BOGGS, Circuit Judge.

This is an age discrimination case involving questions of disparate treatment and disparate impact. When Federal Forge, Inc. learned that former employees of one of its closed plants were demanding seniority pension benefits if they were rehired, it imposed a moratorium on any hiring of those workers. Raymond Abbott and 35 other former employees sued, charging that the moratorium amounted to a refusal to hire on the basis of age, in violation of the Age Discrimination in Employment Act (ADEA), 29 U.S.C. § 621, et seq., because the workers affected by the moratorium included a higher percentage of workers 40 years of age or older than the' workers actually hired. Federal Forge was under no legal or contractual obligation to rehire any of these workers. They now appeal the district court’s grant of summary judgment for Federal Forge. Because appellants failed to establish a prima facie case of either disparate treatment or disparate impact and because Federal Forge established a legitimate and non-discriminatory reason for the moratorium, we affirm the district court’s grant of summary judgment.

Federal Forge, Inc. (the company) is a Michigan corporation that formerly operated two forging plants in the Lansing area — the “Federal Division” plant and the “Lansing Division” plant. The appellants in this case worked in the Federal Division plant. Employees in both plants were represented by the United Auto Workers (UAW) Union, but were in separate bargaining units with separate collective bargaining agreements, pension plans, and seniority systems. In November 1982, the Federal Division plant closed permanently for economic reasons and the employees were laid off. 1 The Lansing Division plant continued to operate.

The union interpreted the collective bargaining agreement between it and Federal Forge to require that employees laid off at the Federal Division plant be given preferential hiring status for openings at the Lansing Division plant for twelve months after the closing of the former plant. Federal Forge disagreed with that interpretation as a matter of law, but complied with it. It also continued to hire only former Federal Division employees at the Lansing plant even after the twelve-month period expired on November 19, 1983.

In September 1983, the UAW filed a grievance claiming that all former Federal Division employees hired at the Lansing Division plant were entitled (pursuant to the collective bargaining agreement) to retroactive seniority and pension benefits based on their Federal Division pension seniority (or “credited service”). Federal Forge initially understood this grievance to refer only to the employees hired at Lansing during the twelve-month preference period. Believing that none of the rehired *871 Federal Division employees was entitled to credited service, Federal Forge rejected the grievance and denied past seniority to all of the already-hired employees. Under the terms of the collective bargaining agreement, the grievance would eventually be resolved through binding arbitration.

In March 1984, the company was made aware that the Union’s position in its grievance was that all former Federal 'Division employees who were hired at the Lansing Division plant were entitled to past seniority pension benefits, whether or not they were hired during the preference period. The company immediately declared a moratorium on the hiring of former Federal Division employees at the Lansing plant until the arbitration on the union’s grievance was resolved. At that point, Federal Forge had already rehired 45 of the former Federal Division employees; 107 former employees had not yet been rehired. The arbitrator decided the union’s grievance in January 1985, ruling that only the former employees hired during the twelve-month preference period were entitled to past seniority pension benefits. Federal Forge requested a clarification of the arbitrator’s decision, which it received in July 1985. In May 1985, after it was satisfied that former Federal Division employees hired at the Lansing Division plant after the twelvemonth preference period would not be entitled to past seniority pension benefits, the company lifted the moratorium.

During the 14 months of the moratorium, Federal Forge hired 27 new employees at the Lansing Division plant. Twenty-three of those were “off-the-street,” and four were former Federal Division employees. 2 The off-the-street hires were generally unskilled workers with no prior hot forge experience. The parties do not dispute that, in terms of job experience alone, the former Federal Division employees were better qualified for employment as forge workers than the off-the-street hires.

According to statements of Nelson Henry, president of Federal Forge, and other officers of the company, the purpose of the moratorium was to save the company money should the arbitration result in a decision that Federal Division workers hired after the twelve-month preference period were entitled to past seniority pension benefits. Federal Forge felt it was within its rights in instituting the moratorium because all parties agreed that continued hiring of Federal Division employees after the preference period was discretionary. It is also undisputed, from the statements and testimony of Henry and other officers, that at the time it imposed the moratorium, Federal Forge considered the Union’s demand to be illegitimate.

Forty-seven of the 107 former Federal Division employees still unhired at the time of the moratorium were 40 years of age or older — or attained 40 years of age at some point during the moratorium. Thirty-eight of those employees brought the present action against Federal Forge, alleging violation of the ADEA and various pendent state claims. On cross motions for summary judgment after discovery, the district court ruled for the company. This appeal followed. Two of the original 38 plaintiffs withdrew from the case, leaving 36 appellants.

Disparate Treatment

In order to make out a prima fa-cie case under a theory of disparate treatment, the plaintiff must show that the employer acted with the intention of discriminating on the basis of age. We are satisfied that the district court was correct that there was no genuine issue of material fact as to Federal Forge’s motive. The motive for imposing the moratorium was Federal Forge’s belief that the union’s claim that workers were entitled to past seniority pension benefits upon rehire was meritless. *872 Appellants offered no other circumstantial evidence of intention to discriminate on the basis of age.

Other evidence supports the conclusion that Federal Forge’s intentions in imposing the moratorium did not include a desire not to hire older workers. Prior to the moratorium, Federal Forge had filled the openings at the Lansing Division plant exclusively with former Federal Division workers. In all, 45 workers were hired. Of that group, over half (23) were 40 years or older at the time they were hired. Employee lists show that had Federal Forge intended to fill the openings in the Lansing Division plant with workers under 40, there were enough among the Federal Division employees for it to do so. Indeed, over 60% of the remaining 107 Federal Division employees were under 40 as of March 1984. Moreover, the dates of hire at the Lansing Division show that Federal Forge displayed an early preference for older workers; all of the first eight and 11 of the first 14 workers hired prior to the moratorium were 40 or older at the time of hire.

Ill

Disparate Impact

Appellants also advanced a disparate impact theory by comparing the ages of the persons not hired because of the moratorium with the ages of the persons hired during the moratorium. Under a theory of disparate impact, the plaintiff may establish a prima facie case of discrimination by showing the existence of an employment practice which, although neutral on its face, has the effect of disproportionately affecting persons in a legally protected group. Griggs v. Duke Power Co., 401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971). In a disparate impact case, it is not necessary to present evidence of the employer’s subjective intent to discriminate, as required in a disparate treatment case. Although disparate impact analysis was developed by the Supreme Court in Title VII race discrimination cases, and has been used by the Court only in that context, other courts have widely applied disparate impact analysis to age discrimination cases brought under the ADEA. Laugesen v. Anaconda Co., 510 F.2d 307, 311 (6th Cir.1975); Geller v. Markham, 635 F.2d 1027, 1032 (2d Cir.1980).

In order to establish a prima facie case of age discrimination by showing disparate impact, the plaintiff must first identify a specific employment practice being used by the employer. Watson v. Fort Worth Bank and Trust, 487 U.S. 977, 994, 108 S.Ct. 2777, 2788, 101 L.Ed.2d 827 (1988). The employment practice in this case was the fourteen-month moratorium on hiring former Federal Division employees. Next, the plaintiff must show an adverse effect caused by the employment practice by offering “statistical evidence of a kind and degree sufficient to show that the practice in question has caused the exclusion of applicants for jobs ... because of their membership in a protected group.” Id. at 994, 108 S.Ct. at 2788-89. The protected group in this case is persons between the ages of 40 and 70, as specified in the ADEA.

Once the plaintiff establishes a prima facie case of disparate impact by statistical evidence, the defendant must articulate a legitimate, non-discriminatory reason, such as business necessity or job-relatedness, for the employment practice. At that point, the burden shifts back to the plaintiff to show either that the employer’s reason is a pretext for discrimination, or that there exists an alternative employment practice, without the disparate impact, that also serves the employer’s legitimate interests. Id. at 998, 108 S.Ct. at 2790. The cost to the employer of such an alternative practice is a factor in determining whether it serves the employer’s legitimate business goals. Ibid.

Statistical evidence

The statistical evidence in this case consists of lists of the names, ages, and dates of hire of the former Federal Division employees who were affected by the moratorium (referred to as “List B”) and the persons who were employed during the moratorium (“List C”). Appellants claim that a *873 significant disparity exists between the number of persons 40 or older in the two groups. There were 107 former Federal Division employees on List B; 47 of them attained the age of 40 years (and thus, membership within the protected class) by the end of the moratorium. Four of the former Federal Division employees were hired during the moratorium, of whom one was over 40 years of age. Thus, 103 persons can be said to have been barred from employment at the Lansing Division plant because of the moratorium, of whom 46, or 44.7%, were in the protected class. Among the 27 workers hired at the Lansing Division plant during the moratorium, five attained the age of 40 years within the relevant time. Thus, of the group hired during the moratorium, 18.5% were within the protected class. Appellants argue that the disparity in these percentages makes out a prima facie case of age discrimination.

Challenges to the statistics

The Supreme Court has rejected rigid mathematical formulas in analyzing statistics purporting to show a disparate impact. Watson, 487 U.S. at 994-95, 108 S.Ct. at 2789. Courts are left to decide on a case-by-case basis whether statistics that purport to show disparate impact are in fact sufficient to the task. There are many ways to assess the significance or sufficiency of statistical evidence. For instance, a defendant may question whether the disparity is great enough to create a presumption that it does not result from random chance ("statistical disparities must be sufficiently substantial that they raise such an inference of causation.” Watson, 487 U.S. at 995, 108 S.Ct. at 2789); whether the sample size is large enough for the statistics to be significant; or whether proper groups were used for comparison. In this ease, the company argues that the disparity in the percentage of older workers is not sufficient to make out a case of discrimination, and that appellants’ statistics compare the wrong groups.

Federal Forge first argues that the correlation between age and those affected by the moratorium is not close enough to demonstrate age discrimination. Even though the group affected by the moratorium had a higher percentage of those 40 and over than the group hired, a majority of the former group was still under 40. The company argues that a moratorium that adversely affects a greater number of those under 40 than those 40 and over can hardly be considered discriminatory on the basis of age. That argument, however, does not foreclose a finding of disparate impact. The significance of the percentage of protected class members in the group affected by the moratorium cannot be measured in isolation; it requires comparison with another relevant group. It is possible to demonstrate disparate impact against a protected group in a facially neutral moratorium that affects more nonmembers of the protected group than members, as long as enough members of the protected group are affected to reduce substantially the number of members of the group available to be employed.

Federal Forge next argues that appellants’ statistics compare the wrong groups. In Wards Cove Packing Co. v. Atonio, — U.S. -, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989), the Supreme Court emphasized the importance of identifying the proper groups for comparison in a disparate impact analysis. In Wards Cove, the plaintiffs showed a disparity between the percentage of minority workers in different job categories within its plant. The Court held that the proper comparison was between the proportion of members of the protected class in the employer's work force and the proportion of members of that protected class in the qualified labor pool available to the employer.

The statistics offered by appellants compare the group of former Federal Division employees with the group of workers actually hired during the moratorium. It is not clear that this is the relevant comparison. Once the twelve-month hiring preference for former Federal Division employees had expired, the former Federal Division employees were situated no differently from the general public, as far as hiring decisions were concerned. Thus, Federal Forge argues that the relevant group for *874 disparate impact comparison is the qualified labor pool in the Lansing area. There are two ways to make such a comparison. Neither, however, establishes disparate impact in this case. The approach most closely resembling that in Wards Cove would be to compare the 27 workers actually hired during the moratorium with the qualified labor pool (which would include the 107 former Federal Division employees). That comparison would show whether the percentage of selected applicants who were 40 or over was significantly lower than the percentage of qualified workers who were 40 or over. Nevertheless, even if a disparity were found, such a comparison would fail to establish any causal relationship between the disparity and the moratorium; it might show a differential pattern in hiring, but it would not explain how that effect was caused by the removal of the Federal Division employees from consideration for employment.

Another approach would be to compare the percentage of those 40 or over in the available labor pool, including the former Federal Forge employees, with the percentage of those 40 or over in the remainder of the qualified labor pool. In theory, once the 107 Federal Division employees were eliminated from consideration for the Lansing Division openings, that left the positions open to all others who wanted to apply. A statistic that showed the percentage of those 40 or over in the qualified labor pool minus the Federal Division employees to be significantly lower than the percentage in the qualified labor pool including the Federal Division employees would indicate that the moratorium could have had a disparate impact on older workers by removing a significant number of them from consideration.

However, that could be true only if the former Federal Division employees made up a significant portion of the total qualified labor pool. The burden, of course, would be on appellants to show that was the case. It is, in any case, extremely doubtful, since Federal Forge filled most of the openings with unskilled workers. The removal of 107 workers from the total pool of skilled and unskilled workers in the Lansing area could hardly have significantly altered the percentage of those 40 and over in the latter group.

In fact, appellants offered neither of the described comparisons. The statistical evidence offered by appellants fails to establish a prima facie case of disparate impact caused by the hiring moratorium because they compare the wrong groups. The disparity identified by appellants does not show that the moratorium had the effect of preventing Federal Forge from hiring a percentage of workers 40 and over that was representative of the qualified labor pool. The moratorium only removed 103 persons from consideration for the Lansing Division jobs, and only 46 persons who had reached the age of 40.

Appellants’ statistics fail to establish that the removal of a small number of potential applicants from a metropolitan labor pool had the effect of causing Federal Forge to hire a disproportionately low number of workers over 40. It could be that even after removing the 107 Federal Division employees from consideration, Federal Forge hired enough workers 40 and over from among the remaining labor pool to match the percentages in the composition of the pool itself. The burden was on appellants to show otherwise. Their statistics do not do so. Even if Federal Forge did not in fact hire a representative percentage of workers 40 and over—and, considering the evidence in the light most favorable to the appellants, the low number of such workers among the 27 actually hired suggests that it may not have done so 3 —appellants’ statistical comparison does not show that the moratorium or any other employer policy was the cause of the disparity.

Federal Forge apparently hired everyone who applied for a position at the Lansing Division plant during the moratorium. The percentage of workers 40 and over may simply reflect the ages of those *875 who applied. Moreover, when considering the group of 27 employees hired during the moratorium in isolation, the small sample size lessens the significance of the statistics obtained. Appellants’ statistics show that a group with a relatively higher percentage of workers 40 and over was barred from employment and a group with a relatively lower percentage of workers 40 and over was hired. But post hoc, ergo prop-ter hoc is not a rule of legal causation. At best, appellants’ statistics show that Federal Forge in fact hired fewer workers 40 and over than might have been expected had it hired only from among the Federal Division employees. But the mere presence of a disparity in an employer’s work force is not enough to establish a prima facie case of disparate impact. Disparities occur naturally in work forces for a multitude of non-discriminatory reasons. Recognizing that fact, disparate impact doctrine requires that a specific employment practice be shown to have caused the disparity. By not establishing a connection between the specific employment practice and the demonstrated disparity, appellants failed to meet the standards of a disparate impact case.

Legitimate business interest

An alternative ground for our conclusion that appellants failed to establish a prima facie case of disparate impact is Federal Forge’s articulation of a legitimate business reason for the moratorium and appellants’ failure to rebut that reason. Federal Forge explained that it imposed the moratorium because of the dispute with the UAW over whether the Federal Division employees hired beyond the twelvemonth preference period were entitled to retain their past seniority for pension purposes. Federal Forge took the position, later vindicated in arbitration proceedings, that they were not. The issue went to arbitration in March 1984, shortly after ap-pellee learned of the demand for past seniority. At that point, Federal Forge decided that rather than continuing to hire former Federal Division employees for the Lansing Division plant — a practice that was discretionary once the twelve-month preference period had ended — and risk having to pay the increased pension costs should the union win on the seniority issue at arbitration, it would cease to hire former Federal Division employees until the resolution of the dispute.

Thus, Federal Forge’s explanation for the moratorium is essentially economic. The statement of affirmative defenses in its answer to the complaint stated that the decision was “based upon possible increased personnel costs.” Hiring the Federal Division employees would subject Federal Forge to a contingent cost (the cost of the added pension benefits multiplied by the probability of losing the arbitration) in addition to the cost at which it had intended to hire workers for the Lansing Division plant. That the added cost was contingent makes it no less real as a matter of economic calculation for the employer. The business interest served by the moratorium was to avoid that contingent cost.

To qualify as serving a legitimate business interest, the challenged employment practice need not be “essential” or “indispensable” to the employer's business. Wards Cove, 109 S.Ct. at 2126. Minimizing the cost of labor is a legitimate business consideration. The imposition of the moratorium served, in a significant way, that legitimate interest by eliminating the contingent cost of added pension benefits for Federal Division employees.

Although it is a legitimate business consideration, an employment practice directed at minimizing the cost of labor can run afoul of age discrimination laws if it systematically and adversely affects older employees or employment applicants. Appellants challenge the legitimacy of the stated economic reason for the moratorium on the ground that the group affected by the moratorium was selected on the basis of a characteristic — seniority—that has been recognized as an illegitimate basis for employment decisions when it adversely affects older employees. In most employment situations, there is a natural correlation between seniority and age of employ *876 ees. Courts have recognized this correlation. 4

The employer’s setting of a wage rate is the most obvious type of “employment practice” that will affect the composition of a work force by influencing the pool of workers who actually apply. In general, a job that offers a relatively low wage will attract a higher proportion of workers that are inexperienced, with lower skills, and/or without family responsibilities. Each of these factors will tend to be correlated with age and will thus attract an actual pool of applicants that “underrepresents” older workers. However, it has never been contended that an employer must offer a wage for a job that is adequate to attract a pool of applicants representative of the entire labor force (or even the entire "qualified” labor force, which would be nearly the same for unskilled work). Even in the case of Metz v. Transit Mix, Inc., 828 F.2d 1202, 1208-10 (7th Cir.1987), where the Seventh Circuit held that an employer could not choose a younger, cheaper manager to replace an older, more expensive one, the court suggested that age discrimination could be avoided by allowing the older manager to compete equally for the job, at the lower wage rate 5 By offering the job to older employees on the same terms as younger, the employer would demonstrate that the employment practice is the cost-saving measure it purports to be and not a pretext for age discrimination.

The ADEA is directly aimed at the evil of taking age into account in making employment decisions, or using age as a proxy for some legitimate factor, with which it is somewhat, but not totally, correlated. In this case, the evil would arise if and only if the former Federal Division workers had offered to waive the outcome of the arbitration, thus placing themselves in exactly the same position as other workers, and Federal Forge had still refused to consider them for the positions. They did not do this, and the Federal Forge management can hardly be faulted for failing to make an explicit offer to that effect, since the offer itself could easily have been interpreted as an unfair labor practice, undertaken in an effort to split the union.

Alternative employment practice

Once the employer identifies a legitimate, non-discriminatory business reason for the employment practice in question, the burden shifts back to the plaintiff to show that the reason is pretextual or to show the existence of an alternative employment practice that reduces the disparate impact but serves the employer’s legitimate interests. Watson, 487 U.S. at 998, 108 S.Ct. at 2790 (citing Albemarle Paper Co. v. Moody; 422 U.S. 405, 425, 95 S.Ct. 2362, 2375, 45 L.Ed.2d 280 (1975)). Appellants do not argue that the stated reason for the moratorium is a pretext; rather, they suggest an alternative practice. The alternative employment practice was for Federal Forge to continue to hire the former Federal Division employees and take its chances on the arbitration of their past seniority claims. Then, if Federal Forge were justified in its belief that the employees were not entitled to past seniority, it would win at arbitration; if not, the employees would be entitled to the extra benefits they would receive at arbitration.

*877 Hiring the Federal Division employees while the issue of past seniority pension benefits was in arbitration would not have served Federal Forge’s legitimate interest in eliminating the contingent cost of added pension benefits. The alternative practice suggested by appellants therefore cannot be said to be “equally as effective as the challenged practice in serving the employer’s legitimate business goals.” Watson, 487 U.S. at 998, 108 S.Ct. at 2790.

Because Federal Forge showed a non-discriminatory legitimate business reason for the imposition of the hiring moratorium and because appellants failed to show that the stated reason was pretextual or that there existed another hiring practice that would have had a less adverse effect on older workers and would have served Federal Forge’s legitimate business interests, the grant of summary judgment to Federal Forge on this issue was proper.

Section 4(f)(2) of the ADEA

In the alternative, appellants maintain that Section 4(f)(2) of the ADEA (29 U.S.C. § 623(f)(2)) creates a direct prohibition on employers refusing to hire on the basis of a pension plan. The statute says:

It shall not be unlawful for an employer, employment agency, or labor organization ... to observe the terms of a bona fide seniority system or any bona fide employee benefit plan such- as a retirement, pension or insurance plan, which is not a subterfuge to evade the purpose of this chapter, except that no such employee benefit plan shall excuse the failure to hire any individual....

(Emphasis added.) Appellants argue that, in imposing the moratorium, Federal Forge violated the emphasized language in the statute by using a pension plan as an excuse not to hire.

The district court rejected appellants’ interpretation of the section, finding that the relevant phrase does not directly proscribe failures to hire on the basis of a seniority system, but that it only acts as a limitation on the defense that an employer was observing the terms of an employee benefit plan. Thus, the emphasized language in the statute would only come into play where the employer invokes § 623(f)(2) as a defense. Since Federal Forge did not attempt to defend the moratorium on the ground that it was merely observing the terms of an employee benefit plan, the district court reasoned, the statutory section was inapplicable to this case.

Since the district court’s decision, the Supreme Court has addressed that section of the statute in Public Employees Ret. Sys. of Ohio v. Betts, 492 U.S. -, 109 S.Ct. 2854, 106 L.Ed.2d 134 (1989). Appellants read Betts to state that § 623(f)(2) does create an affirmative prohibition against refusing to hire on the basis of an employee benefit plan. We do not agree. The issue in Betts was the meaning of the phrase “which is not a subterfuge to evade the purpose of this chapter.” The Court did not address whether § 623(f)(2) creates a direct prohibition on refusals to hire based on the terms of a benefit plan. Appellants cite language from the opinion which, taken out of context, seems to support their position, 6 but which was written in the context of justifying the Court’s holding that § 623(f)(2) helps to define the elements of a plaintiff’s prima facie case of *878 age discrimination. The decision in Betts does not disturb the validity of the district court’s interpretation of § 623(f)(2).

For the foregoing reasons, the district court's grant of summary judgment to Federal Forge is AFFIRMED.

. The Federal Division plant contained both a hot forge and a cold forge and only the hot forge closed at that time. A few Federal Division employees remained at the cold forge, which closed some months later.

. Federal Forge hired the four as part of a settlement with the union after the union filed an unfair labor practices charge with the National Labor Relations Board. Although one of the four was over 40, all had very low seniority at the Federal Division plant. Each had worked less than six years at the plant, while the average for all the workers was approximately 13.5 years.

. Nationally, about 40% of the work force is 40 years of age or over. U.S. Bureau of the Census, Statistical Abstract of the United States: 1988 (108th edition), p. 368.

. Metz v. Transit Mix, Inc., 828 F.2d 1202 (7th Cir.1987) (higher costs of older employee directly related to years of experience); Reichman v. Bonsignore, Brignati & Mazzotta P.C., 818 F.2d 278 (2d Cir.1987) (employee dismissed ten months before pension rights would fully vest); Dace v. ACF Industries, Inc., 722 F.2d 374 (8th Cir.1983) (senior employee demoted, rather than younger employee); Leftwich v. Harris-Stowe State College, 702 F.2d 686 (8th Cir.1983) (preference for retaining non-tenured over tenured professors in consolidation of two colleges); Geller v. Markham, 635 F.2d 1027 (2d Cir.1980) (cost-cutting policy of hiring only teachers below a certain experience level).

. This is consistent with the "least-detrimental-alternative" standard, applied by this court in a forced retirement situation. E.E.O.C. v. Chrysler Corp, 733 F.2d 1183 (6th Cir.1984). There, we held that the employer would have to give the employees it had subjected to forced early retirement (i.e., those over 55) the same right of recall in the event of improved business conditions that it gave to younger (under 55) employees, who had been permitted to accept "layoff status" rather than early retirement.

. The Court held that a plaintiff seeking to challenge an employee benefit plan as a subterfuge bears the burden of proving that the discriminatory provision of the plan actually was intended to serve the purpose of discriminating in some non-fringe benefit aspect of the employment relation. In dicta, the Court stated:

... 4(f)(2) is not so much a defense to a charge of age discrimination as it is a description of the type of employer conduct that is prohibited in the employee benefit plan context.

Despite the fact that § 703(h), like § 4(f)(2), appears on first reading to describe an affirmative defense, we have "regarded [§ 703(h) ] not as a defense ... but as a provision that itself ‘delineates which employment practices are illegal and thereby prohibited and which are not.' ”

Betts, 109 S.Ct. at 2868 (brackets in original) (citations omitted).