Alone among the thirty or so employees in her department, Lena Barricks did not receive a raise in 2003. Barricks, who had worked as a chemical operator at Eli Lilly and Company (“Lilly”) since 1977, thought that discrimination was behind this, so after retiring in 2004 she sued her former employer for age and gender discrimination. The district court granted summary judgment to Lilly and Barricks appeals. Because Barricks cannot show that Lilly’s stated reason for declining to give the raise — her performance — is a pretext for discrimination, we affirm the judgment of the district court.
I. BACKGROUND
Lilly employs a somewhat involved methodology to determine which employees in Barricks’s department should receive raises (or “merit increases” as Lilly calls them). In response to interrogatories and through the testimony of the department’s human resources representative, Lilly explained that the process begins with the employee’s performance evaluation from the previous year, which is determined by the shift supervisor with limited input from other members of management. The evaluation includes a number from one (lowest) to five, which is fed into a computer algorithm along with information about the employee’s current salary level and the overall budget for raises. The computer produces for each employee a “range of allowable merit increases” — for instance, between $50 and $100 per month — from which the shift supervisor, the human resources manager, and the department head decide on a raise. They begin with the range midpoint — $75 in the above example — and give exemplary employees raises toward the high end of the range, and weaker employees raises toward the low end, offsetting any dollar amounts above the midpoint with lower-than-midpoint raises. In other words, if the hypothetical employee with a midpoint of $75 received a raise of $80 per month, the $5 per month “deficit” above the midpoint would be offset by giving another employee a raise $5 below that employee’s midpoint. Beginning in 2002, the department also instituted an unwritten policy of declining to give raises of $20 per month or less, because of what the human resources manager called an “insult factor” — an employee might prefer no raise at all to a very small one.
The present lawsuit is confined to Lilly’s decision not to give Barricks a raise in 2003. In her performance evaluation for 2002, upon which the decision was largely based, she received an overall rating of two out of five. (Lilly stated in response to Barricks’s administrative complaint that it was “a low level 2 performance” — in other words, just above a one.) The evaluation summary notes four “hits,” such as the fact that she had no infractions during the year and trained new chemical operators, and four “misses,” including a need to focus on computer and communication skills. The evaluation also listed three satisfactory “performance behaviors,” but four where improvement was needed, including “create external focus,” “anticipate changes and prepare for the future,” and “achieve results with people.” The evaluation noted that based on her review, Bar-ricks was eligible for a raise but a disclaimer stated that she was not guaranteed one.
*559The computer produced a range between $0 and $30 per month for Barricks’s raise. Based on her low evaluation, her high pay grade level (34 out of a possible 36 for her position), the need to offset other raises above employees’ midpoints, and the $20 de minimis policy, Barricks’s supervisors did not give her a raise for .2003. Barricks testified in a deposition that in her twelve years as a senior chemical operator, she had received a raise six times. She filed suit but the district court granted summary judgment to Lilly on unspecified grounds, and this appeal followed.
II. ANALYSIS
It is said that you can find a statistic to prove anything. In Lilly’s view of the case, four of the five women in Barricks’s department received raises for 2003, as did ten of the eleven employees over age fifty. But Barricks points out that none of the four women received raises above their midpoints, while many of the male employees did. On the other hand, the four women received performance ratings of three, and a raise below the midpoint was common for the men with threes. Depending on the statistic under consideration, discrimination was either perfectly obvious or utterly nonexistent.
We have frequently discussed the dangers of relying on raw data without further analysis or context in employment discrimination disputes. See Hemsworth v. Quotesmith.com, Inc., 476 F.3d 487, 491-92 (7th Cir.2007); Hull v. Stoughton Trailers, LLC, 445 F.3d 949, 951-52 (7th Cir.2006); Radue v. Kimberly-Clark Corp., 219 F.3d 612, 616, 619 (7th Cir.2000). So, rather than play the numbers game, we review the district court’s decision within the McDonnell Douglas burden-shifting framework, by which Barricks has elected to attempt to prove discrimination indirectly. See McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802-04, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973). Under this familiar approach, Barricks must first make out a prima facie case of discrimination by showing that (1) she is a member of a protected class; (2) her performance met her employer’s legitimate expectations; (3) despite this performance, she was subjected to an adverse employment action; and (4) her employer treated similarly situated employees outside of the protected class more favorably. Ptasznik v. St. Joseph Hosp., 464 F.3d 691, 696 (7th Cir.2006). If she succeeds, the burden shifts to the employer to articulate a legitimate, nondiscriminatory reason for its decision, which the plaintiff can then attack as a pretext for discrimination. Id. This approach applies to claims of gender discrimination under Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e-2(a)(1), as well as claims under the Age Discrimination in Employment Act of 1967, 29 U.S.C. § 623. Raymond v. Ameritech Corp., 442 F.3d 600, 610 (7th Cir.2006). We review the district court’s grant of summary judgment de novo. Jackson v. County of Racine, 474 F.3d 493, 498 (7th Cir.2007).
Lilly concedes that as a woman over the age of forty, Barricks is a member of protected classes, and that she was meeting the company’s expectations. (Her ratings were low, but they were not unacceptable, it says.) Lilly also does not dispute that the denial of a raise — as opposed to missing out on something more transient, like a bonus — qualifies as an adverse employment action. See Farrell v. Butler Univ., 421 F.3d 609, 614 (7th Cir.2005); Hildebrandt v. Ill. Dep’t of Natural Res., 347 F.3d 1014, 1030 (7th Cir.2003); Hunt v. City of Markham, 219 F.3d 649, 654 (7th Cir.2000). That leaves the fourth factor, whether similarly situated employees who *560were not members of the protected class were treated more favorably. The “similarly situated” test is a flexible, commonsense inquiry whose requirements vary from case to case. Humphries v. CBOCS West, Inc., 474 F.3d 387, 404-05 (7th Cir.2007). Its purpose is to determine whether there are enough common factors between a plaintiff and a comparator — and few enough confounding ones — to allow for a meaningful comparison in order to divine whether discrimination was at play. Id. at 406; see also Keri v. Bd. of Trs. of Purdue Univ., 458 F.3d 620, 644 (7th Cir.2006) (plaintiff must demonstrate that there is someone directly comparable in all material respects).
Barricks’s claim of age discrimination fails at this point in the analysis. She makes no attempt to point to a younger employee who was similarly situated and received more favorable treatment, and instead relies on the fact that everyone besides her got a raise. The record does not disclose the age of most of the other employees, so it is impossible to know whether or not they are members of her protected class. Barricks mentions the age of a handful of employees in passing, but they were not supervised by Platt, who, Bar-ricks concedes, played the decisive role in giving performance evaluations and determining raise amounts. See Radue, 219 F.3d at 618 (noting importance of showing common supervisor, because different supervisors make employment decisions in different ways).
On her claim of gender discrimination, Barricks identifies one potential comparator, Lawrence Swick, who was also a senior chemical operator, shared Bar-ricks’s pay grade and level of experience, and was supervised by Platt. Lilly notes that in 2002 Swick received a performance rating of three compared to Barricks’s two, and contends that he is therefore an inadequate comparator. See Anders v. Waste Mgmt. of Wis., Inc., 463 F.3d 670, 676 (7th Cir.2006). Given the important role of the evaluation in Lilly’s calculation of raises, Barricks should have included in the record Swick’s evaluation so that we could rule out performance as the basis for Swick’s receiving a raise and Barricks’s disappointment. Nevertheless, we do know from the number ratings and from deposition testimony that Swick’s performance, like Barricks’s, was relatively low. Platt explained that Swick had been subject to various disciplinary infractions that had affected his eligibility for a raise. Even though Barricks’s low performance rating stemmed from deficiencies in her work rather than disciplinary issues, there is no evidence that the computer recognized this difference. See Hill, 445 F.3d at 952 (comparator must have performance history “somewhat comparable” to plaintiffs). Since Swick received a raise in 2003 and Barricks did not, he received more favorable treatment. See Bio v. Federal Express Corp., 424 F.3d 593, 598 (7th Cir.2005).
Even assuming that Swick and Barricks are similarly situated, however, Barricks has failed to show that Lilly’s stated reason for denying her a raise was pretextual. The focus of a pretext inquiry is whether the employer’s reason is honest, not whether it is accurate or wise. Ptasznik, 464 F.3d at 696; see also Burks v. Wis. Dep’t of Transp., 464 F.3d 744, 754-55 (7th Cir.2006) (“An employee’s attempt to avoid summary judgment cannot succeed unless the employee puts forth evidence suggesting that the employer itself did not believe the reasons for [the adverse employment action].”). Lilly states that given Barricks’s potential raise range of $0-30 per month, she would have needed a raise of at least 67% of her maximum possible raise in order to make it over the
*561$20 de minimis level. (By contrast, Swick received only 60% of his maximum possible raise, and his performance evaluation was higher than Barricks’s.1) Barricks does not argue on appeal that the $20 policy was bogus, although she does characterize it in the facts section of her brief as “the invisible $20 ‘policy.’ ” She produced no evidence, however, not even her own affidavit, to suggest that the policy was a fabrication. In fact, at one point in her deposition she conceded that when in the past she and a female colleague received raises of $20 or $30 per month, it was “pretty insulting,” which tends to support Lilly’s basis for instituting the policy. If she wished to create a genuine issue of material fact as to the policy’s existence, she would have to produce at least some evidence or argumentation. See Piraino v. Int’l Orientation Resources, Inc., 84 F.3d 270, 274-75 (7th Cir.1996) (fact issue as to existence of employer’s unwritten pregnancy policy where managers did not mention policy in discussions with pregnant employee and then suddenly produced written directive codifying allegedly informal policy); Sarsha v. Sears, Roebuck & Co., 3 F.3d 1035, 1040 (7th Cir.1993) (fact issue as to existence of company’s no-dating policy where plaintiff had met ex-wife while both worked for defendant, and defendant had even thrown wedding party for the two); see also EEOC v. Yellow Freight Sys., Inc., 253 F.3d 943, 957-58 (7th Cir.2001) (en banc) (Wood, J., dissenting in part and concurring in part) (discussing evidence needed to create issue on existence of unwritten policy).
Instead of challenging the $20 policy, Barricks argues that Lilly has inconsistently described the role of various managers in the merit increase process, and that this proves pretext. Specifically, she contends that at times, Lilly stated that the shift supervisor alone made the key decisions, and at other times that he decided in concert with others. We find no inconsistency. In the pleadings and during discovery, Lilly explained that the shift supervisor has primary responsibility for issuing a performance evaluation, and then in conjunction with the human resources representative and the department head, the shift supervisor selects a raise amount from the computer generated range.
Barricks also argues that Lilly intimidated one of her female co-workers from signing an affidavit in support of Bar-ricks’s motion for summary judgment, and that this shows that the company knew that its case was weak. But the affidavit, even if it had been signed and admitted as evidence, would not have affected the result at summary judgment, for its assertions, including a claim that Platt expected female employees to do more “housekeeping” (i.e., cleaning and maintenance) work than men, made it into evidence in other ways, including through Barricks’s own affidavit. Cf. Great Am. Ins. Co. v. Horab, 309 F.2d 262, 264-65 (8th Cir.1962) (Blackmun, J.) (where interference with witness did not affect result, excluding evidence of interference was not abuse of discretion). Finally, Barricks contends that Platt was reprimanded after pornography was found on his computer and a book of off-color jokes was found in his desk, and that this proves that he denied her a raise because of her gender. But she does not explain how these vague transgressions had anything to do with the decision not to offer her a raise. See Rozskowiak v. Village of Arlington Heights, 415 F.3d 608, 612-13 (7th Cir.2005); Sanghvi v. St. Catherine’s Hosp., Inc., 258 F.3d 570, 575-76 (7th Cir.2001); Schaffner v. Glencoe Park Dist., 256 F.3d 616, 622-23 (7th Cir.2001). In*562deed, some of Barricks’s allegations concern Platt’s inappropriate conduct after he was transferred to another facility and ceased to be her supervisor, so those could hardly bear on his role in the decision. In short, Barricks has failed to present enough evidence to cast the shadow of pretext on Lilly’s explanation for denying the raise. See Keri, 458 F.3d at 645.
III. CONCLUSION
For these reasons, the district court’s decision granting summary judgment to Lilly is AjffiRMEd.