dissenting:
Trying to avoid summary judgment, plaintiffs (five male professors) simply threw rocks at VCU’s salary equity study, rocks that either missed or glanced off harmlessly. On the record before us, VCU’s multiple regression study establishes that there was a manifest imbalance in pay between men and women faculty members (women were underpaid) at that state university. Plaintiffs argue that VCU’s study should have included more variables, but they fail to demonstrate that the inclusion of those variables would have eliminated the statistically significant effect of gender on salaries. Thus, plaintiffs’ response does not satisfy .either Supreme Court precedent (Bazemore) or settled summary judgment principles. With all respect for the majority, I therefore dissent because there is no material fact.in dispute.
I.
VCU’s study (a multiple regression analysis) mathematically estimated the effect of eight independent variables on salaries VCU paid to 770 tenured and tenure-track faculty members. The eight independent variables were: (1) national salary average (same discipline and rank), (2) doctorate or not, (3) *664tenure status, (4) quick tenure (within four years of appointment) or not, (5) years of experience at VCU, (6) academic experience before VCU, (7) service, if any, as department chair, and (8) gender. When the effect of the seven variables other than gender was taken into account, the effect of gender on faculty salaries was a statistically significant $1,354 in 1989 and $1,982 in 1991. Specifically, women faculty members, because they were women, on average were paid that much less than men. These findings were confirmed by VCU’s expert statistician, Dr. Rebecca Klemm. Dr. Klemm testified (in deposition) that she took VCU’s raw data and ran new regression studies using models different from the one used by VCU. Dr. Klemm’s new studies consistently showed a gender difference in salary at VCU of a magnitude “very similar” to that found by VCU’s original model. In moving for summary judgment, VCU relied on its original regression study and Dr. Klemm’s testimony to establish the existence of a manifest imbalance in pay tied to the impermissible factor of gender.
Against this specific evidence, plaintiffs offered the deposition testimony and an affidavit (filed post-judgment) from Dr. Fred McChesney, an economics professor from Emory University. Unlike VCU’s expert, Dr. McChesney did not perform a statistical study in this ease. Indeed, he admitted that he had never performed a pay study such as the one at issue here. Nevertheless, Dr. McChesney opined that VCU’s model should have included additional independent variables, such as performance1 and any prior service as an administrator. This omission was a fatal flaw, he said. When pressed, however, Dr. McChesney agreed that gender could affect salary at VCU. He also admitted that, even if the variables he suggested had been included, it was possible that the study results would have been essentially the same. In short, Dr. McChesney could not say that adding his variables would have had a statistically significant effect on the results of the VCU study.
On this record the majority finds a genuine issue of material fact. In doing so, the majority gives Dr. McChesney far too much credit. Dr. McChesney’s opinion does not justify the majority’s conclusions that “failure to include a faculty member’s status as a former administrator could easily have caused a salary differential that was not attributable to sex,” ante at 663 (emphasis mine), and that “inclusion of the performance factors could very well alter the results of the multiple regression analysis,” id. at 663 (emphasis mine).
Dr. McChesney’s untested conjecture does not undermine the validity of VCU’s multiple regression analysis. To avoid summary judgment, plaintiffs must do more than list a few other variables that might have been included in the regression analysis. Instead, plaintiffs must demonstrate that including an omitted variable would have eliminated any statistically significant disparity between the salaries of men and women. Bazemore v. Friday, 478 U.S. 385, 399-403, 106 S.Ct. 3000, 3008-3010, 92 L.Ed.2d 315 (1986).
Bazemore is controlling here. In Baze-more, certain employees of the North Carolina Agricultural Extension Service sued various state and local officials, alleging that discriminatory differences in pay between black and white workers violated Title VII of the Civil Rights Act of 1964. 478 U.S. at 391, 106 S.Ct. at 3004. To help prove discrimination, the plaintiff-employees offered multiple regression analyses, which both the district court and our court rejected. The Supreme Court took the case and said that our court had erroneously concluded that “ ‘[a]n appropriate regression analysis of salary should ... include all measurable variables thought to have an effect on salary level’ ” Id. at 399, 106 S.Ct. at 3008 (quoting Bazemore v. Friday, 751 F.2d 662, 672 (4th Cir.1984) (alterations in original)). After pointedly noting that the employees’ expert testified that the disparities shown by the regressions were statistically significant, a unanimous Supreme Court said:
The [Fourth Circuit’s] view of the eviden-tiary value of the regression analyses was *665plainly incorrect. While the omission of variables from a regression analysis may render the analysis less probative than it otherwise might be, it can hardly be said, absent some other infirmity, that an analysis which accounts for the major factors “must be considered unacceptable as evidence of discrimination.” Normally, failure to include variables will affect the analysis’ probativeness, not its admissibility.
Importantly, it is clear that a regression analysis that includes less than “all measurable variables” may serve to prove a plaintiffs case. A plaintiff in a Title VII suit need not prove discrimination with scientific certainty; rather his or her burden is to prove discrimination by a preponderance of the evidence.
Id. at 399-400, 106 S.Ct. at 3009 (internal citations and footnote omitted).
The majority cites this passage from Baze-more for the undisputed proposition that “a study must include all major factors.” Ante at 663. The majority then goes on to conclude that the variables omitted here (such as performance) are “crucial” factors because it believes their “inclusion ... could very well alter the results of [VCU’s] multiple regression analysis.” Id. at 663 (emphasis mine). According to the majority, this creates a dispute of material fact “as to the validity of the study.” Id. The record does not support this conclusion.
Plaintiffs never demonstrated the statistical significance of the omitted variables, and this is the fatal flaw in their response. It is fatal because Bazemore teaches that statistical significance must be the wedge that divides “major” (or “crucial”) factors from other “measurable” factors. The Bazemore Court criticized the governmental respondents for a trial strategy that made no attempt “ — statistical or otherwise — to demonstrate that when these [allegedly important] factors were properly organized and accounted for there was no significant disparity between the salaries of blacks and whites.” 478 U.S. at 403 n. 14, 106 S.Ct. at 3010 n. 14. In addition, the Court noted three times that the regression analyses in the Bazemore record showed salary disparities that were “statistically significant.” Id. at 399 n. 9, 401, 404 n. 15,106 S.Ct. at 3008 n. 9, 3009, 3010 n. 15. Bazemore therefore sends a clear signal: “major” factors are statistically significant factors. In the case before us, the omitted variables cannot be characterized as “major” because plaintiffs did not demonstrate that their inclusion would eliminate the statistically significant disparity between the salaries of men and women at VCU. Thus, under Bazemore the district court was correct in rejecting plaintiffs’ claim.
II.
Even without Bazemore, VCU is entitled to summary judgment. Under settled Rule 56 jurisprudence, a defendant is entitled to judgment as a matter of law when the plaintiff fails to adduce facts on an element of his case for which he bears the burden of proof. Celotex Corp. v. Catrett, 477 U.S. 317, 322, 106 S.Ct. 2548, 2552, 91 L.Ed.2d 265 (1986); Anderson v. Liberty Lobby, Inc., 477 U.S. 242, 248-49, 106 S.Ct. 2505, 2510, 91 L.Ed.2d 202 (1986). In this case, the majority acknowledges that plaintiffs bear the burden of establishing that VCU’s affirmative action plan was implemented in the absence of a manifest imbalance in salary between male and female faculty members. Ante at 662. To meet this burden, plaintiffs simply assert that VCU’s multiple regression analysis failed to include enough factors. But, as I have already pointed out, plaintiffs have not shown the statistical significance of even one of the factors they say was erroneously excluded from VCU’s study.2 In fact, their expert, Dr. McChesney, indicated that he did not know whether including the factors he identified would change the result of VCU’s studies in a statistically significant way. To avoid summary judgment, plaintiffs must do more. The complaining male professors must produce firm evidence showing that, after adjusting for an omitted factor, a statistically significant gender effect on salary no longer exists. Nothing else can create a *666genuine issue of material fact concerning the study’s validity.
This case should be decided on the record. The record establishes that women faculty members at VCU were paid less because they were women. VCU made adjustments for the disparity based on a study that has not been discredited. The district court’s award of summary judgment to VCU should be affirmed.
. VCU maintains that the variables in its study relating to tenure, experience and any service as a department chair are “broad measures of performance.”
. The district court put it this way: "no record evidence exists that including these additional variables would have altered the results of VCU’s study.” Smith v. Virginia Commonwealth University, 856 F.Supp. 1088, 1093 n. 15 (1994).