More than a half dozen years have passed since we last saw this case. It is a racial discrimination lawsuit arising out of a reduction in force at the Illinois Department of Employment Security (IDES) that the district court had dismissed on a grant of summary judgment to the defendants. We reversed and remanded the case for further proceedings. Council 31, AFSCME v. Ward, 978 F.2d 373 (1992). Those proceedings are now completed, the plaintiffs lost again, and today we consider their second appeal.
IDES is the administrative agency charged with operating the state’s unemployment insurance system. It is headquartered in Chicago and operates field offices throughout the state. IDES employs both full-time and intermittent employees; intermittent employees work only when called, on an as-needed basis and no more than 1,500 hours per year. But they do the same jobs and earn the same salaries and benefits as full-time employees, though pro rata depending on the number of hours worked. Both full-time and intermittent employees are represented by Council 31 of the American Federation of State, County and Municipal Employees, AFL-CIO, and bound by the collective bargaining agreement (CBA) between Council 31 and IDES.
In 1985 IDES decided to cut its work force (in litigation parlance, a RIF, meaning a reduction in force) by 25 percent. IDES further decided to target most heavily its Cook County offices, which coincidentally (or not) had a proportionately higher black employee population than its downstate offices. *1071Prior to the RIF, IDES’ total work force was about 40 percent black; the Cook County work force was about 60 percent black. IDES’ RIF strategy (which IDES followed) was to cancel overtime work and to get rid of intermittent employees before laying off full-time employees; that way, full-timers would lose their jobs only as a last resort. IDES identified the positions to be cut from each office’s work force but did not name the specific individuals who had to go. Indeed, the collective bargaining agreement between Council 31 and IDES had seniority and bumping provisions that made such identification impossible. The CBA set out an elaborate process whereby a targeted employee could save his or her own job by pushing the axe farther down the union’s seniority ladder. A targeted employee could bump (i.e., take the job of) a more junior employee in the same position classification either in the same office or, if there was no one in the same office, in the office’s “bump unit” (a cluster of offices in a localized region). If no bump was available in the same position classification, a targeted employee could bump into a lower position classification either in the same office or in the bump unit. Or a targeted employee who had been previously certified in any other position classification could bump into that classification either in the same office or in the bump unit. Finally, if no bump was available under any of these scenarios, the targeted employee could opt to take intermittent status in lieu of being laid off. The bumped employees could then exercise the same bumping rights and the whole process would start all over again until the work force had been shaved by 25 percent. All of this suggests that IDES really could not have known which specific employees would be left standing when the music stopped.
Before the RIF, IDES prepared an adverse impact analysis report as required by the Illinois Department of Human Rights. Actually, it prepared three reports; two comparing layoff rates among the races and one comparing retention rates among the races. IDES submitted the retention rate report to the IDHR, which raised no stink, and the RIF proceeded. When the dust settled, IDES had cut 114 employees from its Cook County starting lineup; 81 were black.1
The black employees sued, alleging both that IDES intentionally discriminated against them and that the RIF had a disparate impact on black employees. On the first go-round the district court dismissed the intentional discrimination claim for failure to state a claim. The court granted summary judgment for IDES on the disparate impact claim, reasoning that a disparate impact claim requires the identification of a “specific employment practice” and a single layoff decision cannot qualify as an employment practice. We reversed, finding that a single decision can give rise to both intentional discrimination and disparate impact claims:
[T]he district court was concerned that if single employer decisions are actionable then “every single act, intentional or not, which has an adverse impact on a protected class is actionable under Title VII.” But we fail to see the obvious evil in the court’s hypothesis. Intentionally discriminatory acts are of course actionable under Title VII, single or otherwise. As for unintentionally discriminatory acts that have an adverse impact on a protected class, they too are actionable under Title VII unless they have a business justification. But it is difficult to make out a prima facie showing of adverse impact: the affected group *1072may be too small for any valid statistical comparisons, and obvious or common nondiscriminatory reasons for the apparent disparities must be taken into account. These requirements immunize most single decisions from disparate impact challenges. To the extent that members of a protected class can show significant disparities stemming from a single decision, however, there is no reason that decision should not be actionable.
Ward, 978 F.2d at 378 (citations omitted). We held, in short, that the employees must at least be given a chance to prove that IDES’ single employment decision — the decision to lay off 25 percent of its employees— was discriminatory.
On remand, the district court (Judge Paul E. Plunkett) granted summary judgment to IDES on the intentional discrimination claim, finding the employees’ evidence could not, as a matter of law, establish that IDES intended to discriminate against the black employees. Judge Plunkett held a bench trial on the disparate impact claim, which boiled down to a battle of the experts. The employees’ expert, Professor Robert LaLonde, looked at the impact of the RIF on full-time employees and concluded that black employees were affected disproportionately. IDES’ expert, Professor John Donohue, looked at the impact of the RIF on all employees (full-time employees and intermittents) and concluded that the RIF affected blacks and non-blacks proportionately. Both experts offered statistical analysis to support their views. The judge ultimately found Professor Dono-hue’s statistical analysis to be more-credible and entered judgment for IDES.
On appeal, the employees argue the district court should not have entered summary judgment on the intentional discrimination claim because their evidence showed IDES intended to discriminate against black employees. They also argue the court should not have entered judgment for IDES on the disparate impact claim because they established that blacks were affected disproportionately by the RIF. Finally, they challenge two evidentiary rulings: the decision to admit evidence of IDES’ business justification for the RIF and the decision to exclude Professor LaLonde’s fifth expert report.
The first time this case came before us, we assumed the employees’ intentional discrimination claim challenged IDES’ initial decision to concentrate the layoffs in Cook County, where the employee population was significantly more black than white. And at oral argument, counsel for the employees confirmed that this was the crux of the intentional discrimination claim. But the employees offered almost nothing in the district court to suggest that the original decision to target Cook County was racially motivated. They hung their hat solely on IDES’ decision to submit to the IDHR the adverse impact report comparing retention rates instead of the reports comparing layoff rates. The employees theorized that the layoff rate comparison revealed an adverse impact on black employees, so IDES scrapped reports using those numbers to hide its discriminatory motive. The district court didn’t buy this theory, and neither do we. There was nothing inherently deceptive in the decision to use the retention rate analysis. In fact, the employees conceded in the pretrial order that IDES’ comparison of retention rates was consistent with EEOC guidelines, which use that method of analysis to calculate any adverse impact in employment terminations. Moreover, we don’t see how the use of retention rates could have disguised the true impact on black jobs. As Judge Cudahy noted in our previous decision in this case: “Mathematically speaking, retention rate analysis and layoff rate analysis are identical, like describing a glass half-full or half-empty.” 978 F.2d at 379 n. 3. These concerns aside, IDES’ decision to use the retention rate analysis simply does not demonstrate an intent to discriminate.
“Different kinds and combinations of evidence can create a triable issue of intentional discrimination .... ” Troupe v. May Dept. Stores Co., 20 F.3d 734, 736 (7th Cir.1994). Direct evidence — evidence that can be interpreted as an acknowledgment of discriminatory intent by the defendant or its agents — can do the trick. Id. But so can circumstantial evidence — “suspicious timing, ambiguous statements oral or written, behavior toward or comments directed at other employees in the protected group, and other bits and pieces from which an inference of *1073discriminatory intent might be drawn.” Id. “[P]ieces of evidence none conclusive in itself but together composing a convincing mosaic of discrimination” are enough to survive summary judgment. Id. at 737. IDES’ preparation of multiple adverse impact analysis reports shows, to be generous, that IDES knew a disproportionate number of black employees would be affected by the RIF. But “[kjnowledge of a disparity is not the same thing as an intent to cause or maintain it ....” American Nurses’ Ass’n v. State of Illinois, 783 F.2d 716, 722 (7th Cir.1986). Without more, the employees cannot show IDES intended to discriminate; the “mosaic” of circumstantial evidence just isn’t there. We therefore affirm the court’s entry of summary judgment on the intentional discrimination claim.
The disparate impact claim turned on whose statistical analysis the court accepted and, as we said, the court accepted Professor Donohue’s approach and entered judgment for IDES. We review that decision for clear error, giving due regard to the trial court’s opportunity to see and hear the witnesses. See Fed.R.Civ.P. 52(a); Piraino v. International Orientation Resources, Inc., 137 F.3d 987, 990 (7th Cir.1998).
The question before us is not whether the statistical evidence showed a disparate impact, but whether the court clearly erred in accepting Professor Donohue’s' statistical analysis. The parties agreed that if you look at all employees (full-time employees and intermittent employees), the RIF affected blacks and nonblacks proportionately. The parties also agreed that if you look only at full-time employees in Cook County, blacks were significantly more likely (statistically speaking) to be adversely affected. It is hardly surprising that IDES urged the court to accept the former approach and the employees urged the latter. Professor La-Londe opined that blacks were more likely than whites (LaLonde compared blacks to whites, while Donohue compared blacks to nonblacks) to be adversely affected by the RIF. He calculated that 9.9 percent of black employees were' adversely affected by the RIF, but only 3.2 percent of white employees were adversely affected. Yet he later admitted that his numbers were skewed because he included intermittent employees in the denominator of the percentage equation but did not include them in the numerator. La-Londe further conceded that black representation in full-time employee positions at IDES remained largely unchanged after the RIF. The full-time population was 45.9 percent black before the RIF and 44.8 percent black after the layoff. In his reports he opined that this fact did not necessarily undermine the disparate impact claim because IDES could simply have hired a bunch of new. black employees after firing the old black employees. At trial, however, he admitted that he had no evidence to'suggest IDES had done this, and, in fact, Professor Donohue testified that IDES had not done so.
Professor Donohue considered the effect of the RIF on all employees. He viewed full-time and intermittent employees in terms of “full-time equivalents” and then assessed the impact on full-time equivalents across racial lines. In his first report Professor LaLonde rejected the inclusion of intermittents; more accurately, he didn’t know how intermittents figured into IDES’ employment mix so he excluded them from his analysis. Later, after studying Professor Donohue’s analysis, he adopted the full-time equivalents approach. In fact, when asked at trial about his first report, LaLonde testified: “[Mjetho-dologically it’s sound but there is this institutional issue [the full-time equivalents concept] which I would have wanted to take into account were today June 1993 [the date he prepared the first report].” In a supplemental report, LaLonde opined that Donohue was right (i.e., there was no disparate impact) if you assumed two things: (1) that the total reduction in intermittent hours was caused by the RIF; and (2) that there were no- rises in full-time equivalent" intermittent hours among blacks to offset the reduction in full-time black employees. In other words, if the decrease in intermittent hours was caused by seasonality and if IDES hired new black intermittent workers to replace the full-time black workers it fired, Donohue’s approach would be- flawed. When tested at trial, however, LaLonde could give the court no reason to reject these assumptions.
At trial, Professor Donohue explained that he analyzed both layoff rates (the percent*1074ages of black and nonblack employees affected by the RIF) and retention rates (the percentage changes in black and nonblack employees working at IDES before and after the RIF). Either way, he concluded, the RIF affected black employees proportionately. . Professor Donohue explained that it was important to consider intermittents because they played such an integral role in IDES’ work force. The court found Dono-hue’s approach to be more credible because it saw no reason to exclude intermittents as Professor LaLonde had done. The record supports that decision, and we have no reason to second-guess it. See Rennie v. Dalton, 3 F.3d 1100, 1106 (7th Cir.1993) (“If there are two permissible views on the evidence, the district court’s choice between them cannot be clearly erroneous.”). Additionally, the court’s decision to look at the entire employee population was consistent with the Supreme Court’s analysis of relevant populations. See Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 653, 109 S.Ct. 2115, 104 L.Ed2d 733 (1989) (“Racial' imbalance in one segment of an employer’s work force does not, without more, establish a prima facie case of disparate impact with respect to the selection of workers for the employer’s other positions, even where workers for the different positions may have somewhat fungible skills .... ”). The district court’s decision to adopt Donohue’s statistical analysis was not clearly erroneous.
Finally, the employees challenge two of the district court’s evidentiary rulings: (1) the decision to admit evidence of IDES’ business justification for the RIF; and (2) the decision to exclude Professor LaLonde’s fifth expert report. We review both decisions .under an abuse of discretion standard, Mid-America Tablewares, Inc. v. Mogi Trading Co., 100 F.3d 1353, 1362 (7th Cir.1996); Grassi v. Information Resources, Inc., 63 F.3d 596, 603 (7th Cir.1995), meaning we reverse only if the decision was “fundamentally wrong.” Williams v. Chicago Bd. of Educ., 155 F.3d 853, 857 (7th Cir.1998). We cannot say that either decision was fundamentally wrong.
The employees objected to the introduction of business justification evidence because IDES failed to plead business justification as an affirmative defense in its answer. The district court admitted IDES’ business justification evidence over the employees’ objection because it did not believe the employees could have been surprised or prejudiced by IDES’ failure to plead business justification as a defense to the employees’ claims. Given the law in this area, we agree. Disparate impact analysis is a two-step process. First, the plaintiff must make a prima facie showing that the challenged employment practice — even if facially neutral — had a disparate impact on a protected class. Wards Cove, 490 U.S. at 657, 109 S.Ct. 2115. If the plaintiff clears this hurdle, the case shifts to any business justification the defendant offers for its use of the challenged practices. Id. at 658, 109 S.Ct. 2115. The employees should have known that business justification would be an issue and that such evidence would be coming down the pike.
Beyond that, the business justification issue played no role in the court’s ultimate disposition of the disparate impact claim. In its initial decision the -court found that the numbers revealed no disparate impact. The business justification issue came up later, when the employees moved to amend the judgment. The court denied the motion and, as an aside, noted that even if the employees were right, they would still lose:
Even if we had accepted the plaintiffs’ argument that the impact should be measured only by the loss of full-time jobs, they would not have fared any better. Defendant could not have been held liable under a disparate impact theory if it had a business justification for -effecting the RIF in the manner it chose. Defendant had such a justification: economic necessity.
March 25, 1998, order, at p. 6 (citation omitted). In other words, the district court found the employees never cleared step one of the disparate impact analysis; step two — business justification — didn’t even come into play.2 That finding was not fundamentally *1075wrong; rather, it was firmly grounded in the unequivocal trial testimony.
Nor was the decision to exclude La-Londe’s fifth report fundamentally wrong. After the district court denied the employees’ motion to bar IDES’ business justification evidence, the employees tried to admit a fifth report from Professor LaLonde. The employees claimed the report was necessary to rebut IDES’ business justification claim and could not have been filed earlier because they didn’t know IDES would offer evidence of a business justification. This argument is not convincing. The court’s decision to bar LaLonde’s report was reasonable because the employees had already filed four reports from Professor LaLonde and the fifth one contained no newly acquired information. Even on appeal the employees fail to explain how the fifth report would have rebutted IDES’ claim that the RIF was necessary because of cutbacks in federal funding. Finally, as we have noted, we view this entire business justification issue as little more than a red herring. The district court didn’t decide the case on that basis and the employees never seriously challenged the necessity of the RIF.
A lot of people — people of all races — lost their jobs in IDES’ reduction in force. Our decision is in no way intended to diminish the loss those people suffered because of IDES’ decision. But our role in this incident is limited: we ask not whether IDES’ layoff decision was fair, but whether the evidence showed it was based on race. See Carson v. Bethlehem Steel Corp., 82 F.3d 157, 159 (7th Cir.1996) (citing Pollard v. Rea Magnet Wire Corp., 824 F.2d 557 (7th Cir.1987)). Because the evidence falls far short of showing that IDES discriminated against black employees, we affirm the district court’s judgment.
. Determining the precise numbers of black and nonblack employees adversely affected by the RIF would seem to us to be central to the disparate impact claim. Yet the parties have mucked around so much with the statistics and calculations that it’s difficult to pin down the appropriate set of numbers to use. The parties' experts discussed the numbers imprecisely — often comparing apples and oranges to reach a desired conclusion. Professor John Donohue made a rather insightful comment at trial. He said: "[Tjhere is a danger in a disparate impact case of being able to just sort of carve up the numbers in a way that makes something look bad that won’t necessarily be the case if you have a more comprehensive view.” We add the converse: it’s also possible to pull the lens so far back that the picture becomes obscured. But these aren’t so much “dangers” as inevitabilities, where parties are attempting to paint the statistical evidence in the light most favorable to their respective claims. The court's task in such cases is to see through the math manipulations, select the best statistical approach, and interpret the numbers accordingly.
. The employees also argue that IDES’ business justification evidence did not support the courts conclusion that the RIF was justified by economic necessity. We do not share the employees' view of the business- justification evidence. IDES' witnesses testified that reductions in fed*1075eral funding and decreases in unemployment claims precipitated the RIF, and the employees offered nothing to refute that testimony. In fact, the employees never questioned the necessity of cutting the work force by 25 percent; they’ve questioned only the manner in which IDES accomplished that goal.