REAVLEY, Circuit Judge:
This is a discrimination case in which the employer defendant has prevailed and will now prevail again. Notwithstanding the voluminous record constructed by the parties in the district court and the extensive findings of fact and conclusions of law entered by the district judge,
our decision in this appeal involves little more than the straightforward application of well settled principles of employment discrimination law, chosen from our extensive writings, to a handful of subsidiary findings of fact over which there is no dispute. As a result, several issues raised in this appeal, principally those concerning the use of multiple regression and other sophisticated computer assisted models, require no attention.
I.
Procedural History
Claiming that he was the victim of impermissible employment discrimination, Riley D. Pouncy (“Pouncy” or “the appellant”), a black, brought an action against his employer, The Prudential Insurance Company of America (“Prudential” or “employer”).
His complaint contained two claims.
One, that Prudential had discriminated against him in 1975 when it promoted a coworker, Olga Aschenbeck, a white, to the position of assistant supervisor in the duplicating section of the office services division at Prudential’s Southwestern Home Office in Houston, Texas. The other, that Prudential had discriminated against its black employees as a class by systematically failing to promote them within its work force and otherwise affording them the same conditions of employment as given to whites. A third issue — that the appellant was discharged in retaliation for filing a charge with the EEOC — was raised at trial.
The district judge provisionally certified the plaintiff class to include all present and future black employees, other than part-time employees, wage band employees, and commissary workers, in the administrative section at Prudential’s Houston office,
Pouncy v. Prudential Insurance Co. of America,
499 F.Supp. 427, 431 (S.D.Tex. 1980);
see also id.
at 442-45. After a 13 day trial limited to the issue of liability, in which both parties offered anecdotal, documentary, and statistical evidence, the district judge found
inter alia
that Pouncy had not been denied the promotion to assistant section supervisor due to his race, that
Pouncy’s subsequent discharge by Prudential was not in retaliation for filing his charge of employment discrimination with the Equal Employment Opportunity Commission, and that Prudential had not maintained a system of promotion or advancement that had discriminated against blacks as a class.
Id.
at 466-67.
In this appeal, Pouncy raises two contentions. First, he disputes the district judge’s conclusion that he was not the victim of racial discrimination. Second, he argues that the statistical evidence presented through his two expert witnesses coupled with the other evidence demonstrates that Prudential treats blacks, as a class, less favorably than whites.
We hold that the subsidiary findings of fact made by the district judge lead to the conclusion that -Pouncy was neither the victim of racial discrimination by his employer nor received a retaliatory discharge. We also hold that the appellant’s evidence on the class claims, mostly statistics, failed to show that Prudential’s black employees were treated differently from white employees in terms of promotions, compensation, and in their use throughout Prudential’s work force.
II.
Individual Claims
The district judge predicated his decision that the appellant had not been the victim of racial discrimination on two grounds. First, he concluded that Pouncy had not established a prima facie case of discrimination since he was not qualified for the promotion to assistant supervisor in the duplicating section.
Id.
at 433-34, 466. “Although he had achieved technical proficiency in his performance in the duplicating section,” the district judge found that Pouncy “lacked the necessary communication, leadership and training skills to perform as a supervisor.”
Id.
at 434. Alternatively, the district judge held that even if a prima facie case had been proved, Prudential had established that the decision to promote Olga Asehenbeck rather than the appellant was made for non-racial reasons and the appellant “failed to satisfy [his] ultimate burden of establishing that he was discriminated against on the basis of his race.”
Id.
at 435.
Likewise, the district judge anchored his decision that the appellant was not discharged in retaliation for filing a charge with the EEOC both on Pouncy’s inability to make out a prima facie case and, alternatively, on a finding that Prudential “ha[d] adequately rebutted plaintiff’s case by proving that plaintiff’s termination was based on a legitimate, non-discriminatory reason.”
Id.
at 467. According to the district judge, Pouncy either did not establish that he was discharged “for an infraction for which non-minorities were not,” or, assuming a prima facie case had been made, “failed to sustain the inference that his discharge was the result of racial discrimination.”
Id.
at 437, 438.
The ultimate issue — whether the appellant was the victim of racial discrimination — is a question of ultimate fact to which the clearly erroneous rule does not apply.
Cousin v. Board of Trustees,
648 F.2d 293, 296 & n.3 (5th Cir. 1981). Otherwise, our review of the district court’s subsidiary findings is constrained by the clearly erroneous standard.
Id.
at 295. “A finding is ‘clearly erroneous’ when although there is evidence to support it, the reviewing court on the entire evidence is left with the definite and firm conviction that a mistake has been made.”
United States v. United States Gypsum Co.,
333 U.S. 364, 395, 68 S.Ct. 525, 542, 92 L.Ed. 746 (1948).
We see no reason to decide whether Pouncy established a prima facie case of discrimination since it is clear from the
district judge’s findings that as a plaintiff Pouncy was unable to overcome Prudential’s evidence and discharge “the ultimate burden of persuading the court that [he had] been the victim of intentional discrimination.”
Texas Department of Community Affairs v. Burdine,
450 U.S. 248, 256, 101 S. Ct. 1089, 1095, 67 L.Ed.2d 207 (1981). The record fully supports Judge Bue’s finding that Olga Aschenbeck was chosen as assistant supervisor of the duplicating section for a legitimate, nondiscriminatory reason. She, unlike Pouncy, possessed initiative and the ability to communicate with and train her coworkers. Equally supported by the evidence is the district judge’s conclusion that Pouncy was discharged due to his failure to perform a legitimate work assignment: to rewrite a memorandum received from Prudential’s corporate office. As in most individual Title VII actions that reach trial, the issue of racial discrimination was in hot dispute. The district judge was charged with the task of finding the facts. Given the findings that he made, the conclusion that Pouncy was not the victim of racial discrimination is inescapable.
III.
Class Claims
In the district court, Pouncy presented four sets of statistics to show that Prudential discriminated against its black employees. First, he presented evidence showing that for employees hired between 1973 and 1977 the mean weekly salary of white employees was greater than the mean weekly salary of black employees hired in the same year.
See
499 F.Supp. at 449. The appellant also introduced evidence showing that in each year from 1973 to 1975, the percentage of blacks receiving promotions to managerial and supervisory positions was less than the percentage of blacks comprising Prudential’s total work force.
See id.
at 454. For each year from 1973 to 1977, Pouncy demonstrated that, with one exception in 1975, prior to promotion to levels 5 through 12, black employees had a greater mean number of years of service with Prudential than did white employees.
Id.
at 454-55. Lastly, Pouncy introduced evidence showing that from 1973 to 1975 blacks were clustered in the lower levels of Prudential’s work force and “underrepresented in upper levels of [Prudential’s] workforce when compared with their percentage in [Prudential’s] total workforce and with the workforce as a whole.”
Id.
at 459.
Pouncy argued that three of Prudential’s employment practices were responsible for the disparities shown in his statistics. First, he complained that job vacancies at Prudential were not “posted” or otherwise made known to the employees. Rather, managers and other supervisory personnel, a majority of whom were white, selected employees for promotion using minimal objective criteria. Pouncy also complained that Prudential’s level system, through which clerical employees often are hired at entry level positions and subsequently promoted through the employer’s work force into better jobs,
retains black employees at the lowest paying and least skilled jobs. Lastly, Pouncy singled out Prudential’s use of subjective criteria in employee performance evaluations.
. A.
The Disparate Impact Model
By structuring his case to establish that three of Prudential’s employment practices accounted for the disparities in the racial composition of Prudential’s work force, the appellant attempted to fit his proof into the disparate impact model of an employment discrimination case. This theory of recovery under Title VII is used to attack employment selection criteria that are facially neutral yet fall more harshly on a protected class of employees.
International Brotherhood of Teamsters v. United States,
431 U.S. 324, 335 n.15, 97 S.Ct. 1843, 1854 n.15, 52 L.Ed.2d 396 (1977). Title VII “proscribes not only overt discrimination but also practices that are fair in form, but discriminatory in operation. The touchstone is business
necessity. If an employment practice which operates to exclude [blacks] cannot be shown to be related to job performance, the practice is prohibited.”
Griggs v. Duke Power Co.,
401 U.S. 424, 431, 91 S.Ct. 849, 853, 28 L.Ed.2d 158 (1971).
In order to establish a claim of racial discrimination under this theory, a plaintiff need only show that a facially neutral employment practice produces a significantly adverse impact on one race.
Dothard v. Rawlinson,
433 U.S. 321, 329, 97 S.Ct. 2720, 2726, 53 L.Ed.2d 786 (1977). The employer’s intent to discriminate against a class of employees is not at issue.
When coupled with an identifiable employment practice, neutral in form but discriminatory in operation, statistics that show a significant disparity in an employer’s work force raise “an inference that employment decisions are tainted by intrusion of illegitimate concerns.”
Rivera v. City of Wichita Falls,
665 F.2d 531, 535 (5th Cir. 1982). A prima facie case is shown by identification of a neutral employment practice coupled with proof of its discriminatory impact on the employer’s work force.
Johnson v. Uncle Ben’s, Inc.,
657 F.2d 750, 753 (5th Cir. 1981).
The discriminatory impact model of proof in an employment discrimination case is not, however, the appropriate vehicle from which to launch a wide ranging attack on the cumulative effect of a company’s employment practices. Nor may just any employment practice be challenged under this model simply because an uneven racial balance exists in an employer’s work force. As originally conceived in
Griggs v. Duke Power Co.,
401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158, an action in which a group of black employees challenged their employer’s requirement of a high school diploma and a satisfactory score on two aptitude tests for positions in several departments of a power generating facility, the disparate impact theory applied to an “overt, clearly identified nondiscretionary selection criterion] that [was] applied at a single point in a selection process.” D. Baldus & J. Cole, Statistical Proof of Discrimination § 1.23, at 12 (1981 Supp.). Although some courts have used the disparate impact model of proof to challenge multiple employment practices simultaneously,
see id.,
this is an incorrect use of the model. The disparate impact model applies only when an employer has instituted a specific procedure, usually a selection criterion for employment, that can be shown to have a causal connection to a class based imbalance in the work force. Thus, a disparate impact analysis may be used to challenge aptitude and intelligence tests,
see Griggs,
401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158, educational requirements,
see id.,
height and weight requirements,
see Dothard,
433 U.S. 321, 97 S.Ct. 2720, 53 L.Ed.2d 786, an employer’s refusal to employ persons who use methadone,
see New York City Transit Authority v. Beazer,
440 U.S. 568, 99 S.Ct. 1355, 59 L.Ed.2d 587 (1979), who have a poor credit rating,
see
3 A. Larson & L. Larson, Employment Discrimination § 73.00 (1981 & 1981 Supp.), or who have a record of arrests,
see id.,
so long as a resulting disparate impact may be shown.
We require proof that a specific practice results in a discriminatory impact on a class in an employer’s work force in order to allocate fairly the parties’ respective burdens of proof at trial. The aggrieved party must prove a disparate impact due to the selection procedure. The employer then has the burden of proving that the selection procedure is justified by a legitimate business reason.
Johnson,
657 F.2d
at 753. Identification by the aggrieved party of the specific employment practice responsible for the disparate impact is necessary so that the employer can respond by offering proof of its legitimacy. “Knowledge of a legitimate business reason is uniquely available to the employer who is . . . required to persuade the court of its existence by a preponderance of the evidence.” Id. We do not permit a plaintiff to challenge an entire range of employment practices merely because the employer’s work force reflects a racial imbalance that might be causally related to any one or more of several practices for to do so
would allow the disparate impact of one element to require validation of other elements having no adverse effects. The burden of determining the validity of a screening procedure, weighing not only on the employer but also on the limited resources of the district court, will not be imposed where proof of an absence of discriminatory effect attributable to the procedure shows it to be unwarranted.
Rivera,
665 F.2d at 539.
As we have noted, at trial the appellant tailored his proof on the class claims to fit the disparate impact model. The findings and conclusions of the district judge also reflect the use of the disparate impact analysis. On appeal, both sides have briefed the class claims as presenting a disparate impact case. We conclude, however, that the proof adduced at trial does not fit the disparate impact model. Even if we accept all of the appellant’s statistics as accurately reflecting the condition of black employees in Prudential’s work force, a step the district judge was unwilling to take, it is clear that Pouncy has not shown that a facially neutral employment practice used by Prudential falls more harshly on black employees.
None of the three Prudential “employment practices” singled out by the appellant — the failure to post job openings, the use of a level system, and evaluating employees with subjective criteria — are akin to the “facially neutral employment practices” the disparate impact model was designed to test. Unlike educational requirements, aptitude tests, and the like, the practices identified by Pouncy are not selection procedures to which the disparate impact model traditionally has applied.
See
3 A. Larson & L. Larson, Employment Discrimination § 73.00 (1981 & 1981 Supp.).
Moreover, the appellant cannot make a showing that the Prudential employment practices that he has identified have caused the racial imbalance in Prudential’s work force. The disparate impact model requires proof of a causal connection between a challenged employment practice and the composition of the work force. Aptitude tests, height and weight requirements, and similar selection criteria all may be shown to affect one class of employees more harshly than another by controlling for the impact of the employment practice on one class in the employer’s work force so that it can be measured. For example, in
Griggs
the plaintiff proved not just a racial imbalance in the employer’s work force, but also that the employer’s educational requirement resulted in disqualifying a higher percentage of black employees than white employees. 401 U.S. at 430 n.6, 91 S.Ct. at 853 n.6. In an action challenging an employer’s height and weight requirements, the plaintiff demonstrated that, as compared to men, a disproportionate number of women were automatically disqualified from obtaining jobs as corrections counselors.
Dothard,
433 U.S. at 329-30 & n.12, 97 S.Ct. at 2726-2727 & n.12.
By contrast, Pouncy has not shown, nor can he show, that independent of other factors the employment practices he challenges have caused the racial imbalance in Prudential’s work force. The statistics presented by the appellant do show that, on the whole, blacks are overrepresented in the lower levels of Prudential’s work force. But this might result from any number of causes. Absent proof that the disparate
impact is caused by one or more of the challenged employment practices, we do not require the employer to justify the legitimacy of any (or all) employment practices. In sum, the nature of the evidence presented by the appellant at trial could not establish a case of employment discrimination based on the disparate impact model.
B.
The Disparate Treatment Model
Although the appellant structured his evidence to fit a disparate impact case, a review of the record makes clear that Pouncy also claimed that Prudential discriminated against its black employees by treating them differently from its white employees. Thus, the appellant also attempted to prove a classwide disparate treatment case. In contrast to a disparate impact case, motive must be proved in a disparate treatment case.
Wilkins v. University of Houston,
654 F.2d 388, 395 (5th Cir. 1981). Moreover, to succeed in a disparate treatment case, Pouncy “had to prove more than the mere occurrence of isolated or ‘accidental’ or sporadic discriminatory acts. [He] had to establish by a preponderance of the evidence that racial discrimination was the company’s standard operating procedure — the regular rather than the unusual practice.”
Teamsters,
431 U.S. at 336, 97 S.Ct. at 1855 (footnote omitted).
Statistical evidence is also of central importance in a disparate treatment case. It may be used to show both motive and a pattern or practice of racial discrimination. In a proper case, we may infer racial discrimination if gross statistical disparities in the composition of an employer’s work force can be shown.
Id.
at 335 n.15, 97 S.Ct. at 1854 n.15. When “the statistical showing is sufficiently strong in a disparate treatment action, [the] plaintiffs’ prima facie case can be made without additional evidence establishing that [the] defendant purposefully treated minorities protected under Title VII less favorably than other persons.”
Wilkins,
654 F.2d at 395.
Statistical evidence, however, must be meaningful.
Because of the significant role that statistics can play in discrimination cases and of their inherently slippery nature, it is imperative that they be used properly. While gross statistical dispárities may alone establish a prima facie case of employment discrimination in a proper case, the Supreme Court has cautioned “that statistics are not irrefutable; they come in infinite variety and, like any other kind of evidence, they may be rebutted. In short,.their usefulness depends on all of the surrounding facts and circumstances.”
Id.
(quoting
Teamsters,
431 U.S. at 340, 97 S.Ct. at 1857). With this caveat in mind, we advert to the statistical evidence comprising the appellant’s case.
1.
Salary Evidence
The appellant introduced a series of exhibits comparing the mean salary of Prudential employees hired in 1973 through 1978 and still on the payroll in 1978. The exhibits were prepared by selecting from a 1978 list of Prudential employees all black employees hired in 1973. The weekly salaries of these employees were added together and the sum was then divided by the number of employees producing a mean salary for a black employee at Prudential hired in 1973 and still with Prudential 5 years later. This process was repeated for white employees and black employees hired in each year, 1973 through 1978 inclusive, and still with Prudential. This exercise showed that when compared to black employees hired in the same year, white employees had a greater mean weekly salary.
See
499 F.Supp. at 449.
The district judge characterized the appellant’s series of exhibits as “a naked comparison of average weekly salaries without regard to level of skill, education and training [from which n]o meaningful conclusion can be drawn,” “much less a
prima facie
case.”
Id.
We agree. The
appellant’s statistical evidence is deficient because it “fails to take into account the fact that a number of factors operate simultaneously to influence the amount of salary [an employee] receives.”
Wilkins,
654 F.2d at 402. The discrepancies between the mean salary of black employees and the mean salary of white employees hired in specific years may be explained by any number of nondiscriminatory factors. Different job levels, different skill levels, previous training, and experience: all may account for unequal salaries in an environment free of discrimination. The appellant’s exhibits make the baseless assumption that the mean salary of employees, by race and by year hired, should be equal regardless of the level at which the employee originally was hired. Under the appellant’s analysis, an attorney hired at level 20 with a salary of $20,000 would be compared with a file clerk of a different race with a salary of $6500 if both were hired in the same year. From this inconclusive evidence we are unable to hold that Prudential’s black employees, as a class, were the victims of employment discrimination.2 *******
2.
Promotion Evidence
The appellant also introduced two series of exhibits focusing on the allocation of promotions within Prudential’s work force. One series of exhibits revealed that, with few exceptions, in each year from 1973 to 1975 the percentage of black employees receiving promotions to supervisory and managerial positions, levels 10 through 84, was less than the percentage of blacks in Prudential’s work force. The other series of exhibits showed that, with one exception in 1975, when comparing, by race, employees promoted to levels 5 through 12 in the years from 1973 to 1977, black employees had a greater mean number of years of service with Prudential than white employees. The district judge was persuaded by neither of the appellant’s series of exhibits that Prudential discriminated against its black employees in awarding promotions.
499 F.Supp. at 455. We agree with the district judge that the appellant’s promotional evidence does not demonstrate that Prudential has discriminated impermissibly against its black employees.
The series of exhibits that show the percentage of employees, by race, promoted to managerial and supervisory positions at Prudential in each year were prepared simply. The number of promotions to each level, from level 10 to level 82, for each year from 1973 to 1975, were tallied. Then, each sum was analyzed to determine the number of white employees and the number of black employees receiving a promotion to that level that year. Percentages of white employees and black employees were then calculated.
Although the percentage of black employees that had received promotions to levels 10 through 82 generally is far less than the percentage of black employees in Prudential’s work force, we refuse to infer that Prudential treated its black employees less favorably in awarding promotions. The appellant’s argument that this evidence demonstrated that Prudential discriminated against its black employees in promotions says nothing of the required comparison to a qualified pool of employees presumptively eligible for promotion.
See Wilkins,
654 F.2d at 401. Instead, the appellant makes the baseless assumption that
all
employees are equally qualified for
all
positions — an assumption specifically rejected by the district judge as not supported by the evidence. 499 F.Supp. at 454.
The appellant’s other series of exhibits on promotions at Prudential was pre
pared by finding the sum of the number of years of service at Prudential for each employee, by race, promoted to each level, 5 through 12, at the time of promotion. These sums were then divided by the number of employees, by race, promoted to the different levels. The resulting statistic— the mean number of years of service at Prudential for employees, by race, at each level — indicated that, with one exception, black employees were employéd by Prudential longer than white employees before being promoted to each level.
Like the district judge, we refuse to infer from this presentation that Prudential’s black employees are disfavored in promotions.
[T]his exhibit does not reflect the level of hire of the employees whose promotion rates are compared. Such an omission could result in misinterpretation of data. As an example, [the exhibit] indicates that Blacks averaged 7.16 years of service prior to promotion [to level 9 in 1975]. But if these same white employees had been college hires (hired in at level 8) and had taken an average of 5.8 years to advance one level, and the black employees had been entry level hires (level 2) and had required an average of 7.16 years to advance 7 levels, the same data would indicate that Blacks actually enjoyed a more rapid rate of advancement than Whites.
Thus, as defendant’s expert testified and as these contrasting results demonstrate, no conclusions can be drawn from such an analysis of years prior to promotion, without more, because its results cannot be attributed to any one cause — in this instance, it cannot be determined whether a certain number of years prior to promotion results from the relative level of hire or rate of promotion.
499 F.Supp. at 455 (footnotes omitted). We previously have suggested that the mean length of time between promotions may be relevant in proving discrimination.
Wilkins,
654 F.2d at 402. But that was not proved in this case. The appellant’s evidence on promotions did not establish that Prudential’s black employees were the victims of disparate treatment.
3.
Utilization Evidence
The district judge held that the appellant’s utilization statistics, fully described in his findings,
see
499 F.Supp. at 459, established a prima facie case of employment discrimination,
see id.
He went on to conclude, however, that Prudential’s “ ‘more refined, accurate and valid statistics,’ ”
id.,
reflecting a gradual but steady increase in the percentage of black employees in all levels of Prudential’s work force, was “inconsistent with any policy of discrimination during the time period relevant to this lawsuit,”
id.
at 460.
In this appeal, Prudential argues that the appellant’s- utilization statistics did not make out a prima facie case of employment discrimination. We see, however, no reason to decide this question for, whether or not a prima facie case was established, and no matter what burden, if any, shifted to Prudential, it is abundantly clear, considering the evidence that the number and percentage of black employees steadily increased in all levels of Prudential’s work force,
see id.
at 459-61, that on the ultimate issue— whether Prudential had discriminated against its black employees — the district judge was justified in concluding as he did.
See Movement for Opportunity and Equality v. General Motors Corp.,
622 F.2d 1235, 1245 (7th Cir. 1980). Likewise, any claim of the appellant that his utilization evidence mandated a finding of discrimination is without merit.
IV.
Conclusion
Two recent decisions of this court have closed by sounding a word of caution to
those who would enter the courthouse armed with statistics that prove little more than a litigant’s resourcefulness at manipulating numbers.
See Rivera,
665 F.2d at 547;
Wilkins,
654 F.2d at 410. We would add nothing by repeating that warning again. Instead, we note only that from the inconclusive statistical evidence presented during the trial of this action, even without much of the exotic presentation of the defendant, the district judge hardly could have found other than he did.
The judgment of the district court is AFFIRMED.