After nine months of trial in which were examined 58 individual claims of sex discrimination in salary, promotion, and tenure practices by the Oregon state system of higher education, the court granted individual relief in three claims, but found against the class plaintiffs and denied all other claims against the state board of higher education. The plaintiffs appeal as a class, and individually. We affirm.
Twenty-two women faculty members sued the defendant board under 42 U.S.C. §§ 2000e et seq. and 28 U.S.C. § 1983. The Title VII claims were tried. The § 1983 claims were dismissed during pretrial proceedings. The case proceeded as a “pattern and practice” disparate treatment case, comparing the board’s treatment of women faculty and male faculty in terms of rank, pay, promotion, tenure and administrative appointments. Statistical evidence played a major part in the trial.
The trial judge prepared a 487 page memorandum addressing the factual and legal issues. Within a document of that size, disappointed litigants are likely to find a number of points with which to disagree. This case is no exception.
The plaintiffs’ failures of proof were not for want of effort. The trial produced 25,000 pages of testimony and rooms full of exhibits. The subject matter ranged over the history of women in academic callings. The court heard about the economics of supply and demand, and the depressing effect of tenured, high salaried senior faculty, who are frequently older males, upon the salaries of junior faculty, who are frequently females, during times of taxpayer parsimony and shrinking budgets.
*461After the parties rested, the court determined that the state system of higher education had not intentionally discriminated against women. There is no substantial dispute that historical disparity existed. However, historical disparity does not give rise to a successful Title VII claim. The court found that, at worst, the state’s efforts to bring its female employees into parity with its male employees had been impeded more by external economic factors than by lack of effort by the board to redress historic imbalances.
Because we are satisfied that the trial court’s findings of fact are not subject to reversal as clearly erroneous under Fed.R. Civ.P. 52(a), see Anderson v. City of Bessemer, 470 U.S. 564, 573-75, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985), and because we can conceive of no useful purpose in detailed rehearsal of the evidence, we will take up as briefly as possible the principal legal questions presented by the appeal.
I. BURDENS OF PROOF, PERSUASION AND PRODUCTION
Plaintiffs assert that the court erred in allocating and applying the respective burdens in the class case. They contend that the court required too much of them and too little of the board. In essence, their contention is that the court erroneously neglected to adhere strictly to the standards for the order and allocation of proofs set out in McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973). However, the McDonnell Douglas tripartite standard does not inflexibly apply to class action disparate treatment cases. See United States Postal Service Board of Governors v. Aikens, 460 U.S. 711, 715, 103 S.Ct. 1478, 1481, 75 L.Ed.2d 403 (1983); Texas Dep’t of Community Affairs v. Burdine, 450 U.S. 248, 253-54 n. 6, 101 S.Ct. 1089, 1094 n. 6, 67 L.Ed.2d 207 (1981); Gay v. Waiters’ and Dairy Lunchmen’s Union, 694 F.2d 531, 538 (9th Cir.1982); B. Schlei & P. Grossman, Employment Discrimination Law, 244 (2d ed. Supp.1985).
Indeed, by the time a fully-tried class action disparate treatment case reaches appeal, we need not consider whether the parties carried their respective burdens at the appropriate stages unless prejudicial errors, infecting distinct legal rulings, prevented a fair trial. Aikens, 460 U.S. at 715, 103 S.Ct. at 1481; Rios v. The Board of Regents of the University of Arizona, 811 F.2d 1248, 1249 (9th Cir.1987); Benzies v. Illinois Dep’t. of Mental Health and Developmental Disabilities, 810 F.2d 146, 148 (7th Cir.1987); Gottlieb v. Tulane University of Louisiana, 809 F.2d 278, 282 (5th Cir.1987).
In Bazemore v. Friday, — U.S. -, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986), a disparate treatment class action case, the Supreme Court responded to plaintiffs’ assertion that the McDonnell Douglas order and allocation of proof had not been adhered to as follows:
f the defendants have not succeeded in having a case dismissed on the ground that plaintiffs have failed to establish a prima facie case, and have responded to the plaintiffs’ proof by offering evidence of their own, the factfinder must then decide whether the plaintiffs have demonstrated a pattern or practice of discrimination by a preponderance of the evidence. This is because the only issue to be decided at that point is whether the plaintiffs have actually proved discrimination____ This determination is subject to the clearly erroneous standard on appellate review.
106 S.Ct. at 3008 (citations omitted). Thus, the only relevant burden in a class action disparate treatment case, assuming plaintiffs can avoid Rule 41 dismissal, is upon plaintiffs to prove the existence of discriminatory intent.
In this action, at the close of plaintiffs’ case-in-chief, the court denied the board’s motion for Rule 41 dismissal and allowed the board to introduce rebuttal evidence challenging plaintiffs’ statistical and other evidence. The district court then evaluated the evidence to determine whether plaintiffs had demonstrated, by a preponderance, the existence of either a pat*462tern or practice of discrimination or systematic purposeful discrimination. None was found. We hold that in reaching its decision, the district court placed appropriate burdens on plaintiffs for proof of discriminatory treatment.
II. CONSIDERATION OF ANECDOTAL, STATISTICAL, AND NON-STATISTICAL PROOF AND THEIR INTERRELATIONSHIP
Plaintiffs assert that the judge failed to consider their nonstatistical (or “anecdotal”) evidence in the individual and class claims, and that she failed to consider the statistical evidence in the individual cases. Plaintiffs argue that the judge failed to comprehend the interplay between the statistical evidence and the testimony of individual plaintiffs and other class members.
A. Individual Claims
When summarizing her conclusions on the individual claims, the judge noted that each individual plaintiff provided sufficient testimony to require the board to come forward with legitimate nondiscriminatory reasons for its actions. She then wrote:
Considering the quality and quantity of the Board’s explanations, each plaintiff should have come forward with evidence to show that the reasons offered by the Board were a pretext for sex discrimination. In order to carry their ultimate burden of persuasion, plaintiffs had to show that the Board’s explanations were offered to conceal the Board’s true purpose of sex discrimination. This portion of the plaintiffs’ individual cases was entirely inadequate. Plaintiffs offered almost no evidence to impeach the Board’s explanations of the allegedly illegal employment conditions. Without rebuttal testimony to establish that the Board’s reasons were a pretext for sex discrimination, the named plaintiffs’ individual cases, in most instances, had to fail.
While the McDonnell-Douglas scheme is not ritualistic, and while the court was free to decide whether the Board intentionally discriminated against the plaintiffs based upon all of the evidence before it, the case in chief presented by each plaintiff was, for the most part, weak; and the rebuttal case presented by each plaintiff was, for the most part, nonexistent.
In most cases, only the named plaintiffs testified in support of her individual claim, and her testimony rarely did more than describe differences in employment conditions.
The quoted passage indicates that the court may have viewed each individual case within its own evidentiary compartment. But even if the trier did require evidence of pretext relating specifically to each individual plaintiff, the ultimate result of the proof was a failure to establish pretense.
Generally, there are three types of evidence that can be used to show pretext: (1) direct evidence of discrimination, such as discriminatory statements or admissions, (2) comparative evidence, and (3) statistics. B. Schlei & P. Grossman, Employment Discrimination Law 1314 (2d ed. 1983).
Plaintiffs strongly attack the judge’s reaction to their statistical evidence on the issue of pretext for the individual claims. Statistics may be used to show that a defendant’s articulated nondiscriminatory reason is pretextual. McDonnell Douglas, 411 U.S. at 805, 93 S.Ct. at 1825; Lowe v. City of Monrovia, 775 F.2d 998, 1008 (9th Cir.1986). However, while statistics are often probative in determining the existence of a pattern or practice of disparate treatment, their usefulness in proving pretext “depend[s] primarily upon their relevance to the specific decision affecting the individual plaintiff, [and not] those that affect plaintiff’s protected class in general.” Employment Discrimination Law, supra, at 1316.
Proof of general employment policies and practices may be particularly useful in showing pretext. Cooper v. Federal Reserve Bank of Richmond, 467 U.S. 867, 880, 104 S.Ct. 2794, 2801, 81 L.Ed.2d 718 (1984); McDonnell Douglas, 411 U.S. at 805, 93 S.Ct. at 1825; Sweat v. Miller Brewing Co., 708 F.2d 655, 658 (11th Cir.1983) (statistical evidence of employer’s *463general policy and practice concerning minority employment to be considered on remand). In addition, more general comparative evidence may permit a court to fairly draw an inference of pretext. Furnco Construction Corp. v. Waters, 438 U.S. 567, 580, 98 S.Ct. 2943, 2951, 57 L.Ed.2d 957 (1978) (district court entitled to consider the racial mix of the workforce); O’Connor v. Peru State College, 781 F.2d 632, 637 (8th Cir.1986) (inequities between men’s and women’s athletic programs considered); Holsey v. Armour & Co., 743 F.2d 199, 208 (4th Cir.1984) (general atmosphere of discrimination and past practices may be considered), cert. denied, 470 U.S. 1395, 105 S.Ct. 1395, 84 L.Ed.2d 784 (1985).
In this action, the memorandum opinion would have been stronger had it expressly stated, when reviewing the evidence of pretext in the individual cases, that the district judge had reviewed (1) all of the statistical proof, (2) the evidence of sexist attitudes on the part of male administrators and (3) the degrees of compliance with affirmative action policies. It does not follow, however, that there must be a new trial. We think the memorandum opinion indicates that the judge did, in fact, consider the relevant pretext evidence of a statistical and anecdotal nature introduced by plaintiffs. She simply found the evidence unconvincing as proof of pretext.
B. Class Action
In a discriminatory treatment class action, plaintiffs have the burden of proving by a preponderance of the evidence that sex discrimination was defendant’s “standard operating procedure” maintained through systematic intentional discrimination. Bazemore, 106 S.Ct. at 3008; Cooper v. Federal Reserve Bank of Richmond, 467 U.S. 867, 875-76, 104 S.Ct. 2794, 2799, 81 L.Ed.2d 718 (1984); International Brotherhood of Teamsters v. United States, 431 U.S. 324, 336, 97 S.Ct. 1843, 1855, 52 L.Ed.2d 396 (1977); Spaulding v. University of Washington, 740 F.2d 686, 700 (9th Cir.), cert. denied, 469 U.S. 1036, 105 S.Ct. 511, 83 L.Ed.2d 401 (1984).
To establish the requisite discriminatory intent, plaintiffs may offer statistical, nonstatistical, and anecdotal evidence. However, even if a district court decides that plaintiffs have established prima facie evidence of intentional discrimination, for plaintiffs to prevail, the district court must still be persuaded by a preponderance of the evidence that actual actionable discrimination exists. Bazemore, 106 S.Ct. at 3008; Burdine, 450 U.S. at 257, 101 S.Ct. at 1095; Hazelwood School District v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977); Gay, 694 F.2d at 537 n. 4 (“it is clear that the burden which shifts to the defendant upon the establishment of a prima facie case is a burden of production only; the burden of persuasion remains with the plaintiff at all times”).
In ruling on the board’s motion to dismiss the class action, the judge indicated that she had examined the individual examples of discrimination to see whether they bolstered plaintiffs’ statistical case. In her introduction to the class action findings and conclusions, she explained:
Plaintiffs’ statistical evidence is the foundation for this claim [pattern or practice], although the court will also evaluate other evidence presented as it may bolster or detract from the statistical evidence____
The course that the court will follow in regard to the class action claim will be to evaluate the statistical evidence before it, keeping in mind the court’s rulings in the named plaintiffs’ cases, as well as the other evidence.
The court determined that the anecdotal evidence did not establish proof of a pattern or practice of discrimination even though it had established individual violations in three instances. The court noted when summarizing her conclusions on the class claim that:
Whatever the differences in employment conditions between men and women in the Oregon State System of Higher Education, they have been explained by the Board on legitimate, nondiscriminatory reasons, except in three instances.
*464We are satisfied that the court properly considered the relevant anecdotal evidence when it ruled on the class action.
Statistical evidence was the primary foundation of plaintiffs’ class action. Plaintiffs offered copious amounts of evidence, including multiple regression analyses, directed toward proving discriminatory intent. The board responded by challenging the accuracy and importance of plaintiffs’ evidence and by offering its own statistical evidence to show lack of discriminatory intent. The court found that there was a failure of proof as to the existence of discriminatory intent. Plaintiffs assign this finding as error and assert that the statistical evidence, both their own and defendant’s, proves a pattern or practice of sex discrimination “at every institution in every year studied in pay, rank, and tenure decisions.”
In response to statistical evidence presented in an effort to establish discriminatory intent, a defendant is always entitled to offer rebuttal evidence. Rebuttal evidence usually includes a explanation of statistical disparities based on neutral factors. It can, however, be limited to (1) a showing that the plaintiffs’ statistics are flawed, (2) a demonstration that the disparities generated or isolated by the statistics are not statistically significant or actionable, or (3) presentation by the defendant of statistical evidence contradicting that of the plaintiffs. See Teamsters, 431 U.S. at 339-40, 97 S.Ct. at 1856 (“statistics are not irrefutable; they come in infinite variety and, like any other kind of evidence, they may be rebutted”).
In the employment discrimination context, a defendant should also be entitled to challenge the ability of statistical evidence based on mathematical regressions to approximate the actual determinative factors in the employment decision-making process. See T.J. Campbell, Regression Analysis in Title VII Cases: Minimum Standards, Comparable Worth, and Other Issues Where Law and Statistics Meet, 36 Stan.L.Rev. 1299, 1305-1312 (1984) (arguing that courts should not rely on regression analyses whose “goodness of fit” or “adjusted R2” — a measurement of the degree to which the statistical model reflects actual decisionmaking — has been called into serious question). This concern is particularly important where courts examine academic and professional employment and advancement decisions which properly include a high regard for subjective personnel qualities and characteristics. See Spaulding, 740 F.2d at 704; Zahorik v. Cornell University, 729 F.2d 85, 92-94 (2nd Cir.1984); Green v. Board of Regents of Texas Tech. University, 474 F.2d 594, 596 (5th Cir.1973).
The court accepted and analyzed a prodigious amount of statistical evidence proffered by both plaintiffs and defendant in the forms of expert testimony, cross-examination, and documentary evidence. In the end, after reviewing the lengthy course of the trial, the district court decided that neither side’s statistical evidence accurately reflected the complex factors and motivations guiding the board’s employment, salary, and advancement decisions.
In Bazemore, the Supreme Court reversed a lower court decision that had deemed multiple regression analyses that did not include all measurable variables to be “unacceptable as evidence of discrimination.” 106 S.Ct. 3008-09. The Court seemed to address the problem presented here, namely, the weight properly accorded multiple regression evidence.
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 plaintiff’s 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 *465a preponderance of the evidence. Whether, in fact, such a regression analysis does carry the plaintiffs’ ultimate burden will depend in a given case on the factual context of each case in light of all the evidence presented by both the plaintiff and the defendant. However, as long as the court may fairly conclude, in light of all the evidence, that it is more likely than not that impermissible discrimination exists, the plaintiff is entitled to prevail.
106 S.Ct. at 3009 (citations and footnote omitted).
Bazemore, however, does not give blanket approval to the introduction of all evidence derived from multiple regression analyses. See id. 3009, n. 10 (“There may, of course, be some regressions so incomplete as to be inadmissible as irrelevant.”)
Plaintiffs assert that the district judge required too much of their statistical proof. We disagree. The district judge admitted statistical multiple regression evidence at trial and then chose to discount the weight of that evidence during her deliberations following trial. In her memorandum, the district judge expressly noted that in plaintiffs’ statistical analyses important decision-making variables were either missing or inadequately represented. Missing parts of the plaintiffs’ interpretation of the board’s decision-making equation included such highly determinative quality and productivity factors as teaching quality, community and institutional service, and quality of research and scholarship. The district court concluded that variables missing from the model must have had a significant influence on salary and advancement decisions. Without reiterating the details of the district court’s extensive criticisms of the plaintiffs’ statistical evidence, after our own consideration of the evidence, we hold that the district court did not clearly err in concluding that the plaintiffs failed to establish the existence of actionable discrimination by a preponderance of the evidence.1
We note that by finding fault and infirmities in the statistical evidence presented, the district judge was neither evading her responsibility to examine the evidence, nor placing impossible burdens on the plaintiffs.2 The judge was simply unpersuaded. The heavy burden accepted by the plaintiffs in this action was ultimately to prove intentional discrimination across the whole state system of higher education. Although an attempt to prove a narrower case might have prevailed, the plaintiffs failed to prove this case.
III. DENIAL OF A FAIR TRIAL
Plaintiffs assert that they were denied a fair trial by the cumulative effect of numerous pretrial and trial errors. They urge reversal and remand for a new trial before a different judge.
District courts have broad discretion in controlling the conduct of a trial. Railroad Dynamics, Inc. v. A. Stucki Co., 727 F.2d 1506, 1515 (Fed.Cir.), cert. denied, 469 U.S. 871, 105 S.Ct. 220, 83 L.Ed.2d 150 (1984); Kotz v. Bache Halsey Stuart, Inc., 685 F.2d 1204, 1208 (9th Cir.1982). Federal Rule of Evidence 611(a) directs trial judges to exercise “reasonable control” over the presentation of evidence and interrogation of witnesses. “Very few cases outside of the criminal law area support an appellate finding of general judicial misconduct during trial____ The standard for reversal is whether the trial was unfair.” Handgards, Inc. v. Ethicon, Inc., 743 F.2d 1282, *4661289 (9th Cir.1984), cert. denied, 469 U.S. 1190, 105 S.Ct. 963, 83 L.Ed.2d 968 (1985). Some of plaintiffs’ allegations on this issue raise questions, but they do not approach a showing that the trial was unfair.
A. Exclusion of Statistical Experts
Plaintiffs challenge, for example, the court’s order to exclude the testimony of two of their expert witnesses, Doctors Arrow and Olkin. The exclusion was a sanction imposed for violation of court orders. Because scheduling was important in this complex litigation, we defer to the court’s discretion unless it was clearly abused. Obviously, there was a misunderstanding on the part of plaintiffs’ counsel. However, as the court noted, he failed to “memorialize” his understanding. Technically, he violated the court’s scheduling orders. Judge Frye’s finding to that effect is not clearly erroneous. See United States v. National Medical Enterprises, Inc., 792 F.2d 906, 911 (9th Cir.1986) (court’s determination that an order was violated is reviewed under the clearly erroneous standard). No showing of willful disobedience is required for an exclusion order. Societe Internationale Pour Participations Industrielles et Commerciales, S.A. v. Rogers, 357 U.S. 197, 207-08, 78 S.Ct. 1087, 1093, 2 L.Ed.2d 1255 (1958) (party “refuses to obey” simply by failing to comply). When a trial runs on month after month, some tightening of schedules is to be expected.
The challenged preclusion order is justified under Fed.R.Civ.P. 37(b)(2)(B). The order was specifically tailored to the violation and was not an abuse of discretion. See Admiral Theatre Corp. v. Douglas Theatre Co., 585 F.2d 877, 897-98 (8th Cir.1978) (no abuse of discretion in court’s exclusion of exhibits and witnesses not disclosed in compliance with pretrial orders); United States v. Sumitomo Marine & Fire Insurance Co., Ltd.. 617 F.2d 1365, 1369-70 (9th Cir.1980) (order precluding government from introducing evidence as sanction for failing to comply with discovery order was not an abuse of discretion).
B. Rebuttal Time Schedule
Plaintiffs assert that they were unfairly denied an opportunity to rebut defendant’s statistical evidence. There was much disagreement over scheduling of the rebuttal case, complicated by the court’s demand for a termination date in early November 1984 to prepare for another trial.
On September 26, 1984, the court ordered plaintiffs to submit their statistical rebuttal reports on October 3, before the board’s statistical experts testified. She explained that these reports “relate to defendant’s statistical reports, which plaintiffs have had since May.” She noted that plaintiffs already had violated her scheduling order by missing the September 10 deadline.
Later, on October 17, 1984, she admonished plaintiffs for their failure to comply with a scheduling order regarding the submission of new case-in-chief statistical studies. This failure “has resulted in a convolution in the order of proof in this case,” she wrote. Then, she established the rebuttal schedule about which plaintiffs now complain and by written opinion overruled plaintiffs’ objections.
The court has broad discretion in fashioning appropriate scheduling orders. See Rodriguez v. Olin Corp., 780 F.2d 491, 494 (5th Cir.1986) (order of proof committed to district court discretion); Coursen v. A.H. Robins Co., Inc., 764 F.2d 1329, 1340 (9th Cir.1985) (timing of trials and control of docket are left to trial court discretion).
As explained below, the court may have severely restricted plaintiffs in their ability to run new studies on the redefined class. While it is true that the defense was also restricted in its proof, most of the court’s time constraints fell on plaintiffs. Nonetheless, these constraints followed plaintiffs’ repeated failure to meet the court’s deadlines. These facts do not establish reversible error.
C. Late Redefinition of Class
Plaintiffs assign error to the court’s redefinition of the class after six months of *467trial and the refusal to permit additional discovery.
By order filed October 13,1981, the court certified a class “of all those women faculty members who ... were employed as instructor through professor” during the relevant period at the eight Oregon universities. In her December 10, 1982 order denying defendant’s motion to certify an interlocutory appeal of the class certification issue, the judge wrote that she “may alter the class certification decision at any time before a final decision on the merits under Rule 23(c)(1).”
On August 10,1984, she limited the class to “regular teaching faculty,” excluding nonteaching faculty and nonregular instructors. Plaintiffs do not dispute the court’s power to adjust the class definition to conform to the proof. It is the court’s manner of class reduction to which they object.
After the plaintiffs moved for reconsideration and clarification of the class redefinition order, or alternatively for more discovery time to create a new data base for further statistical studies, the court permitted plaintiffs to present new statistical studies by September 26, 1984. That deadline was extended to October 3, but apparently was disregarded by plaintiffs. The court’s attempts to accommodate the late submission of new studies led to the disrupted rebuttal schedule previously discussed.
Again, the question is whether the plaintiffs were deprived of a fair opportunity to prove their case. The court had allowed so much delay and redundant material to enter the case at this point that control was becoming a problem. The plaintiffs do not establish such an abuse of discretion as to warrant another trial with an expanded plaintiff class if one could be established.
D. Daily Transcript
Defendant board arranged for a daily transcript at a cost of $1,000 per day. The court reporter furnished a copy to the court, as is customary in most districts. The reporter did not offer a free copy to plaintiffs. The sale of such copies is a regulated industry. Plaintiffs now assert that the defendant’s action in providing the court with a daily transcript amounts to impermissible ex parte communication. They allege prejudice due to defendant's ability, and their inability, to refer to the transcripts during discussions with the court.
We have found no authority holding that furnishing daily transcripts to the court is an illegal or unethical ex parte communication. There simply is no “communication.” The defense paid the official court reporter to prepare daily transcripts. That reporter was employed by the court. As long as testimony was being transcribed for someone's benefit, the court could gain access to that transcript simply by requesting an extra copy. It was the court’s prerogative to do so. The fact that the board may have initiated the extra copying is irrelevant. Plaintiffs could have purchased any or all transcripts simply by placing orders for them. This court has no basis upon which to find reversible error in the court’s reference from time to time to a transcript of the testimony.
E. Cost of Corrected Data Base
During discovery, plaintiffs acquired copies of the board’s central computer base tapes of faculty information from 1977 through 1982. These tapes had been maintained in the ordinary course of business. These data were the basis for plaintiffs’ statistical studies in the class action phase of the case.
After both sides’ experts had discovered inaccuracies in the original data base, the board allowed plaintiffs access to their files and business records. Plaintiffs moved to compel the board to correct and update the inaccuracies in the tapes. The judge ruled that the board “was not obligated under the Federal Rules to make data compilations more extensive or accurate than those used by defendant in the ordinary course of defendant’s business.” She ordered that the board make available to plaintiffs any revisions to the data base made in prepara*468tion for trial with costs to “be borne equally by plaintiffs and defendant.”
The board chose to correct the inaccurate data base for use at trial, at a cost of approximately $100,000. Plaintiffs declined to accept and pay for half the cost of these corrections. They now assert that the court erred in requiring them to share the costs of correcting the data base. We do not think, however, that the district court had any obligation to require the defendant to subsidize the plaintiffs’ costs of litigation.
Under Fed.R.Civ.P. 26(c)(2), the court may order that “discovery may be had only on specified terms and conditions.” This rule empowers district courts to order that a party seeking discovery pay a portion of the expenses incurred in obtaining discoverable materials. Oppenheimer Fund, Inc. v. Sanders, 437 U.S. 340, 358, 98 S.Ct. 2380, 2393, 57 L.Ed.2d 253 (1978); In re Puerto Rico Electric Power Authority, 687 F.2d 501, 507 (1st Cir,1982); American Standard, Inc. v. Bendix Corp., 71 F.R.D. 443, 448 (W.D.Mo.1976) (defendant ordered to pay one-half of the reasonable costs of preparing for interviews and obtaining transcripts thereof). Such orders are reviewed for an abuse of discretion. National Independent Theatre Exhibitors, Inc. v. Buena Vista Distribution Co., 748 F.2d 602, 609 (11th Cir.1984), cert. denied, 471 U.S. 1056, 105 S.Ct. 2120, 85 L.Ed.2d 484 and _ U.S. _, 106 S.Ct. 544, 88 L.Ed.2d 473 (1985). There was no abuse of discretion here.
CONCLUSION
While the plaintiffs find fault with the trial court’s use of the statistical evidence, and with 20-20 hindsight, this court itself might view some of the statistical evidence differently, we have found nothing in the record to convince us that another trial would produce a different result. The three faculty members who did prove some elements of discrimination in their own cases recovered appropriate judgments. The class as a whole failed to establish by the preponderance of the evidence that the Oregon State Board of Higher Education had any systematic intent to discriminate or that discrimination was a “standard operating procedure” of the board. The statistics may have proved that there were disparities in salaries and other terms of employment, but not that they were the product of disparate treatment. The trial judges’ findings of fact withstand Rule 52 review.
AFFIRMED.
. We note in passing that it is often acceptable in the social sciences to use a statistical model for proof of behavior even where, in an absolute sense, that model does not describe the data well. However, courts are not free to decide legal propositions on hypothetical evidence. This is especially true where courts are asked to draw inferences as to the existence of hidden discriminatory motives from statistical evidence. See Spaulding v. University of Washington, 740 F.2d 686, 704 (9th Cir.), cert. denied, 469 U.S. 1036, 105 S.Ct. 511, 83 L.Ed.2d 401 (1984).
. Plaintiffs seize upon a possibly erroneous belief held by the district judge that the board was correct in its use of faculty rank as an "objective piece of data.” This intermediate shifting of the burden of persuasion, however, even if error, does not appear to have affected the outcome of the trial.