. Hockey Club, 427 U.S. 639, 642-43, 96 S.Ct. 2778, 49 L.Ed.2d 747 (1976).
B. Standing
In order to establish constitutional standing to sue under Title VII, a plaintiff must show that he was “personally injured by the defendant’s alleged discrimination and that his injury will likely be redressed by the requested relief.” Melendez v. Illinois Bell Tel. Co., 79 F.3d 661, 668 (7th Cir.1996) (quoting Lujan v. Defenders of Wildlife, 504 U.S. 555, 560-1, 112 S.Ct. 2130, 119 L.Ed.2d 351 (1992)). In Title VII cases, the plaintiff must merely, at the summary judgment stage, present some proof (for example, an affidavit) demonstrating that he was injured by an alleged illegal practice. See Senter v. Gen. Motors Corp., 532 F.2d 511, 517 (6th Cir.1976) (stating standing requirement for Title VII); Lujan, 405 U.S. at 555, 92 S.Ct. 1113 (stating amount of proof of standing required at the summary -judgment stage).
*397The magistrate judge cited Melendez for the proposition that a plaintiff must establish that he was qualified for the position sought in order to have standing in a disparate impact case under Title VIL The magistrate judge found on this basis that five of the listed plaintiffs lacked standing: four because they failed to file affidavits asserting injury, and one (Denise Goudy) because she did not “specifically state what promotions she was denied or that she was qualified for them.” What Melendez actually requires, however, is that the plaintiff demonstrate either that he or she was denied an employment opportunity because of a practice prohibited by the statute; or that he or.she was qualified for the opportunity sought and was denied it and therefore, by inference, was subjected to discrimination. Melendez, 79 F.3d at 668.
The magistrate judge correctly found that the four plaintiffs who did not submit affidavits, Curtissene Anderson, Richard Goudy, Linda Pettes, and Jean Washington, did not have standing. Denise Goudy, on the other hand, met the requirements of Melendez and Senter: she stated that she was passed over for promotions, and that she observed less qualified, less experienced white employees in her department obtain positions she applied for, in violation of Title VII. Therefore, we reverse the magistrate judge’s holding on standing with respect to Denise Goudy, but affirm it with respect to Curtissene Anderson, Richard Goudy, Linda Pettes, and Jean Washington.
C. Evidence of Disparate Impact
1. Statutory Framework
Title VII makes it unlawful for an employer “to fail or refuse to hire or to discharge any individual, or otherwise to discriminate against any individual with respect to his - compensation, terms, conditions, or privileges of employment, because of such individual’s race, color, religion, sex, or national origin; or (2) to limit, segregate, or classify his employees or applicants for employment in any way which would deprive or tend to deprive any individual of employment opportunities ... [on account of one of the same five grounds listed above].” 42 U.S.C. § 2000e-2. Section 2000e-16 of the same title makes this core discrimination standard applicable to certain federal employers. Id. § 2000e-16(d). .•
The Secretary does not contest here that DFAS-CO constitutes a federal employer subject to suit under Title VII. There are two principal means of proving employer discrimination under Title VII: disparate treatment and disparate impact. Disparate impact causes of action penalize employment practices that are “fair in form but discriminatory in operation.” Griggs v. Duke Power Co., 401 U.S. 424, 431, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971). Thus, disparate impact analysis is intended to make sure that employers do not use “neutral” decision-making mechanisms that in fact work to eliminate a greater portion of otherwise-qualified protected group members than they do members of other groups.
In a summary judgment case, the plaintiff must identify employment practices challenged and show their disparate impact on the protected group. Austin v. Memphis Light, Gas & Water Div., No. 96-6582, 1997 WL 705094 (6th Cir. Nov.5, 1997).8 1991 amendments to Title VII su*398perseded portions of Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 651, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989), and Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988), which held that a plaintiff in a disparate impact case must show the discriminatory effect of specific practices on protected group members. The amendments provide that if the plaintiff demonstrates that “the elements of a respondent’s decision-making process are not capable of separation for analysis, the decisionmaking process may be analyzed as one practice.” 42 U.S.C. § 2000e-2(k)(1)(B)(i). The amendments also clarified that if the plaintiff meets this burden, the burden of persuasion, not just the burden of production, shifts to the defendant to show a business necessity for the practices with disparate impact. Id. § 2000e-2(k)(1)(A). As an initial matter, the magistrate judge accepted plaintiffs’ argument that, instead of isolating the effect of each employment practice, the DFAS-CO’s entire promotion decision process should be considered as a whole to determine disparate impact. See 42 U.S.C. § 2000e-2(k)(1)(B). The Secretary does not challenge this finding on appeal.
2. Identifying Employment Practices
In the complaint, plaintiffs alleged five kinds of employment practices at DFAS-CO with a discriminatory impact: detailing; ignoring rank and grade requirements to promote white employees without the requisite skills; manipulation of performance reviews, employee awards, and disciplinary action to justify not promoting African-Americans; hazy selection criteria permitting subjectivity, nepotism, and favoritism to enter the process; and the use of an overwhelmingly white group of selecting officials. Despite the fact that many of these alleged practices involve subjectivity, they may nonetheless constitute “employment practices” for purposes of the disparate impact analysis. See Wards Cove, 490 U.S. at 648, 109 S.Ct. 2115; Watson, 487 U.S. at 985, 108 S.Ct. 2777.
We have found that similar allegations of favoritism, but with a negative effect on women, were appropriate for the disparate impact analysis. Scales v. J.C. Bradford & Co., 925 F.2d 901, 908 (6th Cir.1991). A practice showing overt prejudice on the basis of a protected ground, on the other hand, must be challenged in a disparate treatment case.
As to whether these practices are “capable of separation for analysis,” the magistrate judge found it was unnecessary to reach this issue because it found that plaintiffs failed to demonstrate a disparate impact. Plaintiffs urge that the DFAS-CO’s promotion process is incapable of separation, despite conceding that it consists of “eighteen separate steps.” The Secretary does not contest this argument. *399Therefore, we assume that the DFAS-CO’s promotion practices are incapable of separation.
1 3. Establishing Disparate Impact
A series of Supreme Court cases has provided guidance on what statistical data are sufficient to meet the plaintiffs burden of showing disparate impact. In cases challenging hiring practices discriminatory toward African-Americans, for example, one compares the percentage of African-Americans in the job to the percentage of African-Americans in the total qualified local labor force or, if labor force data is unavailable, to the percentage of otherwise-qualified African-Americans in the applicant pool. Wards Cove, 490 U.S. at 650-51, 109 S.Ct. 2115. One then contrasts this ratio with the corresponding ratio for non-protected group members. An EEOC regulation provides, as a rule of thumb, that if the percentage selected from the pool comprises 80% or less of the percentage selected from the non-protected group, a discriminatory effect exists. 29 C.F.R. § 1607.4(d). In cases involving promotion policies, the relevant inquiry is comparing the number of protected group members benefitting from promotions with the number seeking them; this figure is then contrasted with the corresponding ratio for the non-protected group. Connecticut v. Teal, 457 U.S. 440, 448, 102 S.Ct. 2525, 73 L.Ed.2d 130 (1982).
Because DFAS-CO systematically destroys promotions data every two years, the promotion certificates stating which employees were selected as finalists for promotions were unavailable. To determine rates of promotion, the- parties’ options were to create pools of similarly situated workers to incorporate merit into the promotion rate (the Claudy approach), or to look at global rates of promotion within the ethnicities to answer the same question (the Paetzold approach).
a. The Magistrate Judge’s Assessment of the Evidence
The magistrate judge found that plaintiffs had failed to show DFAS-CO’s promotion policies had a negative disparate impact on African-Americans. The magistrate found Paetzold’s analysis “unhelpful,” for two reasons: (1) her “promotion rates” data, spanning four years, failed accurately to document what portion of each ethnic group received ’promotions, since it did not screen out multiple promotions for the same person; (2) her analysis failed to incorporate any consideration of which employees were qualified-for each promotion.
The magistrate judge, however, endorsed the approach the Secretary’s expert, Claudy, took when he created “constructed pools” of qualified applicants. The magistrate judge acknowledged that the Claudy approach assumed only employees from the same grade and series groups would be considéred for the same promotions, but stated, “there is no contention that promotion decisions were made wholly without reference to minimal qualifications.”
Put simply, the magistrate judge engaged in an extensive analysis of the parties’ expert reports and found Claudy’s conclusions more credible. This approach of weighing the credibility of the competing expert reports amounts to improper fact-finding. See Rodgers v. Monumental Life Ins. Co., 289 F.3d 442, 449 (6th Cir.2002). Indeed, competing expert opinions present the “ ‘classic battle of the experts’ and it [is] up to a jury to evaluate what weight and credibility each expert opinion deserves.” Cadmus v. Aetna Casualty and Surety Co., 1996 WL 652769 (6th Cir. Nov.7, 1996) (unpublished) (citation omitted).
*400The matter before the magistrate judge was a motion for summary judgment, not a trial, where credibility determinations are not appropriate. Moreover, as explained below, the magistrate judge’s assessment of the evidence was erroneous, or at best incomplete.
b. This Court’s Assessment of the Evidence.
1) Statistical Evidence
The magistrate judge’s reasoning with respect to the expert opinions is flawed in two respects. First, the case law suggests some skepticism toward the “constructed pools” approach is appropriate. As the Seventh Circuit noted in a case in which the defendant’s expert used multivariate regression to incorporate non-race factors into the analysis of disparate impact, “[T]he use of hindsight to construct ‘qualifications’ for a position must be viewed with some suspicion. The trial judge was correctly dubious of the seductive logic of post-hoc explanation.” Mozee v. Am. Commerc. Marine Serv. Co., 940 F.2d 1036, 1045 (7th Cir.1991).
Plaintiffs allege that managers bent rank and grade requirements in order to promote white employees. Claudy’s arguably fails to encompass a situation in which the group of employees considered for one position are lower-graded white employees and higher-graded African-American employees.9
Second, cases concerning promotions appear reticent to apply the Wards Cove requirement of using qualified persons as the benchmark for the disparate impact analysis. In Teal, the Court examined whether a threshold test administered to employees who wished to become supervisors had a discriminatory effect. 457 U.S. at 444, 102 S.Ct. 2525. Disparate impact was determined by what percentage of each ethnic group passed the test; the Court did not find it necessary to control for factors such as the employee’s professional experience. Id. Similarly, Scales, 925 F.2d 901 (6th Cir.1991), a case from this Circuit, found gender discrimination on the basis of evidence that it took women longer than men to be promoted to the first managerial level in the company. Scales, 925 F.2d at 906. The district court had found the data legally insufficient, because they failed to screen out factors other than gender. This court reversed, noting, “The Supreme-Court has rejected rigid mathematical formulas in analyzing statistics purporting to show disparate impact .... the ‘entire evidence’ in the record must be considered in determining whether a . claim for discrimination has been proven.” Id. at 908 (quoting Bazemore v. Friday, 478 U.S. 385, 404, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986)). The Ninth and Second Circuits, further, conclude that the proper inquiry in promotion cases is “the composition of candidates seeking promotion and the composition of those actually promoted,” without reference to the candidates’ qualifications. Stout v. Potter, 276 F.3d 1118, 1123 (9th Cir.2002); see Waisome v. Port Auth., 948 F.2d 1370, 1372 (2d Cir.1991).
Admittedly, it can fairly be said that both expert’s analysis suffers from some flaws; however, such flaws may be unavoidable given the absence of data on who applied for promotions. In any event, such flaws relate to the weight of the reports which is a matter for the trier of *401fact. Importantly, Paetzold’s report provides some basis for finding that African-Americans received less promotions, at least for overcoming summary judgment. Moreover, expert statistical evidence in disparate impact cases is not to be considered in a vacuum, as the only evidence permitting plaintiffs to meet their prima facie test; it must be considered “in light of all the evidence in the record.” Bazemore, 478 U.S. at 401, 106 S.Ct. 3000. Here, we find that other evidence in the record, discussed below, coupled with the expert reports, is sufficient to create a genuine issue of material fact as to whether or not African-Americans were disparately impacted in promotion rates.
2) Non-Statistical Evidence in the Record
The PAT Report introduces enough new information about the impact of DFAS-CO’s promotion practices to allow plaintiffs to meet the initial summary judgment burden. The PAT Report makes the following relevant points: (1) “Promotions” to the top grade levels 13 and 14 almost always occur through an “appointment” procedure that circumvents the merit promotions process and thus does not allow employees to submit applications. Thus, some of the racism alleged in upward mobility prospects for African-Americans is not captured in data on Competitive promotions. (2) 82% of employees surveyed in the study “perceive career advancement opportunities to be unfavorable due to favoritism, pre-selection, nepotism, racism and now downsizing ...” (3) There is evidence of inequity toward' African-Americans in the distribution of awards, high performance reviews, and disciplinary actions, all of which are mandatory factors in merit promotion decisions.10 These findings, contained in a report commissioned by the DFAS-CO, compensate to some degree for plaintiffs’ failure to demonstrate conclusively that they are promoted at lower rates than white employees. Particularly in this situation, where more thorough statistical analysis is impossible because of data lost or destroyed by DFAS-CO, the PAT Report and the statistical data together raise genuine fact issues that make the question of disparate impact inappropriate for summary judgment.
In sum, several factors in the record viewed in combination are sufficient evidence to defeat summary judgment: (1) evidence (albeit largely statistically insignificant) that African-American employees are promoted less frequently than white employees; (2) evidence that mandatory factors in the promotions process, such as employee awards and employee discipline, disadvantage African-American employees because of racial differences in application; and (3) evidence that many employees perceive favoritism and a lack of transparency as pervading the promotions process at DFAS-CO.
D. Rule 37 Sanctions
The magistrate judge found that the Secretary had violated the court’s December 1998 Discovery and Scheduling Conference Order requiring that the agency “locate and gather all documents responsive to [plaintiffs’ discovery requests] and produce documents responsive to it” upon *402the court’s denial of the Secretary’s summary judgment motion. The agency lost the PAT -Report documents and routinely destroyed promotion records. The magistrate judge felt that there was evidence of neither spoliation nor gross negligence, but that the agency had been negligent in failing to secure the requested documents and to hold them pending the outcome of its summary judgment motion. Plaintiffs requested that the magistrate judge enter judgment in favor of the appellants, or alternatively, that the magistrate judge strike the report of the Secretary’s expert and admit plaintiffs’ expert’s view “as a true statement of statistical evidence.”
The magistrate judge found both proposed sanctions too drastic and decided to grant the motion but wait until trial to decide what sanction to impose. However, since summary judgment was entered in favor of the Secretary, no sanctions were awarded.
Rule 37(b)(2) provides that a district court may make any such order regarding the violation of a discovery order as is just. The choice of what sanction to impose is vested in the court’s discretion. The court considers four factors in deciding whether to impose the drastic sanction of dismissal (against a plaintiff) or entering judgment (against a defendant): (1) evidence of willfulness or bad faith; (2) prejudice to the adversary; (3) whether the violating party had notice of the potential sanction; (4) whether less drastic sanctions have been imposed or ordered. Bass v. Jostens, Inc., 71 F.3d 237, 241 (6th Cir.1995).
In considering the factors, the magistrate judge determined that he could not know how much the unavailability of these documents prejudiced plaintiffs’ case until hearing the merits. Even assuming this conclusion was within his discretion, the magistrate judge granted the Secretary’s motion for summary judgment one month later, leaving the sanctions issue unresolved. This was problematic. The magistrate judge never made an assessment of the quality of the evidence lost and did not mention the sanctions issue in its summary judgment order. If anything, the magistrate judge appears to have found that both parties suffered as a result of the lost evidence as it exposed flaws in their expert reports. However, the magistrate judge had previously found that plaintiffs were entitled to sanctions. The magistrate judge’s failure to address the sanctions issue in light of the summary judgment ruling resulted in the imposition of no sanctions of any kind. This was an abuse of discretion under the circumstances.
III. CONCLUSION
For the reasons stated above, the magistrate judge’s decision is AFFIRMED IN PART, REVERSED IN PART, and REMANDED for proceedings consistent with this opinion. The case continues on plaintiffs’ Title VII claim of race discrimination under a disparate impact theory with the following plaintiffs: Willie S. Phillips, Todd Brooks, Yulander Edwards, Emma Fields, Denise Goudy, Kevin Gray, Jackie McGra-dy, Susan McNeal, Carlisa Miles, Mary L. Phillips, Lisa Prater, Marcia Spence, and Lerdón Woodfolk.
. As described in more detail below, the ratio of the two groups’ percentages of awards is 24:35 or .679; thus, the disparity of impact passes the "four-fifths rule” the EEOC uses as a rule of thumb. 29 C.F.R. § 1607.4.
. This disparity, like the disparity in awards, meets the four-fifths rule; the percentage of white employees subject to disciplinary actions is 28% of the number of African-American employees subject to discipline.
. One plaintiff, Lori Dillard, subsequently withdrew.
. Paetzold stated in her deposition that she “did not make any statistical findings specifically with regard to representation rates.”
. Neither party had the benefit of employment records showing which employees applied for and received which promotions; the Department of Defense destroys this data every two years, a point discussed infra.
. The Defense Department database included names, so it allowed Paetzold to ensure that she did not count any employee more than once. However, it did not allow her to avoid counting more than once multiple promotions of the same employee,. '
.She also analyzed the number of promotions in each year as a percentage of the number of employees in that year, finding that African-American men’s promotion rate in 1993 was significantly lower than white men’s (30.77%, as compared to 37.79%); and that African-American women’s promotion rate in 1994 was significantly lower than white women’s (26.5%, as opposed to 29.6%).
. The magistrate judge’s opinion, and the Secretary's brief, cite the First Circuit case EEOC v. Steamship Clerk’s Union, 48 F.3d 594, 601 (1st Cir.1995), for the proposition that the prima facie disparate impact case contains three elements: identification, disparate impact, and causation. Separating "disparate impact” from "causation” assumes *398that the employee challenges a specific component of the employer’s hiring or promotion decisions, and further that the practice alone does not preclude the employee from employment opportunities. "Disparate impact," under Steamship Clerks Union, would be demonstrated by showing that the group is underrepresented in the employer's workforce; "causation” would be established by showing that a specific practice — for example, a screening test, or a height or weight requirement — accounts for the group's under-representation. See Wards Cove, 490 U.S. at 657, 109 S.Ct. 2115; Steamship Clerks Union, 48 F.3d at 601. If the employee challenges the employer's promotion process as a whole, however — as is the case here — then the disparate impact and causation elements merge. The magistrate judge addressed causation in a separate section from disparate impact, when in fact both sections address the same question: whether the evidence in the record supported a finding that African-American employees were promoted at a lower rate than white employees.
. Plaintiffs emphasize on appeal that many of the African-American workers at DFAS-CO began with the agency in 1991, transferring from other federal jobs. Many of the white employees, who are allegedly advancing more quickly, were hired "from the street" during the agency's rapid hiring period from 1991 to 1995.
. The findings of the DFAS-CO's EEO officer also provide support for the contention that two elements of the promotion process, the distribution of awards and imposition of disciplinary actions, had a negative disparate impact on African-American employees. As a percentage of their numbers in the office, African-Americans were only two-thirds as likely as white employees to receive awards; white employees, on the other hand, were less than one-third as likely to be subject to disciplinary proceedings.
. This disparity included more substantial discrepancies within subcategories. For example, the expert found that white women had a 56.07% promotion rate compared to African-American women's 42.34% for the 1992-1996 period. J.A. at 1117.