Tisa Crawford brought this Title VII suit against her former employer, lost when the judge granted summary judgment for the defendant, and appeals. She worked as a "conductor"—the job that used to be called "trainman" or "brakeman"—which meant operating the switches that move the tracks, coupling and uncoupling the rail cars, and warning the engineers about derailment dangers. She was fired after receiving eight reprimands in her first year of work: four reprimands for not reporting for duty, two for failing to properly stabilize a boxcar (and in one of these incidents the boxcar partially derailed and was damaged), one for failing to wear safety glasses, and another one for riding on a rail car in an unsafe manner. She is a black woman— the only one employed by the defendant— and claims that she was discharged because of her race and sex.

Obviously she was not a model employee. But if equally bad or worse white men employed by the defendant as conductors (no white women were employed in that position) were retained despite deficiencies as serious as hers, and the employer failed at the summary judgment stage to offer a nondiscriminatory explanation for the difference in treatment or it did but the plaintiff presented evidence that the explanation was a pretext (that is, false), then she was, prima facie, a victim of discrimination. Flores v. Preferred Technical Group, 182 F.3d 512, 514-15 (7th Cir.1999); Bush v. Commonwealth Edison Co., 990 F.2d 928, 931 (7th Cir. 1993); see also Rodgers v. U.S. Bank, N.A., 417 F.3d 845, 852-53 (8th Cir.2005). ("Prima facie" because the defendant might at trial be able to show that the plaintiff was not fired because of her race. All that "prima facie" means is strong enough to preclude summary judgment for the defendant. Kovacevich v. Kent State University, 224 F.3d 806, 821 (6th Cir. 2000).)

But this assumes that the better-treated workers with whom the plaintiff compares herself are a representative sample of all the workers who are comparable to her. Bush v. Commonwealth Edison Co., supra, 990 F.2d at 931-32; see also Turgeon v. Premark Int'l, Inc., 87 F.3d 218, 221-22 (7th Cir.1996). She must not pick and choose. We know that the defendant employs about 200 conductors and that it fired 10 of them, besides the plaintiff, for safety and other infractions in a two-year period that includes the one-year period in which she worked. Her lawyer insisted at argument that two other conductors whose infractions were as serious as the plaintiff's were not fired. Suppose then that 12 male employees—the 10 who were fired and the 2 who were not—were comparable to the plaintiff; then 5/6 of the comparable males were treated as badly as she was. This means that 100 percent of the "bad" black female workers were fired and 83 percent of the "bad" white males, but since there was only one worker in the first class, namely the plaintiff, the percentage could not be less than 100 percent. Since perfect enforcement of company rules is hardly to be expected, the fact that "only" 83 percent of the "bad" white men were fired does not support an inference that the defendant treats white men better than black women.

The plaintiff says that the number of comparable white men who were treated better than she was 10 rather than 2—the 2 were just the most egregiously bad of the bad male workers, and if that is right then the percentage of bad men who were fired was only 50 percent, and this begins to suggest that men were treated more leniently than the plaintiff was. But only if the men who were retained really were no better than she, a judgment dependent on their being similarly enough situated to her to make comparison meaningful.

There has been a tendency in our cases, and in those of some other circuits as well (a trend resisted, however, by the Eighth Circuit, see Rodgers v. U.S. Bank, N.A., supra, 417 F.3d at 851-52 (8th Cir.2005); Wheeler v. Aventis Pharmaceuticals, 360 F.3d 853, 857 (8th Cir.2004)), to require closer and closer comparability between the plaintiff and the members of the comparison group (the group of 10 in this case). Brummett v. Sinclair Broadcast Group, Inc., 414 F.3d 686, 694 (7th Cir. 2005); Sartor v. Spherion Corp., 388 F.3d 275, 279-80 (7th Cir.2004); Adams v. Wal-Mart Stores, Inc., 324 F.3d 935, 940 (7th Cir.2003); Peele v. Country Mutual Ins. Co., 288 F.3d 319, 330 (7th Cir.2002); Knight v. Baptist Hospital of Miami, Inc., 330 F.3d 1313, 1316-18 (11th Cir.2003); Okoye v. University of Texas Houston Health Science Center, 245 F.3d 507, 514 (5th Cir.2001) ("nearly identical"). The requirement is a natural response to cherry-picking by plaintiffs, the issue with which we began. If a plaintiff can make a prima facie case by finding just one or two male or nonminority workers who were treated worse than she, she should have to show that they really are comparable to her in every respect.

But if as we believe cherry-picking is improper, the plaintiff should have to show only that the members of the comparison group are sufficiently comparable to her to suggest that she was singled out for worse treatment. Goodwin v. Board of Trustees of University of Illinois, 442 F.3d 611, 619 (7th Cir.2006); Ezell v. Potter, 400 F.3d 1041, 1049-50 (7th Cir.2005). Otherwise plaintiffs will be in a box: if they pick just members of the comparison group who are comparable in every respect, they will be accused of cherry-picking; but if they look for a representative sample, they will unavoidably include some who were not comparable in every respect, but merely broadly comparable. The cases that say that the members of the comparison group must be comparable to the plaintiff in all material respects get this right. E.g., Raymond v. Ameritech Corp., 442 F.3d 600, 610 (7th Cir.2006); Wyninger v. New Venture Gear, Inc., 361 F.3d 965, 979 (7th Cir.2004); Durkin v. City of Chicago, 341 F.3d 606, 613-14 (7th Cir.2003); Rivera-Aponte v. Restaurant Metropol # 3, Inc., 338 F.3d 9, 12 (1st Cir.2003); Graham v. Long Island R.R., 230 F.3d 34, 39-40 (2d Cir.2000).

The defendant emphasizes that members of the comparison group had worked longer for the defendant, and had different supervisors, than the plaintiff did. Length of service is particularly relevant if the comparison group is selected on the basis of the total number of reprimands received by each of the members regardless of length of service; a worker who received one reprimand in 10 years would not be comparable to a worker who had received one reprimand in one year. Length of service would be much less relevant if instead the comparison group is defined by reference to the number of reprimands received by each member during the year in which the plaintiff was employed, unless the compared workers had a sterling record before the comparison year. The plaintiff used the first method, and failed to compare the overall safety and absentee records of the members of the comparison group with the plaintiff's record, even though as we just noted two workers with the identical number of total infractions might, given different lengths of service, have a very different number of average infractions. Moreover, one of the members of the comparison group was not a conductor.

A difference in supervisors is important in evaluating a worker's record of reprimands when the supervisors who issue the reprimands have broad discretion (the equivalent of prosecutorial discretion) in deciding whether and when to do so, as assumed in the many cases, such as Radue v. Kimberly-Clark Corp., 219 F.3d 612, 617-18 (7th Cir.2000), that emphasize such a difference as a material circumstance in determining comparability. Here there were multiple supervisors—the plaintiff's eight reprimands were issued by four different supervisors—and the plaintiff failed to show how much or how little discretion they had.

So summary judgment was properly granted.

AFFIRMED.