Throughout the decade of the 1990s, corporate downsizing was a popular strategy for companies that believed they had become indolent, complacent, inefficient, or otherwise unsuited to the ever-increasing pace of competition in their markets. Ameritech Corporation was no exception. With the advent of more competition in the telecommunications market and the promise of much more change to come in the near future, see Goldwasser v. Ameritech Corp., 222 F.3d 390, 392-393 (7th Cir. 2000), its management came to the conclusion in 1992 that drastic measures were necessary if it was to succeed in its new environment. And drastic measures were taken. Both Ameritech Services, Inc. (ASI), a company jointly owned by Ameritech's five independent operating companies, and Indiana Bell Telephone Company, one of those operating companies, slashed their middle management ranks dramatically as part of a comprehensive restructuring and reduction in force.

The two cases we have consolidated for decision today both arise out of these measures. In Adams v. Ameritech Services Inc., Nos. 98-1506, 98-2259, and 98-2307, the nineteen plaintiffs remaining on appeal claim principally that both ASI and Indiana Bell discriminated against them on the basis of age, in violation of the Age Discrimination in Employment Act (ADEA), 29 U.S.C. sec.sec. 621 et seq., and that both companies violated their rights under sections 502 and 503 of the Employee Retirement Income Security Act (ERISA), 29 U.S.C. sec.sec. 1132, 1133. In Allard v. Indiana Bell Telephone Co., Nos. 98-2426 and 98-2522, another set of 35 plaintiffs (plus two would-be plaintiffs to whom the district court denied permission to opt in to the case) claim the identical violations of the law from the same basic course of conduct by Indiana Bell alone; five of the Allard plaintiffs also raise constructive demotion claims. We do not fully understand why the district court agreed to handle these intertwined cases as separate matters, but, at least for purposes of appeal, they are functionally identical. Unless the context requires otherwise, our discussion therefore applies with equal force to both cases.

With the exception of some incidental issues, the district court disposed of all the claims in a series of orders granting summary judgment for the defendants. It then certified those orders as final for purposes of appeal under Fed. R. Civ. P. 54(b). While we appreciate the herculean efforts the district court made to wade through the voluminous materials on summary judgment that both sides presented, we conclude that the plaintiffs presented enough evidence to withstand the defendants' motions. We therefore reverse and remand for further proceedings.

Ameritech is one of the regional telephone companies that was created with the 1984 break-up of the old American Telephone & Telegraph Company system. It became the parent company of the five Bell operating companies in Indiana, Illinois, Michigan, Ohio, and Wisconsin. Those companies in turn created and now own ASI in equal shares. ASI exists to perform services for the operating companies, including "access services" (marketing, sales and service to long distance companies), billing, finance, human resources, information technology, and so forth. Ameritech itself sponsors and administers the Ameritech Management Pension Plan (the Plan), in which employees of all the operating companies and ASI participate. (Ameritech's subsidiaries have some responsibility for administering the Plan.)

Through 1992, both Indiana Bell and ASI considered as part of their "management" workforces all salaried, non-union employees, from secretaries to company officers. Managers were divided into Salary Grades (SG), which reflected knowledge, skill level, and degree of accountability. Jobs with duties that were entirely pre-set were classified as SG1, while jobs with complete independence in achieving goals within company and budgetary constraints were in the higher SG levels, for example, SG5 and above at Indiana Bell. (Administrative and secretarial assistants were classified in a slightly different system.) Not all jobs within a particular salary grade were the same, however, except insofar as the grade reflected increasing degrees of decisionmaking authority.

Richard Notebaert became ASI's president in June of 1992, at a time when Ameritech as a whole was seeking to improve its competitiveness. Shortly before Notebaert took over his new job, Ameritech's operating committee began to discuss what it perceived as a surplus of employees, and it began to explore with ASI what to do about it. It contacted the other Bell operating companies and concluded that they too had a surplus. Both ASI and Indiana Bell (along with the other operating companies) decided that they had to undertake a significant management workforce reduction--in plain English, they had to get rid of large numbers of managers, either by persuasion or by force. We describe here the process that the companies used and the structure of the Plan; we save until later a consideration of the arguments that age discrimination tainted this process.

To carry out its plans, Ameritech (working with a consultant) designed a comprehensive workforce reduction program. Ameritech eventually selected a benefit enhancement package that calculated a manager's service pension eligibility as if she had remained employed until December 31, 1994. The package offered 2% of current pay for each year of service, with a minimum of 8% and a maximum of 50%. Under that package, about 80% of the managers over the age of 50 were eligible for a service pension, and managers began to become eligible between the ages of 45 and 49.

After they received the consultant's report, Indiana Bell and ASI each designed its own resizing program (though the two programs were, for all intents and purposes relevant here, identical). ASI hoped to reduce its management workforce of 6,695 people by 15%; Indiana Bell set a target for reducing its management force of approximately 1,250 by 10%, or between 125 and 150 people. The companies planned on achieving these goals with a combination of voluntary and involuntary components. Ameritech as a whole hoped that its management workforce of just over 21,000 would be reduced by 2,500 as of March 31, 1993, and by as much as 6,000 over the following two to three years.

When the "resizing" was announced, employees were told that if they voluntarily retired during a six-month window (between August or September of 1992 and March 1993), and if they signed a waiver and release, they would receive the full complement of benefit enhancements. If they did not sign the waiver, they would receive a reduced package of enhancements. In addition, at Indiana Bell, employees in SG3 and lower grades were told they could apply for available nonmanagement (or craft) positions. Five of the Allard plaintiffs chose that option.

The companies dubbed the involuntary termination program the Company Resizing Program, or CRESP for short. Gary Morris, an employee of ASI, designed CRESP, which ASI then used. Indiana Bell used a variant of CRESP that was developed by its director for human resources, Robert T. McFeely. The program took a three-stage approach.

Stage 1 was the "mechanical" phase in which employees were placed in groups of similar skill and salary level. At ASI, employees were placed in 102 groups, with the basic salary groupings being SG5 and below, SG6 to SG9, and SG 10 and above. At Indiana Bell, employees in SG3 and below were ranked together as were those in SG4 and SG5. (The workforce reduction of the few Indiana Bell employees in SG6 and above was not handled through this process.) Employees in the groups were ranked and the lowest 30 to 35% of each group--the "at risk" employees--was then identified. (Volunteers were ranked on the "at risk" list along with everyone else, because of the chance that they might revoke their voluntary retirement election.) In implementing the mechanical rankings, ASI used the managers' 1990 and 1991 merit pay data. (For ASI employees in SG6 and above, an additional factor based on leadership development and potential accounted for 20% of the score.) At Indiana Bell the Stage 1 rankings were compiled using the most recent evaluations of employees' managerial skills and "promotion potential," as well as their performance over the past two years as indicated by performance ratings, lump sum merit awards, and salary target rates.1

After the initial Stage 1 lists were made, higher level managers refined them, moving people in and out of consideration for Stage 2; managers were instructed to consider factors such as an employee's recent promotion or upgrade or irreplaceable skill in making these determinations.

Once the Stage 1 "at risk" lists were finalized, the employees on those lists were again evaluated. In Stage 2, management teams met, discussed, and re-ranked the employees on the "at risk" lists. For this purpose, ranking groups were subdivided. At ASI, some of the larger of the 102 CRESP groups were divided along functional and/or geographic lines into smaller groups of 30 to 35 employees. This resulted in approximately 135 ranking groups during Stage 2 at ASI. ASI managers ranked employees in those groups according to job skills, experience, and knowledge; job performance, both in the immediate and distant past; leadership criteria; and "growth potential." At Indiana Bell, employees were divided into eight subgroups--one for each of SG1, SG2, SG4, and SG5, one for administrative employees, and three for SG3. Indiana Bell ranking teams were instructed to consider 1990-91 performance, pay, promotability and skills, performance to date for 1992, and "Ameritech Leadership Characteristics" (also known as "breakthrough leadership characteristics").

Morris added "growth potential" to the Stage 2 evaluation factors in August 1992. CRESP training materials defined this term to include both "advancement potential" and "potential to grow and change with the business." Morris explained that he assumed that older people would score lower on that point. As he put it, older people tended to be ranked lower on that factor "due to the fact that there was a relationship between age, not in merit, but age and the potential appraisal piece." He further indicated that older people "have less potential to move upward. They are at the position they basically should be." Other evidence indicated that a variation of the "growth potential" criterion was initially developed as part of AT&T's managerial appraisal system in order to predict how successfully an employee could operate at a higher responsibility level.

Rankers discussed their personal experiences with the at-risk managers, and looked at the managers' current performance. At Indiana Bell, for example, ranking committees divided each ranking group into three subgroups, discussing those employees in the bottom third in depth. They then gave numerical rankings to the bottom subgroup. At both companies, raters took notes about the ranked employees, but most of those notes were destroyed. Once the committee ranking process was complete, ASI and Indiana Bell company officers moved on to Stage 3, in which they decided how many of the ranked managers in each salary grade and department would be terminated involuntarily. They did so by drawing lines on the lists as they stood after the re-ranking process.

In the end, ASI let go 1,320 managers (19.72%). Of those, 591 volunteered, with 200 of the "volunteers" deciding to do so only after they found out they had been selected for termination. Of the 1,320, 894 (67.73%) were 40 or older. Looking for the moment at its managerial ranks as a whole (though we discuss below the question whether this is a proper lens through which to view this process), ASI selected for termination 12.63% of those aged 40-44, 16.71% of those aged 45-49, 24.58% of those aged 50-54, and 29.19% of those aged 55 and older.

Indiana Bell's total workforce of 1,261 management employees was reduced by 223 through the resizing program. Of the 1,192 employees in the lower salary grades (SG5 and below), 152 took voluntary retirement, 34 transferred to non-management positions, and 48 were voluntarily terminated. Of the 223 total managers who left the company, 203 were over 40, and of the 59 total employees involuntarily terminated, 56 were over 40.

Both Indiana Bell and ASI provided pension benefits for their management employees through the Ameritech Management Pension Plan. Benefits under the Plan were available depending on both the chronological age an employee had reached and years in service. A service pension was thus available to an employee who was either 65 years old with 10 years of service, 55 years old with 20 years of service, 50 years old with 25 years of service, or any age with 30 years of service. For employees who did not qualify for a service pension, there was a deferred vested pension right. The requirement for that was five years of service, but not enough to qualify for the service pension. No pension of any kind was available for an employee with less than five years of service.

Participants in the Plan received their benefits in one of three ways: a single life annuity, a joint and survivor annuity, or a lump sum payment. To receive the last of these, the employee had to make an election within a specified time. In order to determine the proper amount of the lump sum, the company applied the Pension Benefit Guaranty Corporation (PBGC) annuity discount rate in effect on January 1 of the year in which the employee left the company. Alternatively, an employee eligible for a service pension could defer payment of a lump sum until January 1 of the year following retirement. In that event, the company would apply the PBGC rate in effect on January 1 of that year rather than the year of retirement. The lower the discount rate (i.e., the more optimistic the PBGC's assumptions about inflation were), the higher the lump sum payment.

The PBGC rate on January 1, 1992, was 6.5%; the rate in effect on January 1, 1993, was 5.75%. At Indiana Bell's termination meetings, the company gave employees information about how those eligible for service pensions could elect a lump sum up to 30 days following termination. The company also provided information about the deferral process and explained that deferral had to be requested before leaving the payroll.

The Plan spells out a specific written administrative review and appeal process. According to the Summary Plan Description, exhaustion of remedies is required: "[T]he plan provisions require that you pursue all your claim and appeal rights described above before you seek any other legal recourse regarding claims for benefits or net credited service or vesting service."

The principal theories on which both sets of plaintiffs rely relate to the way in which the lay-offs were orchestrated and the interrelation between those selected for termination and their age and pension status (which the plaintiffs assert created a strong financial incentive to terminate people below the chronological age thresholds set by the Plan). In addition, the plaintiffs point to other direct or circumstantial evidence that they believe reveals that age bias lay behind the two companies' decisions. A sample of these statements includes the following:

* Ameritech's president, William Weiss, called for broad changes to "reinvigorate" the company, and another member of senior management commented that the company needed to be "ventilated."

* Thomas Reiman, Indiana Bell's president, stated during his first videotaped conference as president: "The business will absolutely dry up and die if we don't have a continued influx of new young crazy people in our business. And we're going to find a way to do that."

* Robert Knowling, upon becoming general manager of one of Indiana Bell's business customer support groups at age 36, commented that he was "aghast when he came into the department and found 12 white middle-aged managers working for him."

* Ranking documents that were not destroyed described one 59-year-old employee as "retired in place" and a 61-year-old employee as "old school."

* Jim Goetz (age 35), who served as a coordinator during the 1992 selection process for ASI, discussed the reorganization in a company video made on September 10, 1993. Stressing the fact that the company would be hiring significant numbers of people from the outside despite the recent terminations, Goetz said "some of this [new hiring] you'll see are just analysts, entry-level analyst jobs where we want to get back and start bringing in some folks that are under 45 years old." Later, apparently having been prodded by a colleague, Goetz attempted to clarify that remark by saying, "My comment earlier regarding looking for people under 45, all I meant is that we want to be open to hiring people from colleges again, that's all I mean."

Initially, this was one lawsuit, brought on March 30, 1993, by former employees of both companies. On October 4, 1993, a second amended complaint was filed in Adams and a new complaint, brought by 34 former Indiana Bell employees, was filed in Allard. (Several of the Allard plaintiffs had been parties to the Adams litigation; upon the filing of the Allard complaint, they withdrew.)

The district court ultimately granted summary judgment in favor of the defendants on almost all the claims the plaintiffs were presenting. Specifically, it found in defendants' favor on (1) the plaintiffs' claim that both companies had engaged in a pattern or practice of age discrimination, (2) each individual plaintiff's separate ADEA claim for disparate treatment, (3) the constructive demotion claims (Allard only), (4) the ERISA claims under sec.sec. 502, 503, 510, and (5) the state law wage and trade defamation claims. Along the way, however, the court had ruled that the waivers of ADEA claims were invalid under the Older Workers Benefit Protection Act of 1990 (OWBPA), Pub. L. No. 101-433, 104 Stat. 978, codified at 29 U.S.C. sec.sec. 621, 623, 626, for employees who signed after being terminated. It also dismissed Count II of Indiana Bell's counterclaim in Allard, which had sought damages resulting from the plaintiffs' breach of their waiver agreements with respect to the non-ADEA claims, and denied ASI leave to add a similar count to its counterclaim in Adams.

This left for resolution only the claims in Count I of the Adams defendants' amended counter-claim (which seeks restitution from those plaintiffs who received compensation in exchange for their release of their potential ADEA claims), two of the individual Adams plaintiffs' state law wage claims for pre-termination vacation pay, and in Allard Indiana Bell's claim for restitution remains in the district court. The district court certified the remaining decisions for immediate appeal under Rule 54(b).

The Supreme Court's most recent pronouncement on the legal standards a plaintiff must meet in order to prove a case of age discrimination appears in Reeves v. Sanderson Plumbing Products, Inc., 120 S. Ct. 2097 (2000). In Reeves, the Court began by reiterating that in such cases, "liability depends on whether the protected trait (under the ADEA, age) actually motivated the employer's decision." Id. at 2105, quoting Hazen Paper Co. v. Biggins, 507 U.S. 604, 610 (1993). That is, it continued, "the plaintiff's age must have actually played a role in [the employer's decisionmaking] process and had a determinative influence on the outcome." 120 S. Ct. at 2105, quoting Hazen Paper at 610 (bracketed text in Reeves). Like our case, Reeves dealt with a claim of disparate treatment on account of age. In addition to those claims, our case also involves claims of an unlawful pattern or practice of discriminating on the basis of age, which, as the Supreme Court observed in International Brotherhood of Teamsters v. United States, 431 U.S. 324 (1977), is another theory of intentional discrimination, under which the plaintiff bears the burden of showing "by a preponderance of the evidence that [age] discrimination was the company's standard operating procedure--the regular rather than the unusual practice." Id. at 336.

Courts have sometimes struggled with the proper way to apply these different approaches to reduction-in-force, or RIF, cases. They note the obvious point that adverse actions, in the form of job loss, are inevitable for some people in RIF situations, and that the corporation is taking the action for general efficiency reasons. We have explained on several occasions, however, that the fundamental analysis of RIF cases is no different from the analysis appropriate to other forms of discrimination. See, e.g., Thorn v. Sundstrand Aerospace Corp., 207 F.3d 383, 386 (7th Cir. 2000); Bellaver v. Quanex Corp., 200 F.3d 485, 493-94 (7th Cir. 2000). In Bellaver, we acknowledged that it makes no sense in a RIF case to require the plaintiff to show that the job remained open. Instead, the plaintiff must show that "the employer carried out the RIF in a discriminatory way." 200 F.3d at 494. Conceptually, one can think of a RIF as a situation (not unlike "zero-based budgeting") in which the employer decides whom from a defined group it will "re-hire" or retain, considering all existing employees as roughly like applicants for retention. See Matthews v. Commonwealth Edison Co., 128 F.3d 1194, 1195 (7th Cir. 1997). Discriminatory hiring practices have been analyzed for years under the various frameworks we have described, and RIFs do not require anything significantly different.

In a discrimination case proceeding under a so-called disparate impact theory, statistical evidence is obviously of central importance, because in such a case the plaintiff is trying to show that a particular job requirement has a differential impact on a defined protected group. But, as the district court correctly held, disparate impact is not a theory available to age discrimination plaintiffs in this circuit. See, e.g., Maier v. Lucent Technologies, Inc., 120 F.3d 730, 735 & n.4 (7th Cir. 1997); EEOC v. Francis W. Parker School, 41 F.3d 1073, 1077-78 (7th Cir. 1994). See also Blackwell v. Cole Taylor Bank, 152 F.3d 666, 672 (7th Cir. 1998) (citing cases on both sides of issue from various circuits). We must therefore consider the ways in which statistical evidence can be brought to bear in a disparate treatment or "pattern or practice" case.

Our general framework in this connection is set forth in the Supreme Court's decisions about the admissibility of expert testimony, like the statistical evidence proffered here. The leading case, mysteriously not cited in the plaintiffs' briefs, is Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). Daubert made clear that expert testimony should not be considered in a case unless the expert has genuine expertise (there, scientific knowledge) and that expertise will assist the trier of fact to understand or determine a fact in issue. Id. at 592. The Court elaborated on Daubert's framework in Kumho Tire Co. v. Carmichael, 526 U.S. 136 (1999). The twin requirements for expert testimony, as it explained in Kumho Tire, are relevance and reliability. The expert must "employ[ ] in the courtroom the same level of intellectual rigor that characterizes the practice of an expert in the relevant field." Id. at 152. The trial court, exercising its gatekeeping function, must examine (among other things) the expert's qualifications, the methodologies she used, and the relevance of the final results to the questions before the jury. We review the district court's decisions deferentially, see General Electric Corp. v. Joiner, 522 U.S. 136 (1997).

At the threshold, the defendants have questioned whether statistical evidence as a whole can ever be useful in a case alleging disparate treatment or a discriminatory pattern or practice, as opposed to a disparate impact case. The short answer is yes: statistical evidence can be very useful to prove discrimination in either or both of those two kinds of cases, but it will likely not be sufficient in itself. See generally David C. Baldus and James W.L. Cole, Statistical Proof of Discrimination, sec.sec. 9.02, 9.42 (1980). The Supreme Court discussed the use of statistical evidence in a pattern or practice case in Hazelwood School District v. United States, 433 U.S. 299 (1977), in which the Court approved of the use of statistics to help show that the school district was discriminating on the basis of race in its faculty hiring decisions. See id. at 307. Hazelwood also underscored the importance of looking to the proper "community" or group when making statistical comparisons. The relevant labor market was a question of fact for the trial court to consider, although it was probably something like the market for school teachers in the area. Id. at 310-12. Further emphasizing the factual nature of this inquiry, the Court also held that the employer school district was entitled to an opportunity to rebut any inference of discrimination raised by the plaintiffs' statistical showing. Id. at 309-10.

The Court returned to the topic of statistics in Bazemore v. Friday, 478 U.S. 385 (1986). Bazemore involved both a pattern or practice claim brought by the United States and individual plaintiffs' discrimination claims, all arising out of alleged racial discrimination in employment and the provision of services by the North Carolina Agricultural Extension Service. The court of appeals had rejected the plaintiffs' statistical evidence because it had not taken into account every factor that might have affected the employees' salary levels. The Supreme Court disagreed, holding both that the statistical study should have been admitted and that "[n]ormally, failure to include variables will affect the analysis' probativeness, not its admissibility." Id. at 400. The Court also noted that the statistical evidence had to be evaluated in the light of the remaining evidence in the record.

This court, too, has had a number of opportunities to discuss the use of statistical evidence. In Radue v. Kimberly-Clark Corp., 219 F.3d 612 (7th Cir. 2000), we upheld summary judgment for an employer in an ADEA case where the employee's prima facie case was primarily composed of statistics that showed that older employees were treated less favorably than younger employees in various RIFs the employer had carried out. But those statistics were flawed in a number of ways, and the plaintiff had little else with which to support his case. For one thing, the statistics looked at a completely different part of the company from the one in which the plaintiff worked and they involved an earlier RIF, not the one in which the plaintiff lost his job. Id. at 616. For another, the statistics failed to address the essence of the plaintiff's claim: he did not allege that the RIF was age-based; he claimed instead that the transfers awarded in the wake of the RIFs were given preferentially to younger employees. Id. Finally, the court noted that the statistics standing alone could not prove causation; they could only show a relation between the employer's decisions and the affected employees' traits. Id. See also Tagatz v. Marquette Univ., 861 F.2d 1040, 1044 (7th Cir. 1988) ("Correlation is not causation.").

There is no presumption that statistical evidence has no useful role to play in disparate treatment employment discrimination cases--indeed, we are hard pressed to see how anyone could take such a position consistently with the Supreme Court's guidance on the matter and this court has not done so.

Thus, for example, this court in Mister v. Illinois Central Gulf Railroad Co., 832 F.2d 1427 (7th Cir. 1987), the plaintiffs successfully relied on a statistical showing that the railroad hired a much larger proportion of white applicants than African-American applicants to prove disparate treatment. The employer, ICG, had complained that the data was based on inaccurate assumptions about the labor market, but we rejected that criticism because the expert had used the best data available, which had come from ICG itself. Id. at 1430. The opinion then discussed at some length what one can learn from statistical evidence and what its limitations are. Id. at 1430-31.

Studies in employment discrimination cases typically begin by defining the relevant labor market, and then ask what the results would be for the salient variable (race in Mister, age in our case) if there were no discrimination. That is called the "null hypothesis." If the relevant market is 40% African-American, for instance, one would expect 40% of hires to be African-American under the null hypothesis. If the observed percentage of African-American hires is only 20%, then the statistician will compute the "standard deviation" from the expected norm and indicate how likely it is that race played no part in the decisionmaking. Two standard deviations is normally enough to show that it is extremely unlikely (that is, there is less than a 5% probability) that the disparity is due to chance, giving rise to a reasonable inference that the hiring was not race-neutral; the more standard deviations away, the less likely the factor in question played no role in the decisionmaking process. See Hazelwood, 433 U.S. at 308 n.14; Castaneda v. Partida, 430 U.S. 482, 496 n.17 (1977); see also Tagatz, 861 F.2d at 1044 (discussing the "significance of statistical significance"); Palmer v. Schultz, 815 F.2d 84, 90-97 (D.C. Cir. 1987) (discussing cases and forms of statistical proof). See generally Baldus and Cole, supra, sec. 9.03 (1980 & Supp. 1987).

Mister also noted the importance of making sure that any testing adequately accounts for the real variables that the employer took into account. The plaintiffs' evidence there was wanting because it assumed that the job applicants were identical in every respect except for race, even though other factors like physical condition, employment history, or other non-racial variables might have entered into the employer's calculus. 832 F.2d at 1431; see also Tagatz, 861 F.2d at 1044. But, importantly, it was the ICG that had the responsibility of offering alternative explanations. The only one it actually advanced was that it was concerned about the distance its employees had to travel to get to the job site; for reasons pointed out in the opinion, this was singularly unpersuasive. (It is also important to note that Mister reached this court after a full trial on the merits; nothing in the opinion suggests that the plaintiffs' statistical showing was so weak that summary judgment for the defendant would have been appropriate.)

We are now in a position to consider whether, under the Daubert standards, the statistical evidence plaintiffs have offered in the two cases before us (coupled with the other evidence they presented) was sound enough methodologically (i.e., reliable enough) and relevant, such that the district court should have taken it into account in evaluating their claim. Several points are important to bear in mind. First, the question before us is not whether the reports proffered by the plaintiffs prove the entire case; it is whether they were prepared in a reliable and statistically sound way, such that they contained relevant evidence that a trier of fact would have been entitled to consider. No one piece of evidence has to prove every element of the plaintiffs' case; it need only make the existence of "any fact that is of consequence" more or less probable. See Fed. R. Evid. 401 (emphasis added). See also, e.g., United States v. Porter, 881 F.2d 878, 887 (10th Cir. 1989) ("'An item of evidence, being but a single link in the chain of proof, need not prove conclusively the proposition for which it is offered.... It is enough if the item could reasonably show that a fact is slightly more probable than it would appear without that evidence.... A brick is not a wall.'"), quoting McCormick on Evidence, sec. 185 at 542-43 (E. Cleary 3d ed. 1984) (footnotes omitted). Put a little differently, the issue is whether the criticisms of the Wertheimer reports and the plaintiffs' other statistical evidence affected the admissibility of those materials, or only, as the Supreme Court put it in Bazemore, their "probativeness" or weight.

Second, the question before the district court, and before us on de novo review from the summary judgments, is not whether we find one set of expert reports more persuasive than another. See Casey v. Uddeholm Corp., 32 F.3d 1094, 1099 (7th Cir. 1994); Roberts v. Sears, Roebuck & Co., 723 F.2d 1324, 1338 (7th Cir. 1983). It is whether, taking the facts in the light most favorable to the plaintiffs, a trier of fact should be permitted to make that choice.

Third, embedded within the debate about the relative merit of the plaintiffs' versus the defendants' experts are a number of factual issues. One is the same kind of question that concerned the Supreme Court in Hazelwood: what is the proper level of aggregation or disaggregation at which ASI's and Indiana Bell's actions should be assessed? At one extreme, one could perhaps look at the two companies' entire workforces, management and non-management alike; at the other extreme, one could take a highly individualistic view of humanity and conclude that no two people are exactly alike and statistics are therefore worthless. Neither approach has much to recommend it, of course, but the thought experiment suggests the outer possibilities.

What Wertheimer did, briefly put, was to examine the correlations that existed between the ages of employees and the companies' decisions to terminate. What he did not do (and, as far as we can tell, what the defendants' experts did not do either) was to run a multiple-regression analysis that would have isolated the relevance of age as a factor in the companies' decisions. While this omission strikes us as odd, we are not prepared to hold as a matter of law that nothing but regression analyses can produce evidence that passes the Daubert and Kumho Tire thresholds. Statisticians might have good reasons to look at data in different ways. (For example, as additional variables are introduced into a regression, the less likely it is that any of them will be statistically significant, a fact that causes its own problems.) We thus evaluate here what Wertheimer did, rather than hypothetical tests that he or another expert might have done.

In the ASI reports, Wertheimer initially grouped all management employees together and did an aggregated study, while the defendants' expert focused on each particular CRESP group, even though some of those groups were as small as 30 people. Next, in his "rebuttal" report, Wertheimer looked at groups across the 13 vice-presidential groupings, which (according to defense expert McCabe) "reflect[ed] the different activities that are the responsibilities of these vice presidents." Also in the rebuttal report, Wertheimer both criticized the CRESP groups as being too small to serve as meaningful sets for analysis, and he pointed out that even a CRESP-oriented analysis showed significantly higher selection rates for terminations of older workers. Finally, in a "declaration" that also addressed McCabe's conclusions, Wertheimer analyzed the data by salary group, looking particularly at Stage 1 of the process.

We summarize here Wertheimer's findings at each of those levels. After making some comparisons between the age profile of ASI's workforce as a whole and that of the entire country (which we do not find particularly useful and thus disregard), he calculated the termination rates of employees by age. He did this by dividing the number of terminated workers in each age category by the number of workers in that category before the terminations. The results for the November 1992 terminations were as follows:

Age < 30: 6.2%            Age 45-49: 12.6%
Age 30-34: 9.9%           Age 50-54: 12.5%
Age 35-39: 9.9%            Age 55-59: 9.5%
Age 40-44: 10.5%          Age 60-64: 26.3%                 
Table D-2, Wertheimer Report.

Another way he reviewed the same data was to compare the share of terminations accounted for by employees at least age 40 with the share of the under-40 group: the former accounted for 62.6% of the 1992 terminations, even though they were only 58.1% of the workforce. For the 1993 terminations, the differences were greater: employees at least 40 accounted for 79.3% of the terminations, but only 61.3% of the workforce. Last, in this report he considered the way that the selection process worked to see how the designation of the at risk employees in Stage 1 and the termination candidates at Stage 2 correlated to age. At Stage 1, the selection rate for the under-40 employees was 28.6%, and for the older employees it was 35.3%; at Stage 2 the selection rate for the younger group was 38.1%, compared with 46.7% for the older employees. These differences exceeded two standard deviations and, Wertheimer concluded, were thus statistically significant. (By that, he meant that the probability that the difference would have been observed even if the hypothesis being tested were false is less than 5%. Most likely, some other factor--perhaps age, perhaps something else correlated to age--explains at least some of the difference.)

Perhaps, however, there were non-age-related differences among the management employees as a whole that should have been taken into account, as McCabe argued. Recall that ASI and Indiana Bell defined management very broadly, to include practically all white-collar workers. Wertheimer's rebuttal report was also before the district court. In it, he criticizes McCabe's use of CRESP groups on the ground that at least many of them were too small to yield useful information. He pointed out that 69.6% of the CRESP groups had fewer than 50 employees. When samples are that small, he argued, it is hard to find statistically significant differences between two selection rates. (This statement necessarily assumes that the smaller groups are not the appropriate reference groups; a large group cluttered with too many unlike individuals is not preferable to a smaller group of similarly situated people.) Wertheimer opined that one would need at least 300 employees in a CRESP group to be confident that useful information about the Stage 1 selection rate for older employees could be detected. (Aside from reflecting the truism that it is always better to have more data than less, this comment does not tell the full story; there are statistical techniques that can be used for small samples, although neither Wertheimer nor the defendants' experts appear to have used them.) Only four CRESP groups met Wertheimer's preferred size criterion.

Wertheimer's rebuttal also analyzed the selection rates for employees by vice-presidential group and by age within each group. That exercise showed that selection rates were higher for managerial employees at least age 40 in 12 out of the 13 groups. The difference was statistically significant in six of those 12 categories, in the sense that there was a small (less than 5%) probability that the difference was due to chance. His later declaration did much the same thing, using salary groups instead of vice presidential groups.

Wertheimer's work on the Indiana Bell reports was similar. Initially, he analyzed the managerial workforce as a whole and found that a disproportionate number of older managers were selected for termination. Once again, in a rebuttal report he criticized the fragmentation he found in the McCabe study, which for Indiana Bell subdivided the employees in question into salary grades and some sub-grades, and then made seven comparisons for each group. Again, this led to very small groups and subgroups, with 28% containing fewer than 10 employees. In Wertheimer's view, the final result of the Indiana Bell process was the proper focus. He also analyzed the selection rates of the Indiana Bell managers by overall salary grade level, however, looking at the end result for each salary grade rather than after each of the seven steps along the way to determining who would be let go. This process showed that age correlated with the results. In the SG1 group, no employees below age 40 were selected for termination, while 25% of those 40 and above were selected. This difference was also statistically significant, in the sense that it meant that there was less than a 5% chance that it was explainable by chance.

There is far more detail in the record, but this description is enough to give an idea of what Wertheimer was doing. The district court found his reports inadmissible for several reasons: (1) the underlying information about the RIF programs was not reliable; (2) the reports only showed that the difference in treatment between the over and under 40 aged individuals was not due to chance, but they did not affirmatively indicate what caused that difference; (3) the analysis did not take into account or control for other non-age related variables; (4) Wertheimer relied on the plaintiffs' description of the RIF and did not himself become familiar with the procedures used; and (5) the jury would find the reports so confusing that they should be excluded under Rule 403.

The underlying information about the RIF came from the defendants ultimately, and as such we see no problem in Wertheimer's decision to rely on it. Nor do we see anything wrong in the way Wertheimer familiarized himself with the process. The more serious objections the court had were its second, third, and fifth. The theme of numbers two and three was that Wertheimer's analysis, standing alone, was not enough to show that age was the reason why ASI and Indiana Bell took the actions that they did. That much is true: the statistical analyses were enough to rule out chance, but the real reason for the decisions may have been age or it may have been some other factor or factors positively correlated with both advancing age and the likelihood of termination. But ruling out chance was an important step in the plaintiffs' proof, even if it was not a single leap from the starting line to the finish line. If this is all the plaintiffs had introduced, we would agree with the district court that the record would have supported summary judgment against them. It was not, however, and in our view the other items of evidence, if believed by a jury, could have done the rest of the job: that is, it could have ruled out factors other than age.

Importantly, however, the "other factor" cannot itself be something that is a euphemism or proxy for age discrimination. Morris's assumption that people over 40 are already working in the highest job they will ever reach could be viewed by a trier of fact as just such a stereotype. A trier of fact might conclude that it and its analogues--"advancement potential" and "growth potential"--infected the process of determining who would be terminated in the downsizing, meaning that covert age discrimination was at play. (Taking the inferences from this statement in the light most favorable to the plaintiffs, it reflected Morris's ex ante assumption that older people have no growth potential; we realize that a trier of fact might also construe it as an innocuous ex post observation that there is a correlation between people whose growth potential has been exhausted and older workers, but at this stage we cannot draw inferences in favor of the party moving for summary judgment.)

The disputes that the parties have highlighted between Wertheimer and McCabe (and among the other experts on both sides) went to the weight of the evidence each was presenting, not to its admissibility. Once Wertheimer was analyzing vice presidential groups or salary groups, he was talking about people sufficiently similarly situated that general conclusions could be drawn. Indeed, this is exactly what ASI and Indiana Bell argue when they challenge the district court's decision that the waivers certain employees signed did not comply with the OWBPA. The district court found the waivers invalid in part because the waivers referred only to the salary grades of individuals eligible for or selected for the termination program, not their precise job titles. The specificity of job titles was necessary given the purpose of that statute, according to the district court. ASI and Indiana Bell objected, defending the utility of the salary grade information. So, at least in some contexts, even the companies believed that salary grades were the appropriate groupings to use.

Last, we address the district court's alternative ruling that the statistical evidence should be excluded under Fed. R. Evid. 403. We would be more receptive to this possibility if it were not for two problems. First, while it is not unheard of to exclude evidence under Rule 403 at the summary judgment stage, see, e.g., Weit v. Continental Illinois Bank & Trust Co., 641 F.2d 457, 467 (7th Cir. 1981), normally the balancing process contemplated by that rule is best undertaken at the trial itself. Second, we simply cannot tell from this distance whether the court's Rule 403 decision would have been the same or not, if it had recognized the limited but important role the plaintiffs' statistics could play. As the court reconsiders the case on remand in accordance with this opinion, and particularly if the case goes forward to a trial, it will be free to re-weigh the admissibility question under Rule 403 and come to a fresh conclusion. In the short term, however, it must also bear in mind that the Supreme Court's recent decision in Reeves was a cautionary note not to grant summary judgment too readily when facts are susceptible to two interpretations, and not to dismiss as irrelevant damaging remarks like the one there (plaintiff so old he "must have come over on the Mayflower," 120 S. Ct. at 2110), or, in our case, Goetz's proclaimed desire to hire people under the age of 45.

The appeal we have before us, we reiterate, came from a decision to grant summary judgment. We do not rule out the possibility that either or both companies will be able to prevail before a trier of fact. We hold only that this evidence met the standards of admissibility set by the Federal Rules of Evidence and thus should have been counted on plaintiffs' side for summary judgment purposes.

It is critically important that the plaintiffs had other evidence to satisfy other parts of their prima facie case. Wertheimer also studied the question why ASI and Indiana Bell would have had a motive to engage in intentional discrimination against older workers. The answer, briefly, lay in the structure of the Ameritech pension plan. Unlike the plan in Hazen, which was phrased only in terms of years of service, the Ameritech Plan used chronological age as an important variable in determining pension eligibility. By terminating employees who had not yet reached age 55, for instance, Ameritech was able to cut off the possibility of their moving from the "deferred vested pension" category to the service pension category. For ASI, Wertheimer calculated that the pension system gained $32,686,607 by terminating deferred vested pension employees, while it lost $4,896,505 by terminating "extra 2 service pension employees" and $8,518,087 by terminating regular service pension employees, for a net gain of $19,272,016. The net gain to the pension system from Indiana Bell's terminations was $1,202,689, according to Wertheimer.

The plaintiffs also had evidence showing that the CRESP program itself may not have been age neutral in design or implementation. This ranged from the explicit comments about the desire to change the demographic make-up of Ameritech's workforce, to possibly tainted criteria (i.e., the "potential" factors), to (suspicious) destruction of evidence. We are not concerned about remarks to the effect that the company needed to be "reinvigorated" or "ventilated." As we have commented in other cases, there is no reason why those goals cannot be accomplished with talented older workers or with talented younger workers. Reiman's comment about wanting "new young crazy people" is harder to sanitize, especially because as president it would be hard to say he was not in the direct decisionmaking line for the structuring of a company-wide restructuring program. Worst of all was Goetz's explicit statement that ASI wanted to go out and hire people under the age of 45 again. True, Goetz tried to limit the damage by spinning the remark into one that expressed a desire to hire from colleges again, but a trier of fact could find that this recharacterization did little to change the meaning. We note that statistics published by the National Center for Education Statistics at the U.S. Department of Education indicate that only about 8% of the students enrolled full-time in institutions of higher education in 1993 were over 35 years of age (643,000 out of a total of 8,128,000). Digest of Educational Statistics, 1999, 204, Table 177. As of 1997, looking at undergraduate students 40 and over in two- and four-year degree granting institutions, a mere 3.4% were 40 and over. Id. at 205, Table 178. These numbers and a little common sense suggest that a trier of fact could see a desire to hire "from colleges" and a desire to hire "under 45-year-olds" as equivalents.

The district court analyzed the cases as McDonnell Douglas indirect proof cases and pattern or practice cases. The record was full of objective evidence--far more than the plaintiffs' own evaluations of their performances--suggesting that not only were they doing their jobs well, but (more importantly) that in the comparative ratings that are required in a RIF, they could have survived had the criteria been age-neutral. The size of the merit bonuses and raises the companies paid them is one example of such evidence, as is documentation of employees' job skills, experience and past job performance.

Before turning to the ERISA claims, we add a final word about the way this case has been handled thus far. We have the greatest respect for the conscientious way in which the district court worked its way through the evidence, and we have every confidence that it will continue to manage this difficult case well on remand.

We can dispose of the ERISA claims in both Adams and Allard with very little discussion, because very little discussion is all they received in the briefs on appeal. Indeed, the briefs pay so little attention to this theory of the case that we could rest our decision on the ground that the plaintiffs have not preserved these theories as a separate ground for relief (as opposed to evidentiary support for their ADEA claims). Even if we reach the merits, we agree that summary judgment in the defendants' favor was proper on these claims. Briefly, both the Adams and Allard plaintiffs had argued that they were entitled to benefits under ERISA sec.sec. 502 and 503. 29 U.S.C. sec.sec. 1132, 1133. The district court rejected these claims on the ground that the plaintiffs who had originally raised them failed to exhaust their remedies under the Plan, and the one Allard plaintiff (Meyers) who did exhaust failed to include this as a ground of relief until it was too late. The standard of review for a dismissal for failure to exhaust Plan remedies is abuse of discretion. Gallegos v. Mount Sinai Med. Ctr., 210 F.3d 803, 808 (7th Cir. 2000); Janowski v. International Bhd. of Teamsters Local 710 Pension Fund, 673 F.2d 931, 935 (7th Cir. 1982). We see none here, either in the decision concerning exhaustion or in the decision to deny permission to Meyers to amend the pretrial contentions to add this point long after the time for putting claims on the table had expired.

Both sets of plaintiffs also presented claims under ERISA sec. 510, which prohibits an employer from discriminating against an employee "for the purpose of interfering with the attainment of any right to which such participant might be entitled under the plan." (Only the Adams plaintiffs raise this issue on appeal, however.) One might think from the earlier discussion that they were pursuing a broad version of this theory, but that is not the case. Instead, the Adams plaintiffs sec. 510 claim alleges that the defendants terminated them in November 1992 in order to prevent them from learning what the January 1993 PBGC rate would be in time to make an election. The district court initially denied the defendants' motion for summary judgment on these claims, finding that the plaintiffs had presented enough evidence to show that the CRESP evaluation process may have been tailored to disfavor older workers. The court later reversed itself after it reconsidered ASI's argument that one should not equate pension eligibility and age and that knowledge of a potentially adverse effect was not the same thing as intent. Both those propositions are true, but we are not certain either one has great applicability here. On the other hand, plaintiffs have pointed to no evidence in this record that would support a finding that the timing of the terminations was designed to manipulate the discount rate. For that reason, we also affirm summary judgment for ASI on this theory.

ASI and Indiana Bell have attacked the district court's ruling in the plaintiffs' favor that the waivers many of them signed were invalid under the OWBPA. This point is relevant to the ADEA claims (because if the waivers were valid, the ADEA claims are barred), and so we must address it. The Act states that "an individual may not waive any right or claim... unless the waiver is knowing and voluntary." 29 U.S.C. sec. 626(f). In addition, if the waiver is sought in connection with an early retirement program, it must meet the following additional requirements:

(H) if a waiver is requested in connection with an exit incentive or other employment termination program offered to a group or class of employees, the employer... informs the individual in writing in a manner calculated to be understood by the average individual eligible to participate, as to(i) any class, unit, or group of individuals covered by such program, any eligibility factors for such program, and any time limits applicable to such program; and

(ii) the job titles and ages of all individuals eligible or selected for the program, and the ages of all individuals in the same job classification or organizational unit who are not eligible or selected for the program.

29 U.S.C. sec. 626(f)(1)(H).

The waivers that the plaintiffs signed in this case did not refer to the job titles of those selected for the program; instead, they mentioned only salary grade. As we noted above, in this context the defendants have suddenly decided that salary grade gives plenty of detail. We do not need to decide here whether a literal job title is always required. Indeed, as the Sixth Circuit suggested in Raczak v. Ameritech Corp., 103 F.3d 1257 (6th Cir. 1997), cert. denied, 522 U.S. 1107 (1998), such a literal approach to the statute could lead to hypertechnical requirements that have little to do with the purpose of the law. Id. at 1262-64. Here, however, the court found that salary grade was too general to furnish the kind of information the statute contemplated. As a form of worker protection legislation, the OWBPA demands information that allows people to ascertain whether they are being treated fairly vis-a-vis their peers. Congress's use of the term "job title" indicates that it wanted that information to be quite specific (that is, more literal and particular than the functional approach for which the defendants are arguing, which would have been adequate for a statistician). On this record, we see no error in the district court's decision that the particular waivers these plaintiffs signed failed to comply with the Act.

We also affirm the district court's decision that the additional waivers signed by some individuals who sought to join the Allard litigation were valid. Charles Steinmetz and Raymond Winkler sought permission to opt in as plaintiffs in the Allard case. Magistrate Judge V. Sue Shields denied them permission, and the district court adopted the magistrate judge's ruling.

After Steinmetz, Winkler, and nine other similarly-situated persons were terminated by Indiana Bell, they employed an attorney (Mark Waterfill, one of the attorneys in both the Adams and the Allard litigation) and negotiated a settlement. As part of this settlement, Indiana Bell paid them both the standard enhanced benefits package as well as an additional amount. In exchange, these eleven signed both the standard waiver and an additional waiver, or "Settlement and General Release." Although the district court had ruled that the standard waivers signed by Steinmetz, Winkler, and the other plaintiffs were invalid under the OWBPA, the magistrate held that the additional waivers signed as part of the negotiated settlement were valid, even though the additional waiver, like the standard one, failed to mention the job titles of those who had been terminated. The magistrate judge ruled that the additional waivers were valid despite this deficiency because those who signed the additional waivers were in a substantially different position than the other plaintiffs had been when they signed the standard waiver: they had already been terminated, and they were represented by an attorney, who threatened litigation and negotiated the settlement.

We agree with the magistrate judge (and the district court, which adopted the magistrate's ruling) that the requirements of sec. 626(f)(1)(H) do not apply to the additional waivers. Section 626(f)(1)(H) applies "if a waiver is requested in connection with an exit incentive or other employment termination program...." This language implies that the waivers subject to the requirements of subsection (H) must be requested while the early exit program is being implemented. The additional waivers signed by Steinmetz and Winkler were not signed in connection with Indiana Bell's early retirement incentive program; instead, they were signed in connection with post-termination settlement negotiations. We therefore affirm the decision not to allow Steinmetz and Winkler to opt-in as plaintiffs in the Allard case.

The Adams plaintiffs challenge the district court's decision not to allow them to depose Richard Janney, a former ASI General Counsel and Vice President. Janney authored a letter to the editor published in the October 24, 1995, edition of the Chicago Tribune in which he expressed his opinion that stock options offered as part of executive compensation are in large part responsible for "continuing lay-offs and low pay raises" for workers. Ameritech and its subsidiaries were not mentioned in the letter. Three weeks later, and five months after the close of discovery, the Adams plaintiffs moved to reopen discovery so that they could depose Janney. The district court overruled the motion on January 26, 1996.

Although the district court provided no reasons for its decision, discovery is an area over which the district courts have great authority and discretion and we will not disturb the district court's resolution of a discovery dispute unless we find that the court abused its discretion. Corley v. Rosewood Care Ctr., Inc. of Peoria, 142 F.3d 1041, 1052 (7th Cir. 1998); Gile v. United Airlines, Inc., 95 F.3d 492, 495 (7th Cir. 1996). Given that Janney's name and position were known to the plaintiffs well in advance of his letter and that their request to depose him was well after the close of discovery, we find no such abuse of discretion with respect to Janney.

Five of the Allard plaintiffs filed claims that they had been "constructively demoted." Although this is a strange label for a claim, what these five plaintiffs allege at bottom is that they were coerced into applying for and taking jobs with lower pay, benefits, and status because they were at risk of being terminated during the RIF; and because the RIF violated the ADEA, they allege, their decisions to take a demotion were also infected with discrimination.

The district court granted summary judgment in favor of Indiana Bell on these claims, reasoning first that the plaintiffs had failed to produce sufficient evidence to create a genuine issue of material fact that illegal age discrimination motivated the workforce resizing or any individual termination, and second that the plaintiffs' decisions to apply for and accept lower-paying craft positions were voluntary.

As both sides and the district court agree, the critical precedent on this issue is Henn v. National Geographic Society, 819 F.2d 824 (7th Cir. 1987). In Henn, this court ruled that the offer of an early retirement program could be discriminatory and violate the ADEA if those who took the early retirement offer did not have a true choice in the matter:

[The plaintiffs] could prevail only by showing that [their employer] manipulated the options so that they were driven to early retirement not by its attractions but by the terror of the alternative. If the terms on which they would have remained at [their employer] were themselves violations of the ADEA, then taking the offer of early retirement was making the best of things, a form of minimization of damages. Suppose, for example, that one of the plaintiffs expected to earn $200,000 in his remaining time at [his employer], but [it] (in violation of the ADEA) had just cut his salary because of his age, so that he now could expect no more than $100,000. The employee would snap up an early retirement package worth $110,000 but might reject the same old package had he been allowed to work at his old wage.... The decision to reduce one's injury from the employer's violation of the ADEA would not prevent a suit seeking to recover the remainder of the loss.

Id. at 829. See also Karlen v. City Colleges of Chicago, 837 F.2d 314, 317 (7th Cir. 1988) ("Whether a worker takes early retirement because he fears he will be discriminated against on account of his age if he does not, or refuses to take early retirement and indeed is then discriminated against on account of his age, his rights under the Age Discrimination in Employment Act have been violated.").

The Henn dilemma describes what the constructive demotion plaintiffs faced to a "T." Their decisions to accept demotions to lower paying and lower status jobs were taken in the face of this possibly discriminatory action. These plaintiffs were not faced with a truly voluntary choice: because they were already on the "at risk" lists, they knew that they sat between the Scylla of a better job but with low security, and the Charybdis of a lesser job but with high security. If the plaintiffs succeed in demonstrating that the RIF was infected with and motivated by age discrimination, then they should be allowed to proceed with their claim that their decisions to accept demotions were the result of the same discrimination.

The parties have raised a number of additional points in their briefs. We note here only the fact that our decision with respect to the plaintiffs' right to go forward on their discrimination cases covers both the pattern or practice theory and the individual cases. We Reverse the claims in both Adams and Allard that arise under the ADEA and Remand for further proceedings. We Affirm the district court's dismissal of all claims based on ERISA. Each party shall bear its own costs on appeal.