OPINION
MOORE, Circuit Judge.Robert Skalka, Joseph Balnites, William Ponsock, and Charles Conover sued their former employer, Fernald Environmental Restoration Management Corporation (“FERMCO”), and its parent company, Fluor Daniel, Inc., for age discrimination and breach of contract. A jury found for Skalka and Conover on the age discrimination claim and for all four plaintiffs on the breach of contract claim. The jury also awarded substantial damages. The district court accepted the jury’s verdict except for the finding in favor of Conover on the breach of contract claim, and the court entered judgment accordingly.
FERMCO appeals, arguing that virtually all of the jury’s decisions on liability and damages were not supported by sufficient evidence. Conover cross-appeals on the breach of contract claim. We AFFIRM the age-discrimination judgment for Skal-ka and REVERSE the age-discrimination judgment for Conover. We AFFIRM the contract judgment against Conover and REVERSE the contract judgments for the rest of the plaintiffs. We also conclude that there were errors in the plaintiffs’ favor in the calculation of damages, and, therefore, we must REMAND this case so the district court can determine an appropriate remittitur, and, in the event of Skal-ka’s failure to remit the amount determined, order a new trial on the issue of damages.
I. BACKGROUND
The four plaintiffs in this case worked at a site owned by the federal government in Hamilton County, Ohio, where various contractors once produced uranium and others now clean up the waste. FERMCO took over clean-up responsibilities in December 1992, expecting to spend about fifteen years on the project. The following August it laid off approximately 120 salaried employees (and sixty more who left voluntarily). According to an Inspector *419General’s report, this reduction in force was inefficient and did not substantially reduce the workforce, since many of the discharged employees were eventually replaced. FERMCO attributes some of the problems criticized by the Inspector General to the vagaries of the budget process and to the large sums it spent on severance packages in order to accomplish the reduction in force. The plaintiffs imply that such a blatantly ineffective “reduction” in force must have been a pretext for getting rid of unwanted employees.
When it decided to reduce its workforce, FERMCO developed a “forced ranking” process to decide whom to lay off. Employees were classified into “peer groups,” and supervisors for each group evaluated their subordinates on various skills in order to produce a ranked list. Top management officials decided which departments would lose employees, and the manager of each department used the forced rankings to decide which employee's would go. FERMCO assured its employees that the selection process would be fair and objective, that the company would try to find new jobs for people, and that an appeals process would provide for meaningful review of the decision to terminate any particular employee.
At the time of the layoffs, Skalka was 54 years old and had several years of experience at the site. ' He worked in the Remedial Support Operations Group (“RSO”), and he reported to Timothy Huey. His peer group included four other RSO representatives (“RSOs”) who reported to Huey; the others were all younger than Skalka. Although they were classified together, the RSOs worked in separate parts of FERM-CO, each providing support services to a particular department. By the time of this litigation, FERMCO claimed to have lost the rating forms for the rest of Skalka’s peer group. However, Skalka received a near-perfect score (4.74 out of 5.0), and his score was better than Jim Golden’s. J.A. at 499, 502.1
Despite Skalka’s superior rating and performance, he was laid off because the department to which he had been assigned no longer needed an RSO. No other RSOs in his peer group were laid off. Skalka claims that the RSO with the lowest rating should have been laid off and that he should have been moved to that person’s position. He notes that a year later, when the RSO position was eliminated entirely, FERMCO transferred the four younger employees to other jobs.
FERMCO responds that Huey was responsible for ranking Skalka’s peer group only because he was above them in the organizational hierarchy. He actually knew little or nothing about their performance, so FERMCO did not base any decisions on his rankings. In addition, Skalka may have anticipated the elimination of his position, since he worked on his résumé before the layoffs were announced.
■ Conover also had several years of experience at the FERMCO site and was 47 at the time of the layoffs. He introduced some testimony that he did a good job, but he was ranked last in his five-person peer group. Two members of the group were older than he was: Donald Liepold was 48, almost 49; Paul Simons was 47 but a few months older than Conover; Eric Meuser was 40; and Paul Sturgeon was 35. J.A. at 313-16, 701-02. The group was ranked by Gwen Nalls and Louis Henke. Among a number of irregularities in the rating process were numerous mathematical and typographical errors on this peer group’s rating forms. However, it appears that the only errors that actually affected the scores were in Conover’s favor.
*420Conover claims that Henke selectively solicited older employees for “voluntary” termination. He also claims that Henke is somehow implicated in another manager’s statement that some employees would be laid off because they were “too comfortable” in their jobs. Skalka Br. at 5. Conover argues that he should have been retained instead of several younger employees, including his peer group and his subordinates.
Ponsock and Balnites were also laid off after being ranked at the bottoms of then-peer groups.
Ponsock, Balnites, and Skalka all filed internal appeals of their terminations. They allege that the appeals process was a sham, consisted only of asking each appellant’s manager if correct procedures were followed, and did not result in overturning any layoff decisions.
FERMCO conducted a statistical study of the effects of the layoffs, and a preliminary report found “borderline” results in certain divisions, including Conover’s and Balnites’s, indicating a possible disparate impact on older workers. J.A. at 466-68. FERMCO’s expert witness testified that, upon further investigation, the result was found not to be statistically significant. J.A. at 742-43. The plaintiffs criticize this conclusion because it was based on the assumption that all groups should be equally affected, while they assert that older workers, because of their greater experience, should have been disproportionately unaffected by the layoffs. They offer FERMCO’s failure to follow up on this statistical red flag as evidence that the company’s violations of the Age Discrimination in Employment Act (“ADEA”) were willful.
All four' plaintiffs claim that FERMCO discriminated on the basis of age, in violation of state and federal law, and that FERMCO is hable for breach of contract and/or promissory estoppel for breaking its promise that the layoff procedures would be fair and allow for meaningful review. The contract claim and the federal discrimination claim were tried to a jury, which found that FERMCO had discriminated against Skalka and Conover and awarded them back pay and front pay. Because the jury also found that the violations were willful, the back-pay award was doubled. Although the jury rejected Pon-sock’s and Balnites’s age discrimination claims, it found for all plaintiffs on their contract/promissory estoppel claims and awarded damages. With some complications discussed below, the district court eventually entered judgment in accordance with the jury’s verdict, except that it granted FERMCO’s motion for judgment on Conover’s contract claim because he had failed to pursue an internal appeal of his termination.
FERMCO appeals, claiming that there was insufficient evidence to support the findings of liability, and insufficient evidence that any violations of the ADEA were willful. It also has several complaints about the damages calculations. Conover cross-appeals the judgment notwithstanding the verdict on his contract claim.
II. ANALYSIS
A. AGE DISCRIMINATION
A prima facie case of age discrimination consists of four elements: (1) the plaintiff was a member of the protected class, which is people at least forty years old; (2) he or she was discharged; (3) he or she was qualified for the position; and (4) he or she was replaced by a younger person. See Phelps v. Yale Security, Inc., 986 F.2d 1020, 1023 (6th Cir.), cert. denied, 510 U.S. 861, 114 S.Ct. 175, 126 L.Ed.2d 135 (1993). In the context of a reduction in force, the fourth requirement is modified to require the plaintiff to offer some “direct, circumstantial, or statistical evidence tending to indicate that the employer singled out the plaintiff for discharge for impermissible reasons.” Barnes v. GenCorp Inc., 896 F.2d 1457, *4211465 (6th Cir.), cert. denied, 498 U.S. 878, 111 S.Ct. 211, 112 L.Ed.2d 171 (1990). This evidence may include “showing that persons outside the protected class were retained in the same position.” EEOC v. Clay Printing Co., 955 F.2d 936, 941 (4th Cir.1992) (internal quotation omitted).
After a jury verdict, the burden-shifting framework falls away. The question for the court is simply whether there was sufficient evidence to support a finding of age discrimination. See Simpson v. Midland-Ross Corp., 823 F.2d 937, 942 (6th Cir.1987). However, the elements of that framework remain useful ways of thinking about the evidence and how thé jury might reasonably have arrived at its verdict. For example, the elements of the prima facie case, together with the jury’s disbelief of the employer’s explanation for its actions, are sufficient to support a verdict for the employee, although they do not compel such a verdict. See St. Mary’s Honor Ctr. v. Hicks, 509 U.S. 502, 511, 113 S.Ct. 2742,125 L.Ed.2d 407 (1993).
(1) ADEA Violations
(a) Skalka
There was sufficient evidence for the jury to conclude that FERMCO discriminated against Skalka. Skalka was the oldest member of his peer group and was laid off despite being rated the most competent. FERMCO’s explanations of this anomaly are (1) the ranking scores were not used in deciding which RSOs to eliminate because the supervisor who rated them did not actually know anything about their performance and (2) Skalka’s elimination was not really part of the general layoffs because the department to which he had been assigned did not need an RSO.
The jury was entitled to reject the credibility of FERMCO’s first explanation. FERMCO itself constructed the peer groups and assigned reviewers. If it is true that Huey was not qualified to rank the RSOs, then the jury may simply have taken that information as further evidence that FERMCO’s “objective” system of layoffs was a sham. FERMCO relied on the integrity of its forced ranking procedures to justify laying off the other plaintiffs (and many other employees). The jury could understandably have been skeptical when FERMCO conveniently identified flaws in the process that produced Skalka’s score.
To support its second argument, FERMCO cites our decisions holding that “no-bump” layoff policies and different treatment of differently situated employees generally are not sufficient to support an inference of discrimination. See Mitchell v. Toledo Hosp., 964 F.2d 577, 583 (6th Cir.1992); Barnes, 896 F.2d at 1464-66. Barnes held that the plaintiffs prima facie case is somewhat different in a case involving a workforce reduction. To justify an inference that “the most common legitimate reason” for termination was not the true reason, the plaintiff must present some evidence that the employer singled him or her out. See Barnes, 896 F.2d at 1465. Rather than showing that the plaintiff was replaced, “the fourth element [of the prima facie case] may be satisfied by showing that persons outside the protected class were retained in the same position or that there was some other evidence indicating that the employer did not treat age neutrally in deciding to dismiss the plaintiff.” Clay Printing, 955 F.2d at 941 (internal quotation omitted). The plaintiff cannot prevail merely by pointing to other positions for which she was qualified and claiming that the employer should have allowed her to “bump” the occupant of that position. See Barnes, 896 F.2d at 1465; see also Mitchell, 964 F.2d at 583. However, the employer cannot avoid liability by changing job titles or relying on minor differences in an attempt to argue that other employees did not have the “same” job. See Barnes, 896 F.2d at 1465 n. 10.
In this case the jury could reasonably have concluded, as FERMCO would have had them do, that each RSO’s job was *422unique, so that Skalka could not compare himself to the four other members of his peer group. However, this conclusion was not required, and again the jury may have been suspicious of the defendants’ insistence that their own system for planning the layoffs was inappropriate for Skalka. Skalka and the other RSOs had the same title and the same supervisor. FERMCO determined that their jobs required the same skills when it assigned them to a single peer group. To say that an exception was warranted because Skalka’s particular assignment was unnecessary is not a very convincing argument in the context of a large-scale layoff. If a company decides to eliminate scores of employees, it must necessarily decide that some jobs are unnecessary. The company could have chosen to structure the layoffs by identifying the least necessary jobs and eliminating their incumbents, as the employer did in Barnes. FERMCO, however, chose to proceed on the basis of employee skills. The jury did not have to accept as credible its explanations of why it deviated from its own procedure in Skalka’s case. Having concluded that FERMCO’s reason for laying off Skalka was not worthy of belief and having heard .evidence that FERMCO “lost” the peer group’s ranking forms, deviated from its normal procedures, and fired the oldest and most qualified RSO, the jury was entitled to find that discrimination had occurred. See St. Mary’s, 509 U.S. at 511.
(b) Conover
The jury also returned a verdict in favor of Conover on his age discrimination claim. Although Conover introduced what he claimed was evidence that Henke was age-biased, we can only attribute the verdict to confusion. Conover received the lowest score in his peer group, although he did introduce evidence at trial that he did a good job. He also introduced evidence of extreme carelessness with the ranking forms. As discussed above, St. Mary’s permits an inference of discrimination when a plaintiff has made out a prima facie case and the employer’s explanation of its action makes no sense. It does not, however, permit an inference of discrimination merely because the explanation makes no sense and the plaintiff is a member of a protected class. In permitting the inference of discrimination, St. Mary’s assumed that the prima facie case had been satisfied — in particular, that the plaintiff had been replaced by or otherwise treated differently from an employee outside the protected class. Regardless of whether Con-over deserved to be fired, it is hard to see how the jury could infer age discrimination when he was younger than two of the four other people in his peer group.
Of course, Conover was treated differently from the two employees who were substantially younger than he was, and in practical terms he was the same age as Liepold and Simons. We do not suggest that a plaintiff cannot establish age discrimination when only one of several older employees is treated differently from younger employees. Indeed, Conover could make out a prima facie case by comparing himself to the younger members of the peer group. On the ultimate question of whether he has proven discrimination, however, Conover’s evidence is too weak to sustain the verdict. It consists of (1) the fact that Henke discussed the possibility of early retirement with older employees but not younger ones, and two employees in their 50s accepted early retirement; (2) a statement by another supervisor that some employees would be laid off because they were “too comfortable”; and (3) the assertion “the oldest” of that supervisor’s employees were laid off. Skalka Br. at 5. Conover somehow attributes facts (2) and (3) to Henke. These allegations are vague and are unsupported by citations to the record. In addition, we do not believe that the “comfortable” comment is indicative of age bias. See Clay Printing, 955 F.2d at 942.
Most problematic is that Conover relies heavily on the false assertion that he was *423the oldest member of his peer group-. Skalka Br. at 18. Thus, Conover has not really provided any argument to support the verdict. The most likely explanation for it seems to be that the jury made the same mistake that Conover has made-about the ages of his peers and, finding FERMCO’s decision irrational and careless, inferred discrimination. We do, not think that a reasonable jury could have reached -this conclusion on the basis of the correct facts about Conover’s age relative to his peer group.
(2) Willfulness
The ADEA provides for additional liquidated damages if an employer’s violation of the Act is “willful.” See 29 U.S.C. § 626(b). As a result of the jury’s finding of willfulness in this case, the district court doubled the back-pay awards. FERMCO argues that there was insufficient evidence to support the willfulness finding, even assuming a violation of the ADEA.
To find willfulness, the jury had to conclude by the preponderance of the evidence that FERMCO “either knew or showed reckless disregard for the matter of whether its conduct was prohibited” by the ADEA. McLaughlin v. Richland Shoe Co., 486 U.S. 128, 133, 108 S.Ct. 1677, 100 L.Ed.2d 115 (1988). It is not enough to show that the employer knew that the ADEA Was “in the picture” or that the employer “acted without a reasonable basis for believing that it was complying with the statute.” Id. at 132-34 (rejecting these two standards).
To demonstrate willfulness, the plaintiffs rely mostly on FERMCO’s failure to follow up on the statistical survey indicating a possible disparate impact. That argument applies primarily to Conover, not Skalka, since Skalka’s department was not one of the ones flagged by the survey. The plaintiffs also criticize FERMCO for “failing to take a proactive stance in preventing discrimination.” Skalka Br. at 29. In addition, Skglka’s discharge involved a dramatic departure from established company procedures and was therefore highly “visible” to top management. The jury could reasonably have concluded that FERMCO was aware that its oldest RSO was selected for a layoff through irregular procedures and that its failure to investigate and remedy the situation indicated reckless disregard for whether its managers were making discriminatory decisions. We therefore conclude that there was sufficient evidence to support the jury’s finding that FERMCO’s discrimination against Skalka was willful.
B. CONTRACT/PROMISSORY ESTOPPEL
Ohio law does not support recovery for breach of contract or promissory estoppel in this situation. Ohio adheres to the doctrine that employment relationships are generally terminable at will but recognizes that an at-will-relationship can be modified by contract or promissory estoppel. See Mers v. Dispatch Printing Co., 19 Ohio St.3d 100, 483 N.E.2d 150, 153-55 (Ohio 1985); see also Wright v. Honda of America Mfg., Inc., 73 Ohio St.3d 571, 653 N.E.2d 381, 386 (Ohio 1995) (Pfeifer, J., concurring) (recognizing a “limited implied contract” based on employer’s multiple assurances that employee would not be fired for violation of anti-nepotism policy).2 However, an employer’s general policy statements, such as would appear in an employee handbook, are usually not sufficient by themselves to create an implied contract of continued employment. See Rudy v. Loral Defense Sys., 85 Ohio App.3d 148, 619 N.E.2d 449, *424452 (Ohio Ct.App.), cause dismissed, 66 Ohio St.3d 1478, 612 N.E.2d 329 (Ohio 1993); Bartlett v. Daniel Drake Mem’l Hosp., 75 Ohio App.3d 334, 599 N.E.2d 403, 406 (Ohio Ct.App.1991). Similarly, only a specific promise of job security — not generalized praise of an employee’s performance and prospects for advancement-can give rise to estoppel. Compare Heir mick v. Cincinnati Word Processing, Inc., 45 Ohio St.3d 131, 543 N.E.2d 1212, 1216-17 (Ohio 1989) (holding that estoppel applied because employer made specific individual promises to dissuade employee from pursuing other opportunities), with Wing v. Anchor Media, Ltd. of Texas, 59 Ohio St.3d 108, 570 N.E.2d 1095, 1098-99 (Ohio 1991) (holding that promises regarding “future benefits or opportunities” were not sufficient to constitute specific promises of continued employment), and Anders v. Specialty Chem. Resources, Inc., 121 Ohio App.3d 348, 700 N.E.2d 39, 41 (Ohio Ct. App.1997) (holding that “statements assuring [employee] of a ‘secure future’ and ‘secure career’ ” did not “establish express or implied contracts or promissory estop-pel”). In particular, assurances that the employer will be “fair” in its decisions do not constitute a specific guarantee of continued employment. See Snyder v. Ag Trucking, Inc., 57 F.3d 484, 489 (6th Cir. 1995) (holding that promises, including a statement that “ ‘all employees are treated fairly,’” were “not sufficiently specific to induce reliance”); Hanly v. Riverside Methodist Hosps., 78 Ohio App.3d 73, 603 N.E.2d 1126, 1129 (Ohio Ct.App.1991) (holding that stated policy of being “ ‘frank, fair and honest’ ” was not specific enough to create an implied contract).
Although FERMCO certainly wanted employees to believe that the layoff process would be fair, there were no specific promises of continued employment, even for people who were doing a good job. It would have been unreasonable for any employee to treat a promise of fairness in the implementation of mass layoffs as an assurance of continued employment. One would not expect any employer to make layoff decisions randomly, and a good employer would want to explain its plans and methods to its employees. It would strive to be fair and objective, just as it would strive to be fair and objective in the hiring process, when there are often more strong applicants than there are positions. Permitting employees to sue over the outcome of the layoffs on the basis of such general statements by the employer would make virtually every large-scale layoff judicially reviewable. While parties may contract to limit the employer’s ability to discharge, and compliance with the contract can ultimately be determined in court, we will not infer the existence of such a contract whenever an employer establishes a system for internal decision-making.
Because we reject the contract claims brought by all of the plaintiffs, we need not review the district court’s additional reason for setting aside the verdict for Conover on this claim.
C. DAMAGES
The jury awarded Skalka $487,907 in back pay and $529,416 in front pay for his ADEA claim, as well as $225,000 for his contract claim. FERMCO has several complaints about these damages. As an initial matter, we reject its claim that the jury’s decisions reflect “passion and prejudice.” The jury found against the plaintiffs on two counts (Balnites’s and Ponsock’s ADEA claims), and by FERMCO’s own arguments the inflated damages awards are the result of errors rather than unprincipled anger at FERMCO.
In the absence of undue passion and prejudice on the part of the jury, we review for abuse of discretion the district court’s refusal to grant a new trial based on excessive damages or a remittitur. See Roush v. KFC Nat’l Mngmnt. Co., 10 F.3d 392, 397 (6th Cir. 1993), cert. denied, 513 U.S. 808, 115 S.Ct. 56, 130 L.Ed.2d 15 (1994). A verdict is not exces*425sive unless it exceeds “ ‘the maximum that a jury could reasonably find to be compensatory’ for the plaintiffs loss.” Id. (quoting In re Lewis, 845 F.2d 624, 635 (6th Cir.1988)). Here, the district court refused to reduce the damages because the plaintiffs were able to offer a “tenable” explanation for the amounts. J.A. at 118 (3/24/97 Order) (relying on plaintiffs’ calculations at J.A. 87-89). As FERMCO demonstrates in its brief, the jury clearly arrived at Skalka’s back-pay figure by adding the stipulated amounts of salary and non-pension benefits to his projected pension without accounting ' for the amounts Skalka had received in early retirement benefits and in income from other employment. The plaintiffs’ explanation for the damages incorporated the jury’s clearest error, the failure to discount future pension benefits to present value. We therefore find it necessary to remand for the district court to calculate an appropriate remittitur and, if necessary, hold a retrial on the damages issue.
(1) Pension Benefits as Part of Back Pay
Over half of Skalka’s “back-pay” award was compensation for pension benefits he eventually would have received if he had stayed at FERMCO until retirement. FERMCO argues that, by definition, payments expected in the future cannot be part of back pay.3 The plaintiffs do not attempt to defend the inclusion of future payments in the back pay award, arguing only that FERMCO failed to object to jury instructions that defined back pay as including “pension benefits which a plaintiff would have received had the discrimination not occurred.” J.A. at 796. FERMCO responds that the parties intended this to refer only to 401(k) payments that would have been made during the time before the trial.
The conversation at the charge conference clearly treated 401(k) payments as “part of the pension benefits,” J.A. at 780 (emphasis added), but it is not clear whether the parties were treating the future payments as back pay. The plaintiffs’ closing argument discussed back pay, front pay, pension benefits, and 401(k) payments, in that order. J.A. at 786-87. The verdict form, however, had space only for back pay and front pay, leaving the jury to classify pension and 401 (k) benefits as it saw fit. Although plaintiffs’ counsel at one point referred to the claimed future pension benefits as “lost wages,” J.A. at 787, this apparent slip of the tongue did not suffice to put FERMCO on notice that all pension benefits would be considered back pay. We conclude that FERMCO did not forfeit this issue and that only the portion of the damages that truly represents “back pay” should be doubled. Because $269,353 of the jury’s “back-pay” award was compensation for Skalka’s expected pension, that amount should be treated as front pay rather than back pay.
(2) Speculative Nature of Pension Benefits
FERMCO also argues that there was insufficient evidence to support the award of any pension benefits because those benefits were too speculative. FERMCO’s argument on this point focuses on Conover, who in light of our holdings above is not entitled to any damages. FERMCO points out that the project was expected to last only ten to fifteen years and therefore would have ended before Conover turned 65. It claims that after the project is completed “ ‘all personnel will move on to other employment or retirement.’ ” FERMCO Br. at 42 (purporting to quote J.A. at 302). However, the document it cites for this proposition is rather vague. J.A. at 302 (“[W]hether such employees [will] have a 'continued presence at the [site] will depend upon the availability of other work for which they are qualified.”).
*426Skalka was several years older than Conover and would have retired within the projected lifetime of the project. FERM-CO’s arguments that the pension component of the damages was speculative do not apply to him, and the jury was within its rights to award him damages for his lost pension. Thus, except as discussed in the next section, this portion of the award may stand as a component of front pay.
(3) Failure to Discount the Pension to Present Value
Perhaps because it erroneously treated the pension benefits as back pay, the jury failed to discount them to present value. The plaintiffs’ justification of the damages to the district court made the same error, and they do not attempt to defend the error on appeal.
We have in the past been vigilant in correcting juries’ failures to discount damages to present value. See, e.g, Rodgers v. Fisher Body Div., Gen. Motors Corp., 739 F.2d 1102, 1105-06 (6th Cir.1984) (raising this issue sua sponte), cert. denied, 470 U.S. 1054, 105 S.Ct. 1759, 84 L.Ed.2d 821 (1985). We therefore instruct the district court to apply an appropriate discount rate to the pension benefits in calculating the remittitur and carefully to instruct the jury on this issue if there is a re-trial.
(4) Failure to Reduce Back Pay to Account for Other Income
FERMCO claims that the jury unreasonably failed to subtract Skalka’s early retirement benefits and money he earned through other employment from his back-pay award.
Skalka’s back-pay award— $218,554 once the future pension benefits are re-classified as front pay — compensated him for his lost salary and benefits up until the time of the verdict.4 FERMCO argues that this award should be reduced because, several months after his termination, Skalka secured a higher-paying position with another company. Skalka had a duty to mitigate his damages by seeking new employment, but the burden of proving reductions in Skalka’s damages is on FERMCO. See Jackson v. City of Cooke-ville, 31 F.3d 1354, 1359 (6th Cir.1994). Therefore, the remittitur calculated on remand should reflect Skalka’s new employment to the extent that any rational jury would have found this income to be proven. We note that Skalka is not entitled to back pay for any period in which he earned an equal or higher salary than he would have earned at FERMCO, see EEOC v. New York Times Broad. Serv., Inc., 542 F.2d 356, 359 (6th Cir.1976), but that his “excess” earnings are not to be subtracted from the back-pay award for the period of unemployment. See Matthews v. A-l, Inc., 748 F.2d 975, 978-79 (5th Cir.1984); see also Darnell v. City of Jasper, 730 F.2d 653, 656-57 (11th Cir.1984); Leftwich v. Harris-Stowe State College, 702 F.2d 686, 693-94 (8th Cir.1983).
FERMCO also argues that Skalka’s back-pay award should be reduced because he received early retirement benefits during the time between his termination and the trial. Such benefits, however, may constitute payments from a collateral source that are not properly subtracted from the damages awarded to a discrimination plaintiff. See Hamlin v. Charter Township of Flint, 165 F.3d 426, 432-36 (6th Cir.1999). In Hamlin, we set out a test for determining whether pension benefits are collateral. See id. at 435. On remand, the district court should apply Hamlin to determine whether Skalka’s retirement benefits received up until the time of trial should be subtracted from his back-pay award.
(5)Front Pay
FERMCO asserts in its brief that Skal-ka’s front-pay award should be vacated:
*427Under the ADEA, “a substantial liquidated damage award may indicate that an additional award of front pay is inappropriate or excessive.” Weaver v. Amoco Prod. Co., 66 F.3d 85, 89 (5th Cir.1995). But at the same time that the jury was giving Skalka and Conover inflated liquidated damage awards, it was also giving them huge front-pay awards. These front-pay awards ... should be vacated.
FERMCO Br. at 45. This argument is no longer applicable because our decision classifying Skalka’s pension as front pay will have the effect of substantially reducing the “liquidated damages” (i.e., the doubled back pay to which Skalka is entitled because the jury found that FERMCO discriminated willfully). The remainder of this section of FERMCO’s brief is devoted to Conover’s claim, and the arguments are not applicable to Skalka. As FERMCO acknowledges, this circuit has held that the amount of front pay is for the jury to decide, see Roush, 10 F.3d at 398, and FERMCO has not pointed to any identifiable flaw in the jury’s determination. Thus, when the district court determines the remittitur, it should allow the front-pay award to stand as determined by the jury, adjusted consistent with this opinion.
(6) Overlap Between ADEA and Contract Damages
FERMCO claims that it made no sense for the jury to award contract damages, since Skalka was fully compensated for his lost pay (past and future) through his ADEA damages. Although we have held that Skalka’s contract claim fails as a matter of law, the plaintiffs’ calculation of the damages makes the not unreasonable assumption that the jury simply chose to “allocate” some of the total damages to the contract claim. J.A. at 87-89. Therefore, in determining the remittitur, the district court should start with the total amount awarded, rather than just throwing out the contract damages. For example, under both claims the plaintiffs asked the jury to award damages for lost 401(k) payments. The damages figures for the ADEA claims suggest that the jury did not include 401(k) damages in its back-pay calculations, but it may have taken the 401(k) payments into account in the damages it allocated to the contract claim. The court’s duty in calculating a remittitur is to determine the maximum amount the jury could reasonably have awarded. See Katch v. Speidel, Div. of Textron, Inc., 746 F.2d 1136, 1144 (6th Cir.1984). Therefore, the remittitur should assume that the jury awarded damages for 401(k) payments at the rate requested by Skalka and that Skalka is entitled to 401(k) damages whether or not the jury lumped them into its contract claim award..
(7) Summary of Damages Issues
In summary, the district court’s determination of the remittitur should take into account the following. Skalka’s back-pay award properly includes the stipulated value of Skalka’s salary and benefits less severance pay ($218,554), reduced to the extent required by his other employment and early retirement benefits. The front pay award may stand as determined by the jury, increased by an amount to compensate for Skalka’s lost pension ($269,353 in future payments, which should be reduced to present value). In addition, the damages allocated by the jury to Skalka’s contract claim may be re-allocated if appropriate to either back pay or front pay, to the extent that a reasonable jury could have awarded more on the ADEA claim. Finally, the back-pay award is doubled due to the finding of willfulness.
D. POST-JUDGMENT INTEREST
The parties dispute whether post-judgment interest on the jury’s awards accrued from the date on which the district court entered an initial judgment memorializing the verdicts or from the later date on which the court entered a final, appeal-able judgment disposing of all claims. This question of statutory interpretation is *428one of first impression in this court and, so far as we are aware, in any federal court. We hold that interest runs from the date of the initial judgment.
On August 12, 1996, the district court entered an order, which it titled “Judgment Order,” awarding judgment and damages on the ADEA and contract claims. The court, however, specifically reserved judgment on the state age discrimination claims, which had not been given to the jury. A later order disposed of post-trial motions without mentioning the state claims. FERMCO appealed, and Conover filed an untimely cross-appeal. To save the cross-appeal, the plaintiffs persuaded the district court to enter a new judgment order, on the theory that the previous orders were not final because the state-law claims were still pending. In its new order, dated January 23, 1998, the district court specified that its prior orders should be deemed “final” for most purposes. Although the district court did not specifically address the issue of post-judgment interest, the parties agree that the district court intended to require FERM-CO to pay interest from the date of the entry of its partial order on August 12, 1996.
Post-judgment interest is calculated “from the date of the entry of the judgment.” 28 U.S.C. § 1961(a); see Kaiser Aluminum & Chem. Corp. v. Bonjomo, 494 U.S. 827, 835, 110 S.Ct. 1570, 108 L.Ed.2d 842 (1990) (refusing to calculate interest from date of jury verdict). FERMCO relies on the language of Federal Rule of Civil Procedure 54, which says:
(a) ... “Judgment” ... includes a decree and any order form which an appeal lies.
(b) ... When more than one claim for relief is presented ... or when multiple parties are involved, the court may direct the entry of a final judgment as to one or more but fewer than all of the claims or parties only upon [certain express findings and directions]. In the absence of such determination and direction, any order or other form of decision, however designated, which adjudicates fewer than all the claims ... shall not terminate the action as to any of the claims or parties, and the order or other form of decision is subject to revision at any time before the entry of judgment adjudicating all the claims....
Fed. R. Civ. P. 54 (emphasis added). Contrary to FERMCO’s argument, this language does not require the conclusion that any order that is not a final, appealable judgment is not a “judgment.” Rather, the use of “includes” and the selective use of the adjective “final” suggests that there can be “judgments” that are not “final, appealable judgments.” This conclusion is consistent with common usage. For example, when a court grants summary judgment to a defendant on some claims but holds a trial on others, its entry of judgment after trial does not typically refer to the earlier claims — there has already been a “judgment” on them. Rule 54 therefore does not definitively answer the question of whether post-judgment interest begins to accrue when the district court enters a partial judgment.
The Supreme Court’s decision in Kaiser decided that the plain language of § 1961 required interest to be calculated from the date of the judgment rather than the date of the verdict. However, the Court also faced a question not directly answered by the plain language of the statute. In Kaiser, the district court set aside the jury’s initial findings on damages, holding that they were not supported by the evidence. After a limited re-trial, a new judgment was entered. Kaiser, 494 U.S. at 830. In deciding which judgment triggered the accrual of post-judgment interest, the Court looked at when “damages [had] been ‘ascertained’ in [a] meaningful way.” Id. at 836 (citing Foleto v. Consolidated Rail Corp., 826 F.2d 1270, 1280 (3d Cir.1987)). It concluded that because the first judgment was not supported by the evidence, it did not represent a meaningful assessment of the damages. Interest therefore ran *429from the date of the entry of the second judgment. See Kaiser, 494 U.S. at 836.
We have applied Kaiser to hold that post-judgment interest runs from the date of any judgment that is not entirely set aside. Compare Adkins v. Asbestos Corp., Ltd., 18 F.3d 1349, 1351-52 (6th Cir.1994) (holding that post-judgment interest ran from date of judgment after remand, not judgment that had been reversed on appeal), with Coal Resources, Inc. v. Gulf & Western Indus., Inc., 954 F.2d 1263, 1273-75 (6th Cir.1992). In Coal Resources, we disagreed with the damages that were awarded, but the prevailing party accepted a remittitur. See id. at 1273-74. Despite the dramatic difference in the size of the award ($226,563 instead of $8,999,542), we held that damages had been “sufficiently ascertained” on the date of the initial judgment because “the remittitur merely reduced the damages by a distinct amount easily determined from the facts of the case.” Id. at 1275; see also Brocklehurst v. PPG Indus., Inc., 907 F.Supp. 1106, 1109 (E.D.Mieh.1995) (applying same rule when district court reduces damages through remittitur). These cases, however, do not address the issue of whether a partial, non-appealable judgment triggers the accrual of post-judgment interest.
We believe that the better rule is for plaintiffs to be entitled to post-judgment interest from the date of entry of the initial, partial judgment on the ADEA and contract claims, even though that judgment was not yet appealable. This holding is consistent with the text of § 1961 and the common meaning of the word “judgment.” We recognize that it is potentially burdensome to defendants. If a district court enters partial judgment for a plaintiff on one claim but takes a long time to resolve others, the defendant will have to choose between paying the judgment immediately (long before having a chance to appeal) or allowing interest to accrue. On the other hand, postponing the accrual of interest would deprive prevailing parties of the time value of their money. We resolve this dilemma by assuming that Congress has set the rate of post-judgment interest only to compensate for delay, without seeking to punish the losing party. See Kaiser, 494 U.S. at 835-36 (stating that purpose of post-judgment interest is compensatory). Thus, the defendant will not be unfairly burdened by the accrual of interest because it will continue to have the use of the money.
III. CONCLUSION
We AFFIRM the judgment for Skalka on his ADEA claim but REVERSE the ADEA judgment for Conover. On the contract claims, we AFFIRM the- judgment against Conover and REVERSE the judgments for Skalka, Balnites,. and Pon-sock. We REMAND Skalka’s ADEA claim to the district court for determination of an appropriate remittitur and, if necessary, a new trial on damages.
. The plaintiffs assert, without citation, that Skalka had the highest score and Golden the lowest. We note that the plaintiffs’ statement of the facts, which is over seven pages long, contains exactly two citations to the record. This paucity of record and Joint Appendix citations fails to comply with our rules, see Fed. R.App. P. 28(a)(7) & (9) (formerly Fed. R.App. P. 28(a)(4) & (6)), and unnecessarily complicates our job on review.
. In Ohio, the court’s syllabus is generally the official statement of the holding of the case. See Ohio S.Ct. R. for Reporting Ops. 1. In this case, however, the syllabus does not discuss the issue of how an at-will employment relationship can be modified, and no opinion won the support of a majority of the court. We cite Justice Pfeifer’s concurring opinion, because it reached the result on narrower grounds than those relied on in the lead opinion.
. This issue is important to FERMCO because the back-pay award but not the front-pay award is doubled when the jury finds that the defendant’s violation of the ADEA was willful.
. This figure reflects the stipulated value of Skalka's salary and benefits from the time of his termination until the time of the verdict, minus the amount he received as severance pay-