The district court granted summary judgment for the plaintiff's former employer in this age discrimination in employment suit, and the plaintiff appeals. The plaintiff, age 54 when he was hired in 1997, was an information technology consultant with extensive experience in the health care industry, and he was hired specifically to service a large health care client of the defendant. Just a few weeks after the plaintiff was hired, the defendant lost the client, and several months later it decided to discontinue its health care practice. At about the same time it instituted a RIF (reduction in force) targeted on employees who had little prospect for billable work in the forthcoming months, a category that included the plaintiff. Accordingly he was terminated in January of 1998 (after having been employed by the defendant for only five months) along with two other employees out of a total of 32 in the plaintiff's department, of whom 27, including the three who were riffed, were at least 40, the age at which the protections of the federal age discrimination law kick in.

A supervisory employee who happened to have his own suit for age discrimination and retaliation pending against the defendant testified by deposition that the defendant had an age-discriminatory "culture" and that the plaintiff and other older workers were not given the same opportunities as younger workers for choice assignments. This kind of vague, speculative evidence by an employee with his own axe to grind has too little probative value to make out a prima facie case of discrimination, cf. Hoffman v. MCA, Inc., 144 F.3d 1117, 1122 (7th Cir. 1998); United States v. Hooks, 848 F.2d 785, 797 (7th Cir. 1988); Wexler v. White's Fine Furniture, Inc., 246 F.3d 856, 862 (6th Cir. 2001); Feliciano De La Cruz v. El Conquistador Resort & Country Club, 218 F.3d 1, 5 (1st Cir. 2000), and we must see whether there is anything else. One thing else is that according to this same employee, the head of the plaintiff's department, who picked the employees including the plaintiff in that department to be riffed, had once said to this witness, "Bob, you know how hard it is to sell these guys and they just don't look like they work as hard." The meaning of the statement is obscure; but even if it expresses "ageist" prejudice, it was made at an unspecified date before the RIF, and there is nothing, not even coincidence of time, to tie it to the RIF. Fortier v. Ameritech Mobile Communications, Inc., 161 F.3d 1106, 1112 (7th Cir. 1998); Eiland v. Trinity Hospital, 150 F.3d 747, 751 (7th Cir. 1998); Kennedy v. Schoenberg, Fisher & Newman, Ltd., 140 F.3d 716, 724 (7th Cir. 1998).

All that leaves is statistical evidence, namely that all three of the riffed employees were at least 40. But only five of the 32 employees in the department were under 40, which means that, assuming the choice of employees to riff was random, there was a 59 percent chance that all three riffed employees would be at least 40. See David Freedman, Robert Pisani & Roger Purves, Statistics 228-30 (3d. ed. 1998). The statistical evidence tendered by the plaintiff thus actually favors the defendant.

So the plaintiff had no case (and for the further reason that the employer had a compelling noninvidious ground for terminating the plaintiff--the work it had hired him to do, the work for which his experience qualified him, had dried up within weeks of his arrival), and summary judgment was rightly granted for the employer. But we wish to make three observations about the record, for such bearing as they may have on future discrimination cases.

1. The supervisor who riffed the plaintiff was even older than the plaintiff--56--and dicta in a number of cases suggest that this is a factor that should weigh heavily against a finding of age discrimination. See, e.g., Fairchild v. Forma Scientific, Inc., 147 F.3d 567, 572 (7th Cir. 1998); Mills v. First Federal Savings & Loan Ass'n, 83 F.3d 833, 842 (7th Cir. 1996); Wexler v. White's Fine Furniture, Inc., supra, 246 F.3d at 866-67; Brown v. CSC Logic, Inc., 82 F.3d 651, 658 (5th Cir. 1996); Rothmeier v. Investment Advisers, Inc., 85 F.3d 1328, 1337 (8th Cir. 1996); see also Elrod v. Sears, Roebuck & Co., 939 F.2d 1466, 1471 (11th Cir. 1991); but cf. Rea v. Martin Marietta Corp., 29 F.3d 1450, 1456 (10th Cir. 1994). On reflection, we offer the counterdictum that the relative ages of the terminating and terminated employee are relatively unimportant. For it is altogether common and natural for older people, first, to exempt themselves from what they believe to be the characteristic decline of energy and ability with age; second, to want to surround themselves with younger people; third to want to protect their own jobs by making sure the workforce is not too old, which might, if "ageist" prejudice is rampant, lead to RIFs of which they themselves might be the victims; and fourth, to be oblivious to the prejudices they hold, especially perhaps prejudices against the group to which they belong. We emphatically rejected the "same-actor inference" in the race-discrimination setting in Johnson v. Zema Systems Corp., 170 F.3d 734, 745 (7th Cir. 1999), and our conclusion there applies with equal force to proof of age discrimination.

But it is eminently reasonable to doubt that, as in this case, a worker hired at an age well beyond that at which the protections of the age discrimination law click in and terminated within months, that is, before he is appreciably older, was a victim of age discrimination. Rand v. CF Industries, Inc., 42 F.3d 1139, 1147 (7th Cir. 1994); LeBlanc v. Great American Ins. Co., 6 F.3d 836, 847 (1st Cir. 1993); Lowe v. J.B. Hunt Transport, Inc., 963 F.2d 173, 174-75 (8th Cir. 1992). A company that didn't want 54- year-olds on its payroll would be unlikely to hire one rather than to hire one and promptly fire him, thus inviting a lawsuit since terminated workers are much more likely to sue than ones who have merely not been hired, owing to the much greater difficulty of proving damages in the latter case. A worker who is not hired will find it hard to prove that he would have received a much higher salary than in his present job--for why would a new employer have paid him so much more than his market rate? See Olin v. Prudential Ins. Co., 798 F.2d 1, 4-5 (1st Cir. 1986), overruled on other grounds, Gallagher v. Wilton Enterprises Inc., 962 F.2d 120 (1st Cir. 1992) (per curiam).

2. Some cases suggest that statistical evidence is not admissible to show discrimination unless it is significant at the conventional 5 percent significance level (that is, the coefficient of the relevant correlation is at least two standard deviations away from zero), e.g., Anderson v. Douglas & Lomason Co., Inc., 26 F.3d 1277, 1291 n. 26 (5th Cir. 1994); Ottaviani v. State University of New York, 875 F.2d 365, 371-72 (2d Cir. 1989); Segar v. Smith, 738 F.2d 1249, 1282-83 (D.C. Cir. 1984); cf. Bennett v. Total Minatome Corp., 138 F.3d 1053, 1062 (5th Cir. 1998) (significance level not specified)--in other words, unless there is no more than a 5 percent probability that we would observe a statistical correlation between the dependent variable (such as whether terminated) and the independent variable having legal significance (such as age) even if the variables were uncorrelated in the population from which the sample was drawn. Other cases--including our own--reject the suggestion. Bell v. EPA, 232 F.3d 546, 554 (7th Cir. 2000); MacDissi v. Valmont Industries Inc., 856 F.2d 1054, 1058 n. 3 (8th Cir. 1988); Heller v. Shaw Industries Inc., 167 F.3d 146, 158 (3d Cir. 1999); Waisome v. Port Authority of New York & New Jersey, 948 F.2d 1370, 1376 (2d Cir. 1991). The 5 percent test is arbitrary; it is influenced by the fact that scholarly publishers have limited space and don't want to clog up their journals and books with statistical findings that have a substantial probability of being a product of chance rather than of some interesting underlying relation between the variables of concern. Litigation generally is not fussy about evidence; much eyewitness and other nonquantitative evidence is subject to significant possibility of error, yet no effort is made to exclude it if it doesn't satisfy some counterpart to the 5 percent significance test. A lower significance level may show that the correlation is spurious, but may also be a result of "noise" in the data or collinearity (correlation between independent variables, such as sex and weight), and such evidence, when corroborated by other evidence, need not be deemed worthless. Conversely, a high significance level may be a misleading artifact of the study's design; and there is always the risk that the party's statistical witness ran 20 regressions, one and only one of which supported the party's position and that was the only one presented, though, in the circumstances, it was a chance result with no actual evidentiary significance. (Careful pretrial discovery by the other party should unmask this trick.)

But the question whether a study is responsible and therefore admissible under the Daubert standard is different from the weight to be accorded to the significance of a particular correlation found by the study. It is for the judge to say, on the basis of the evidence of a trained statistician, whether a particular significance level, in the context of a particular study in a particular case, is too low to make the study worth the consideration of judge or jury. Cf. Adams v. Ameritech Services Inc., 231 F.3d 414, 425 (7th Cir. 2000); Porter v. Whitehall Laboratories Inc., 9 F.3d 607, 611 (7th Cir. 1993).

3. Last we consider whether statistical evidence alone, supposing it now to be free from any doubts based on significance levels, can ever establish a prima facie case of intentional discrimination (disparate treatment). Some cases say yes, Walther v. Lone Star Gas Co., 977 F.2d 161 (5th Cir. 1992) (per curiam); MacDissi v. Valmont Industries Inc., supra, 856 F.2d at 1058; Chrisner v. Complete Auto Transit, Inc., 645 F.2d 1251, 1259 n. 7 (6th Cir. 1981), Davis v. Califano, 613 F.2d 957, 962 (D.C. Cir. 1980); Reed v. Lockheed Aircraft Corp., 613 F.2d 757, 762 (9th Cir. 1980); some no, Kidd v. Illinois State Police, 167 F.3d 1084, 1101 n. 16 (7th Cir. 1999); Smith v. Horner, 839 F.2d 1530, 1536 n. 8 (11th Cir. 1988); Carmichael v. Birmingham Saw Works, 738 F.2d 1126, 1131 (11th Cir. 1984)--though it is important to note that our own statements of this position, in Kidd and the cases cited there, are brief dicta probably best understood as comments on the weakness of the particular evidence presented in those cases rather than as considered attempts to decide whether statistical evidence might ever suffice in an extreme, an unusual case. Although it is unlikely that a pure correlation, say between age and terminations, would be enough to establish a prima facie case of intentional discrimination, it would be precipitate to hold that it could never do so. If 100 employees in a department of 1,000 employees were riffed and every one of the 100 was 40 years old or older and every one of the 900 retained was under 40, that would, we should think, be enough evidence of age discrimination (the probability of its occurring by chance being inconceivably minute) to place on the employer a burden of explaining, which is all that making out a prima facie case means. But that cannot help the plaintiff in this case, the facts of which are wildly different.

Affirmed.