concurring in part and dissenting in part.
For the reasons that follow, I would affirm the district court’s finding of discriminatory impact, remand for further factual findings regarding discriminatory treatment, and reverse the district court’s finding of liability against LIUNA. I concur in the majority’s affirmance of the district court’s imposition of discovery sanctions against the individual Plaintiffs and Plaintiffs’ counsel.1
This case involves an employment discrimination class action suit filed on December 20, 1984, brought under 42 U.S.C.A. § 2000e, et seq. (West 1994 & Supp.1998) (“Title VII”) and 42 U.S.C.A. § 1981 (West 1994 & Supp.1998). Originally named as defendants were Local 496 of the Laborer’s International Union of North America (“Local 496”) and a number of electric power companies that were later dismissed from the suit. By orders dated October 28,1985, and March 6,1986, individual discrimination suits filed by Plaintiffs Colvin; Lilly; Turner, II; Poue-wells; Coffee, Sr.; Johnson, Jr.; and Rice against Local 496 and Local 496’s Business Manager, Floyd Conrad,2 were consolidated with the class suit.
On January 26, 1988, the court certified a class of
All Black persons who, on or before [January 26, 1988], have sought membership in Local 496 and/or employment either by application or by referral under the policy described in the collective bargaining agreement of March 9, 1973, by which the defendant union agreed to make referrals to employers at the Perry Nuclear Power Plant site.
Cheryl Journigan was the last class member to file an EEOC charge against Local 496; she filed on January 10, 1985. On September 7, 1989, Plaintiffs Colvin *415and Tomblin filed additional EEOC charges alleging that they had recently discovered that Laborer’s International Union of North America (“LIUNA”) was aware of their EEOC charges against Local 496 but had intentionally refused to investigate the matters; these were the first allegations in this case levied against LIUNA.
On January 12, 1990, the class moved to amend their complaint to add LIUNA as a party defendant. The suit originally was set for trial on January 17, 1990, but the trial court found that LIUNA was an indispensable party and on January 19,1990, granted the Class leave to amend their complaint. On January 22, 1990, Plaintiffs Tomblin and Colvin filed “amended” charges of discrimination, again alleging that they had recently discovered that LI-UNA had intentionally failed to investigate the discrimination charges. Colvin received his right to sue letter on March 26, 1990, and on March 30, 1990, the Class filed an amended complaint, adding LIU-NA and Business Manager Conrad as party defendants.
The court bifurcated the liability and damages proceedings. After a 1991 trial on liability issues, on December 10, 1991, the district court found that Local 496 and Conrad had engaged in a pattern or practice of discriminatory treatment and were liable under a disparate impact theory, both in violation of Title VII and § 1981. The court also ruled that LIUNA was similarly hable because LIUNA had an agency relationship with Local 496 and because it had an affirmative duty to oppose its local affiliate’s discrimination.
During the discovery period of the damages phase, the court imposed sanctions on five individual Plaintiffs and on Plaintiffs’ counsel for failure to comply with LIU-NA’s discovery demands. During the damages portion of the- trial, the parties reached a partial settlement wherein Local 496 and LIUNA agreed to make an initial payment to the Class; additional damages payments depend on the outcome of this appeal, and all parties reserved the right to appeal any order other than the judgment journalizing the settlement. The district court approved the parties’ settlement agreement, and on July 19, 1996, the court formally dismissed the class action suit. All parties timely noticed their appeals.
Appellants Local 496 and LIUNA now claim that the trial court committed clear error in finding that Plaintiffs proved race discrimination under either a disparate impact or a disparate treatment theory. Appellant LIUNA also asserts that the trial court erred in finding that LIUNA was hable pursuant to either an “agency theory” or an “affirmative duty” theory. Plaintiffs cross-appeal, claiming that the district court abused its discretion in imposing sanctions against Plaintiffs and their counsel for discovery violations.
I. Factual Background.
A.
In 1973, Local 496 signed a project labor agreement (“PLA” or “1973 agreement”) governing the construction of the Perry Nuclear Power Plant (“Perry”), which is located in Lake County, Ohio. LIUNA was not a signatory to this agreement. In the absence of the PLA, LIUNA’s Uniform Local Union Constitution, which governs Local 496, would require: “In order to be eligible for membership a person must be working at the calling within the territory of the Local Union in which the individual applies for membership.” LUINA Uniform Local Constitution, Art. Ill, § 1(a). In other words, according to the Local Constitution, a nonmember must secure a union job before he or she is eligible to become a union member. If an individual obtained such a job, he or she was required to join Local 496 within eight days of beginning such employment.
The PLA, however, required in part that applicants for work on the Perry project be referred without regard to union mem*416bership.3 Thus, the Perry contract essentially included a contractual waiver of the working-at-the-calling rule. Defendant Conrad admits that pursuant to this agreement, Local 496 operated as an “exclusive hiring hall” for laborers at Perry, and had an obligation to refer both members and non-members for the laborer jobs.
Perry’s construction phase lasted from 1973 to 1985. Thereafter, work at Perry was considered “maintenance” and was performed under the National Maintenance Agreement, effective March 4, 1985. LIUNA signed the maintenance agreement; Local 496 did not. Local 496, however, continued as the referral agent for laborers at Perry.
Local 496’s union hall is located in Madison, Ohio, and its geographical jurisdiction is limited to Lake County. Lake County is bordered to the east by Ashtabula County, to the south by Geauga County, to the west by Cuyahoga County, and to the north by Lake Erie. Local 496’s craft jurisdiction is limited to commercial and industrial building.construction. When Perry’s owners first began to build Perry, a jurisdictional dispute arose between Local 496 and LIUNA Local 860, whose craft jurisdiction is “heavy construction,” which includes roads., sewers, site preparations, bridges, dams, and airport runways, and whose geographic jurisdiction includes Lake, Cuyahoga, and Geauga counties. LIUNA resolved this dispute by splitting the work between the two Locals. Local 496, however, remained the exclusive referral source for laborers at Perry. As the Business Manager, Defendant Conrad personally had exclusive control over union referrals to all Perry contractors.
B.
Local 496’s membership is overwhelmingly white. Between 1980 and 1985, the total number of active members ranged from 509 to 545, while the number of black members ranged from 16 to 20. During this same period, black members comprised between 2.88% and 3.59% of total membership. Approximately 30% (145 of 507) of Local 496 members were related to one another. Between April 1982 and February 1984, 42 new members, none of whom was black, were initiated into the Local 496. From 1980 to 1985, the union accepted a total of 54 new members, one of whom was black.
Over the years, the referral practices of Local 496 have varied. Prior to October 1987, the union did not have in place a written referral policy. Local 496 admits that in response to contractor requests during the pre-1987 period, the union referred only members, in contravention of the PLA. Unemployed members could sign a spiral notebook located in the union hall, and from that notebook, Conrad would compile a master list of unemployed members from which he made referrals. At most times, non-members could also sign the notebook, but they were not included on the master referral list. During this period, Local 496 accepted as new members those who obtained contractor-letters or were hired directly by contractors.
In 1987, however, as part of a settlement of an NLRB charge filed by a white nonmember, Local 496 instituted a new, written, referral policy. Under this policy, *417both members and nonmembers could sign the unemployed list, and referrals were to be made in the order that the names appeared on the list. Each month, the secretary was to prepare a new master list, deleting the names of individuals from the prior month’s list who either (1) had worked more than 40 hours the prior month, or (2) had not contacted the union during the prior month and informed the union that he or she was still unemployed and was still interested in staying on the list. Plaintiffs’ counsel, Edward Kramer, received a copy of these rules on August 30,1989.
Even with the changed referral policy, the union admitted that it never actually referred a non-member to a job at Perry.4 With one exception, however, after May 1985, no class member communicated with Local 496 for the purpose of becoming a union member and/or seeking a referral. Sometime between January 19 and 22, 1990, Plaintiffs Colvin and Tomblin went to Local 496’s hall and signed the unemployed list. At the time of signing, their names were 92nd and 93rd on the list. While they were at the hall, no one informed Colvin or Tomblin about the monthly re-notification requirement.5 Col-vin and Tomblin testified that Mr. Harrington, a union representative, did tell them that to join the union, one must be working at the calling. On February 1, 1990, Local 496’s secretary prepared a new unemployed list on which Colvin and Tom-blin occupied the 65th and 67th positions. Neither Colvin nor Tomblin wrote or otherwise informed Local 496 during February 1990 or thereafter that he remained unemployed and continued to be interested in referrals. When the secretary prepared the March 1990 list, she removed Colvin and Thomas’s names.
C.
LIUNA and Local 496 are separate entities; the locals negotiate their own contracts, spend their own funds, own their own property, etc. At all relevant times, the Local 496 referral system was administered solely by Local 496. The constitutions of LIUNA and Local 496, however, provide LIUNA with supervisory power over Local 496. LIUNA retained the power to suspend or dissolve the charter of Local 496, and testimony established that LIUNA had the authority to correct “blatantly” discriminatory policies of local affiliates.
The district court found that LIUNA “has continually refused to investigate charges of discrimination or take any action to correct [Local 496’s] illegal conduct.”- Local 496'frequently notified LIU-NA of the race discrimination complaints that had been lodged against it, and LIU-NA kept a file of documents pertaining to the discrimination claims. No class member, however, ever contacted LIUNA regarding a discrimination complaint. In an interrogatory answer, LIUNA stated that it never investigated the specific discrimination complaints lodged -with the EEOC by Colvin and Tomblin in 1984 against Local 496, but that it instead mailed copies of the complaints to Thomas Arconti, LIU-NA’s Regional Manager in charge of Local 496. It also stated that this was consistent with LIUNA’s established practice upon receipt of similar complaints, and cross-references its answers to earlier interrogatories.6
*418The evidence of record shows that several times in 1984 and 1985, Thomas Ar-conti met with Local 496 members, a number of whom were black, to discuss problems they were having getting referrals from Local 496. Also at that time, they discussed the discrimination complaints of a non-class member, Donald Robinson. Thomas Arconti died on May 1, 1989, prior to the filing of discrimination charges against LIUNA either with the EEOC or in the Amended Complaint; it is not clear from the record what his investigation disclosed.
II. Racial Discrimination.
Appellants Local 496 and LIUNA assert that the trial court committed clear error in finding that the Plaintiffs proved race discrimination under either a disparate impact or a disparate treatment theory. While I concur in the majority’s conclusion that we must uphold the district court’s finding of disparate impact, I believe that we cannot affirm the findings of disparate treatment because the dearth of relevant factual findings leave us unable to evaluate properly the court’s finding of discriminatory treatment.7
We review a district court’s factual finding of discrimination only for clear error. Jackson v. RKO Bottlers of Toledo, Inc., 743 F.2d 370, 374 (6th Cir.1984). Thus, findings with regard to the evidence necessary to establish a prima facie case of employment discrimination are reviewed under the clearly erroneous standard. See id. at 374-77.
Title VII prohibits discrimination by a union against its members on the basis of race, color, religion, sex, or national origin. 42 U.S.C.A. § 2000e-2(e). Racial discrimination is also prohibited by 42 U.S.C.A. § 1981; claims brought under § 1981 are governed by the same evidentiary framework applied to Title VII claims. Patterson v. McLean Credit Union, 491 U.S. 164, 186, 109 S.Ct. 2363, 105 L.Ed.2d 132 (1989).
In cases where plaintiffs allege that because of their race they were treated differently from Caucasian individuals, plaintiffs bear the initial burden of establishing by a preponderance of the evidence a pri-ma facie case of discrimination. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973). In these “disparate treatment” cases, plaintiffs must prove that the defendant had a discriminatory intent or motive, which in some situations may be inferred from the mere fact of differences in treatment. Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 986, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988); International Bhd. of Teamsters v. United States, 431 U.S. 324, 335 n. 15, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977). Plaintiffs may establish such intent either directly, by producing direct evidence of discrimination, or inferentially, by a showing of the following four elements: (1) plaintiffs belong to a protected class; (2) plaintiffs applied for and were qualified for employment; (3) despite their qualifications, plaintiffs were denied a favorable employment decision; and (4) persons outside of the protected *419class with substantially similar or lesser qualifications received the jobs. McDonnell Douglas, 411 U.S. at 802. Where plaintiffs allege a system-wide “pattern or practice” of discrimination, they must ultimately prove more than the mere occurrence of isolated or sporadic discriminatory acts; they must establish that racial discrimination was the defendant’s “standard operating procedure.” Teamsters, 431 U.S. at 336 & n. 16. Upon the plaintiffs’ satisfaction of their prima facie case, the burden then shifts to the defendant to show a legitimate, nondiscriminatory business reason for acting as it did. If the defendant makes such a showing, the burden shifts back to the plaintiffs to show that the employer’s stated reason is really a pretext for unlawful discrimination. Watson, 487 U.S. at 985-86.
Title VII proscribes “not only overt discrimination but also practices that are fair in form, but discriminatory in operation.” Griggs v. Duke Power Co., 401 U.S. 424, 431, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971). In a so-called “disparate impact” case, the plaintiffs need not prove that the defendant intended to discriminate; instead, plaintiffs must prove that a particular employment practice, although neutral on its face, has caused a disproportionate adverse effect on a protected group. See Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 657, 109 S.Ct. 2115, 2125, 104 L.Ed.2d 733 (1989); United States v. City of Warren, 138 F.3d 1083, 1091 (6th Cir.1998); Scales v. J.C. Bradford & Co., 925 F.2d 901, 907 (6th Cir.1991). Once the plaintiffs have established the adverse effect, the burden shifts to the employer to produce evidence that the challenged practice is a business necessity. Wards Cove, 490 U.S. at 658-59.- The plaintiffs can defeat a defendant’s asserted business justification by showing either that the justification is a pretext, or that there exists an alternative practice with less racial impact that will achieve the same business end. Id. at 658.
Evidence of disparate impact usually focuses on statistical disparities rather than on specific incidents. Watson, 487 U.S. at 987. We have frequently stated that to prove a prima facie case of disparate impact, plaintiffs must (1) identify a specific employment practice; and (2) “show an adverse effect caused by the employment practice by offering ‘statistical evidence of a kind or degree sufficient to show that the practice in question has caused the exclusion of applicants for jobs ... because of their membership in a protected group.’ ” Scales, 925 F.2d at 908 (quoting Watson, 487 U.S. at 994); see also City of Warren, 138 F.3d at 1093.
We have also held, however, that “statistical evidence is not absolutely essential in proving a disparate impact case[; nonetheless,] there must be proof of disparity using the proper standards for comparison.” Gibson v. Frank, 946 F.2d 1229, 1233 (6th Cir.1991); see also Thomas v. Washington County Sch. Bd., 915 F.2d 922, 926 (4th Cir.1990) (statistics are “neither the exclusive nor a necessary means of proof’ in disparate impact cases). The Supreme Court has made clear that the proper comparison in a disparate-impact case is between “ ‘the racial composition of [the at-issue jobs] and the racial composition of the qualified ... population in the relevant labor market.’ ” Wards Cove, 490 U.S. at 650 (alterations in original) (quoting Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 308, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977)).
A. Disparate Impact.
The district court found that Plaintiffs established a prima facie case of discrimination under a disparate impact theory. Plaintiffs identified the members-only referral policy and the imposition of the working-at-the-ealhng requirement, in spite of its waiver in the PLA, as the employment practices causing the adverse effect. Plaintiffs asserted that in conjunction with these practices, two different types of “nepotism” operated to exclude *420blacks from jobs at Perry: (1) members who were already working at Perry could approach a contractor to get a relative or friend a union job; (2) members would approach Conrad to get relatives or friends employment.
Both sides submitted statistical analyses and expert testimony purporting to show that black membership in Local 496 was or was not proportionally low in a statistically significant way. While the experts agreed on the appropriate technique, ie., standard deviation analysis, they disagreed upon how to define the relevant qualified labor force to which Local 496’s membership must be compared.
The disagreement was primarily geographical. The experts agreed that the court should look at the numbers associated with both the General Occupational Category (“GOC”) of “operators, fabricators and laborers,” which included machine operators and tenders (except precision), fabricators, assemblers, inspectors and samplers, transportation occupations, material moving equipment operators, handlers, equipment cleaners, helpers, and laborers, and the Specific Occupational Category (“SOC”) of “handlers, equipment cleaners, helpers, and laborers.” They disagreed, however, on which persons falling under these categories should be included in the relevant labor force pool.
According to the 1980 Census, the residency figures in the four counties from which Local 496 drew 93.5%8 of its members broke down as follows:
[[Image here]]
As an initial matter, I note that the parties stipulated that “1762 persons employed in [the SOC] Category in Cuya-hoga, Lake, Ashtabula and Geauga counties were black.” Despite the stipulation, I think that this number is obviously incorrect. Both experts purported to rely on statistics gathered from the 1980 Census. For the stipulated number to be accurate, the 1980 Census numbers for Cuyahoga County alone would be wrong, as would the resulting percentages. There is no support whatever in the record for this number, and the parties have not used it in their calculations. Thus, I too have ignored it. In addition, I note that the district court found that 23.4% of the SOC workforce in Cuyahoga, Lake, Ashtabula, and Geauga counties was black. That percentage figure does not appear in either party’s calculations and does not find any support in the record. The court did, however, expressly claim to rely on the 1980 Census figures for this information. Therefore, I have used the number actually reported by the 1980 Census, ie., 21.54%, as the comparison figure adopted by the court.
Plaintiffs’ expert, Dr. Pendleton, opined that the relevant labor force numbers should be drawn from the entire four-county area. His opinion was based on several factors: (1) 93.5% of Local 496’s membership lived in these four counties; (2) the pay scale (approximately $14.00/hr for Perry laborers vs. an average of $6.66/hr for all jobs, including professional occupations) and the relatively low skill-level required to perform the job necessitated consideration of an expanded geographic area because people would be willing to travel farther; and (3) the extensive highway system between Cuyahoga and Lake counties increased the distance which people would be willing to travel for work.9 *421Dr. Pendleton testified that he did not “weight” his figures according to commuter patterns/population proportionality because there was evidence of a “chilling effect,” i.e., when there has been either word-of-mouth recruiting or media coverage of past discrimination suits, both of which were present in this case, potential applicants remove themselves from the applicant pool and therefore skew the weights. Applying standard deviation analysis, Dr. Pendleton found that there was a statistically significant underrepre-sentation of blacks in Local 496’s union membership.10
Defendant’s expert, Dr. Martin, opined that the relevant labor force numbers should instead include only those individuals within the GOC or SOC who presently work in Lake County, regardless of residency. She based this opinion on the fact that Local 496’s membership policy states that one must be working in Lake County at the time one applies for membership.11 Using the applicant pool so-defined, Dr. Martin concluded that there was not a statistical underrepresentation of blacks in Local 496’s membership.
Local 496 contends on appeal that the district court’s finding of disparate impact cannot stand because the court failed to identify properly the relevant labor market. Wards Cove makes clear that defining the relevant labor force is of paramount importance. 490 U.S. at 650-51. This determination, however, is a factual question, which we review only for clear error. See EEOC v. O & G Spring & Wire Forms Specialty Co., 38 F.3d 872, 876-78 (7th Cir.1994). We have previously described “clear error” as
when the reviewing court on the entire evidence is left with the definite and firm conviction that a mistake has been committed. The question is not whether the finding is the best or only conclusion that can be drawn from the evidence, or whether it is the one which the reviewing court would draw. Rather, the test is whether there is evidence in the record to support the lower court’s finding, and whether its construction of that evidence is a reasonable one.
Heights Community Congress v. Hilltop Realty, Inc., 774 F.2d 135, 140 (6th Cir.1985).
The district court defined the relevant labor pool as follows:
The relevant geographic area for purposes of comparison of the labor force with the membership of Local 496 is the four-county area from which the union draws the majority of its members, not only the county over which the local has jurisdiction.... The four-county area is comprised of Lake, Ashtabula, Cuya-hoga and Geauga counties.
The court then concluded:
Local 496’s membership and referral policies have resulted in a disproportionately small percentage of black union members as compared to the relevant labor force. Plaintiffs’ statistical evidence shows a gross disparity between the percentage of blacks in the membership of Local 496 and the labor force of the four-county area, both in terms of total number of members during the *422relevant period and in terms of the new members taken in during this period.[12]
The statistical disparities are sufficient to establish a prima facie case of discrimination under the disparate impact theory. In addition, other evidence of discriminatory treatment of class members supports the statistical evidence and together the evidence raises an inference of intentional discrimination.
After reviewing the evidence, I cannot say that the four-county area clearly is too broad a measurement. It is true that earlier in its factual findings, the court described Dr. Pendleton’s definition of the relevant labor pool as “either the GOC or the SOC for the four-county area, rather than limited to the GOC or the SOC working in Lake County,” and stated that “the basis of Dr. Pendleton’s report is too broad.” It thus appears that the court’s statements concerning Dr. Pendleton’s defi-inition of the relevant market and its own definition of the relevant market are, on their faces, internally inconsistent. But looking at the opinion and the record as a whole, I conclude that they are not inconsistent in substance.13
The court explicitly rejected Dr. Martin’s opinion that the relevant market should include only those who presently work in Lake County. I would agree that Dr. Martin’s definition is entirely too narrow. In fact, it begs the question. As the district court observed:
The weakness of Dr. Martin’s argument or conclusion is that it does not take into consideration the fact that there may be discrimination in the hiring of the 6,488 employees in the GOC in Lake County. In other words, if in the hiring of all of the 6,488 employees in the GOC in Lake County there was in fact discrimination, then 5% blacks would not be a proper figure. Dr. Martin’s conclusion assumes that there was no discrimination in the hiring of blacks in the GOC and, thus, her conclusion that the applicant pool is 5% black may or may not be valid.
See Clark v. Chrysler, Corp., 673 F.2d 921, 928 (7th Cir.1982) (recognizing that the danger of “weighting” a relevant labor market calculation to reflect commuter patterns is that the geographic recruiting and employment practices may themselves be tainted by racial discrimination). In addition, I see no reason to assume that persons currently driving to work in Cuya-hoga, Ashtabula, or Geauga Counties would not be willing to drive an equal time or distance to work in Lake County if desired jobs were available. Moreover, Dr. Martin’s definition relies on the existence of the working-at-the-calling requirement. Under the PLA between Perry and Local 496, however, persons not yet working in Lake County and persons not yet members of Local 496 can be referred to Perry and then admitted into membership. Thus,' there is no factual basis for Dr. Martin’s arbitrary limitation.
As listed above, Dr. Pendleton’s testimony provided several reasons why the court should consider the population figures for all four counties, not the least of which was the relatively high wages that laborers at Perry earned. Compare EEOC v. Chicago Miniature Lamp Works, 947 F.2d 292, 302 (7th Cir.1991) (finding it necessary to con*423sider commuting times because the jobs at issue paid low wages and provided little opportunity for advancement, and therefore were more likely to be filled by those living close by). I find these reasons persuasive and sufficient grounds for the court to include the entire four-county area in the relevant labor market.
Moreover, if it is appropriate to include in the relevant labor market the bordering Ashtabula and Geauga Counties, which Defendants do not contest, obviously at least some portion of the bordering Cuyahoga County must also be included. If, for proximity reasons, the court were to limit the portion of Cuyahoga County includable in the relevant labor market to something less than the entirety of Cuyahoga County, in all likelihood, black percentages would increase: the majority of Cuyahoga County’s black population resides on the county’s east side, ie., the side located closer to Lake County and the Perry plant. See, e.g., United States v. City of Parma, 661 F.2d 562, 565-66 (6th Cir.1981); Banks v. Perk, 341 F.Supp. 1175, 1178 (N.D.Ohio 1972), aff'd in part, rev’d in part, 473 F.2d 910 (6th Cir.1973). Thus the court’s definition of the relevant labor market was not clearly erroneous.14
As the court found, whether one compares the relevant labor market population *424percentage (SOC — 21.54%; GOC — 21.79%) to the Local 496’s total black membership numbers (2.88% — 3.59%) or to the number of new black members admitted during 1982-84(0%) or 1980-85 (approximately 2%), the disparity is statistically significant. Thus, despite the confusion as to numbers in the district court’s opinion, I conclude that the court’s finding of a “gross disparity between the percentage of blacks in the membership of Local 496 and the [relevant] labor force” was not clearly erroneous.
Local 496 further argues that the court failed properly to link the statistical disparity shown to the employment practices at issue. In addition to its statistical findings, however, the court cited considerable anecdotal evidence showing that these disparities were in fact caused by the members-only referral policy and the imposition of the working-at-the-calling rule. See Gibson, 946 F.2d at 1233 (a plaintiff may establish disparate impact on a racial minority without statistical evidence).
The court’s list of evidence included: the bargaining agreements; that employers, by custom, practice and agreement, could hire employees without any union oversight regarding discrimination; that the union admitted into membership anyone whom an employer hired; that in 1975, union membership was 100 with 10 black members, while in 1985, membership was 500 with 20 black members; that Perry was a “closed shop” arrangement; that in response to non-specified employer requests for workers, the union sent only members and never sent a non-member; that because of security, the average applicant could not gain access to Perry employers to be hired, while present union members, stewards, and foremen had direct access to such employers and could recommend their relatives and friends to employers for jobs, thereby rendering these relatives and friends “working in the calling”; and that approximately 30% of all union members, and 30% of those hired at Perry, were relatives of existing union members. The court also found that when a contractor specifically asked for a minority worker, instead of sending one of the black non-member plaintiffs, Conrad would instead contact Local 860 to obtain a referral. In addition, as the court noted, Conrad essentially admitted that he knew that employers relied on their white, but not black, superintendents and foremen to acquire additional workers.
Given the extreme statistical disparity between the number of blacks in the relevant labor pool and the number of new black members added to Local 496 between 1980 and 1985, the district court did not clearly err in finding that this disparity was caused by Local 496’s members-only referral policy and the working-at-the-calling rule. Perry is a closed shop. Local 496 admittedly referred only members to Perry during the time period in question. To become a member of Local 496, one had to be working-at-the-calling in Lake County, but during the time in question, Perry was the primary employer of laborers in all of Lake County. Also during the relevant period, all of the class members approached the union, requested membership and/or referrals to Perry, and were denied the same. It is therefore reasonable to conclude that the challenged practices caused the statistical disparity. Consequently, I agree that this court must affirm the district court’s finding of disparate impact.15
B. Disparate Treatment.
The district court held that Plaintiffs proved a prima facie case of disparate *425treatment under the McDonnell Douglas test because: (1) plaintiffs are black; (2) they were available for referral by the union for job opportunities at Perry; (3) they did not receive referrals despite the fact that as nonmembers, they should have been referred according to the PLA; and (4) white non-members were referred to jobs to which plaintiffs had applied and were made members of Local 496. The court also summarily stated that the evidence demonstrated that Local 496 failed to inform minorities of its procedures for membership and job referrals.
On appeal, Local 496 devotes considerable energy to attempting to discredit the court’s conclusions that Local 496 referred white, but not black, non-members to jobs at Perry and that Local 496 faded to inform minorities, but informed white persons, about membership and referral procedures. Plaintiffs respond by pointing to evidence in the record that could support the court’s conclusions. While there is evidence in the record that, if credited, might support the district court’s findings, “[i]t has long been clear that when the court does not make findings which are sufficient to indicate the factual basis for its ultimate conclusion, the appropriate procedure is to vacate the judgment and remand for such findings.” Gonzales v. Galvin, 151 F.3d 526, 532 (6th Cir.1998) (emphasis omitted); accord Deal v. Cincinnati Bd. of Educ., 11 Ohio Misc. 184, 369 F.2d 55, 63-64 (6th Cir.1966) (there must be subsidiary findings to support the ultimate conclusions of the court).
In its findings of fact, without further explanation, the district court summarily stated that the working-at-the-calling requirement was waived “primarily” for white applicants. The only expansion on this conclusory finding is the court’s statement in its conclusions of law that “the evidence in the record shows a pattern and policy of favoritism to friends and relatives.” The court fails to provide us with a single clue as to who these “friends and relatives” allegedly receiving favorable treatment were, or how, when, or how often the working-at-the-calling rule was waived in this manner, or for that matter, whether such waivers were racially disproportionate. Similarly, while the court did find that in 1990, black class members were not informed of the monthly renotifi-cation requirement, the evidence of record reflects only one instance where the union failed to provide a copy of its referral policy. The court failed to make any other findings with respect to the union’s failure to provide membership/referral policy information. Thus, I do not believe that we can assess whether Local 496 maintained a “pattern and practice” of refusing to inform minorities of membership and referral procedures. Moreover, the court has not provided us with any reasons to support its implicit conclusion that white persons were given such information. We simply do not know what evidence the court credited in reaching its conclusion that Plaintiffs had been treated in a discriminatory manner.
[A] district court’s findings “should be comprehensive and relevant to the issues so as to provide a rational basis for the trial court’s decision.” In addition, the “findings should be explicit so as to give the appellate court a clear understanding of the basis of the trial court’s decision, and to enable it to determine the grounds on which the trial court reached its decision.”
Sanders v. Dorris, 873 F.2d 938, 942-43 (6th Cir.1989) (district court’s failure to discuss evidence supporting pattern and practice discrimination) (quoting Grover Hill Grain Co. v. Baughman-Oster, Inc., 728 F.2d 784, 792 (6th Cir.1984)) (internal citation omitted); see also Gonzales, 151 F.3d at 532. In short, due to the court’s complete failure to make relevant factual findings, we cannot discern the grounds for its decision and we therefore have no basis upon which to determine whether its conclusions were clearly erroneous. Therefore, I would remand this issue to the district court to provide explicit factual *426findings explaining the basis for its conclusions.
III. Liability of LIUNA.
We review de novo a district court’s conclusions of law, Waxman v. Luna, 881 F.2d 237, 240 (6th Cir.1989) (per curiam), but review its findings of fact only for clear error, Jackson v. RKO Bottlers of Toledo, Inc., 743 F.2d 370, 374 (6th Cir.1984).
The National Maintenance Agreement, signed by LIUNA and effective at Perry in March 1985, provided in part:
The employer agrees to hire men in any territory where work is being performed or is to be performed in accordance with the hiring procedure existing in the territory where the work is being performed or is performed; however, in the event the Local Union is unable to fill the request of the Employer for Employees within a forty-eight (48) hour period after such request for Employees ..., the Employer may employ workmen from any source.
The Agreement also contained a non-discrimination clause, which stated: “The Union and the Employer agree to abide by all Executive Orders and subsequent amendments thereto, regarding the Civil Rights Act of 1964, pertaining to non-discrimination in employment, in every respect.”
The district court found that “[a]t all relevant times, the referral system was administered solely by Local 496; no International Union representative or agent has participated in its operation.” The court also found that LIUNA “was aware of the discriminatory actions being take by the defendants] Local 496 and Floyd Conrad since May of 1983, but has continually refused to investigate charges of discrimination or take any action to correct this illegal conduct.” It is undisputed, however, that no class member ever contacted LIUNA to register a discrimination complaint or to ask LIUNA to investigate his or her situation.
The district court concluded that LIU-NA was liable under a principal-agent theory for the discriminatory actions of Local 496 because both before and after March 1985, LIUNA was “aware” of Local 496’s actions regarding its refusal to make referrals to class members or to permit them to become union members, and because LIU-NA “acquiesced” in those actions. The court also held that LIUNA had an affirmative duty under Title VII and § 1981 to oppose Local 496’s discriminatory practices because for several years LIUNA had knowledge of the class members’ charges of discrimination against Local 496 and of some resulting EEOC reasonable cause findings, and because LIUNA had signed the National Maintenance Agreement, which regulated labor relations at Perry.
A. Agency Theory.
An international union may be liable for the discriminatory practices of an affiliated local if, in carrying out such practices, the local is acting as the international’s agent. See Berger v. Iron Workers Reinforced Rodmen Local 201, 843 F.2d 1395, 1426-33 (D.C.Cir.1988).16' Thus, to be liable under *427Title VII, an international must “participate in” or “authorize[ ], ratify! ], or approve! ]” of the particular condition about which the plaintiff complains. See id. at 1428-32.17 Contrary to the statements of the district court, mere “acquiescence” is not enough. To be similarly liable under § 1981, the plaintiff must also prove that the relationship between the international and the local is sufficient to impute discriminatory intent. Id. at 1430.
In the present case, the district court’s findings and the parties’ stipulations foreclose a finding that LIUNA “participated in” the discriminatory referral process. The closer question, of course, is whether, by signing the National Maintenance Agreement, and therein obligating contractors to “hire men in any territory where work is being performed or is to be performed in accordance with the hiring procedure existing in the territory where the work is being performed or is to be performed,” when LIUNA knew or should have known of Local 496’s discriminatory practices, LIUNA “authorized, ratified, or approved” Local 496’s referral practices.
In General Building Contractors, like in the present case, the plaintiffs charged that the union had engaged in a pattern and practice of racial discrimination by systematically denying access to union referral lists and by arbitrarily skewing referrals in favor of white workers. The Court held that the mere fact that trade associations and employers had delegated to the union the authority to select workers, and the union, in effectuating that delegation, intentionally discriminated or produced a discriminatory impact, alone was not enough to support an agency relationship. 458 U.S. at 391-94. Likewise, this court cannot find that merely securing in the National Maintenance Agreement a promise that the contractors will follow Local 496’s hiring procedures, whatever they may be, without any concurrent control over those procedures, is sufficient to render LIUNA a principal responsible for Local 496’s discriminatory actions. This is particularly so because the same agreement that Plaintiffs and the district court would read as reflecting LIUNA’s acquiescence in the discriminatory referral procedures also contains an explicit requirement that the union not discriminate.
In Berger, the court found an agency relationship because, in addition to the international union’s constitutional provisions providing the international with *428oversight authority over the local’s membership practices (which alone would not be enough to find the agency relationship, 843 F.2d at 1431), the international had formally endorsed the establishment of the discriminatory practice in question and was actually involved in its implementation. Id. at 1430-32. In the cases cited by Plaintiffs, the provisions in question were themselves discriminatory and were explicitly negotiated by the international. See, e.g., Myers v. Gilman Paper Corp., 544 F.2d 837, 844, 847-48, 850-51 (5th Cir.1977) (agency liability for “line seniority” provision in labor agreement either negotiated by the international or “advised” by international and requiring international’s approval). In the present case, LIUNA was not involved in Local 496’s 1973 agreement to refer both members and nonmembers to laborer positions at Perry. Plaintiffs have failed to present any evidence indicating that LIUNA supported or approved of Local 496’s decision in fact not to comply with the terms of the PLA, or allegedly to make exceptions in their noncompliance for a number of white relatives/friends. LIUNA knew that charges of discrimination had been levied, but not that discrimination actually had occurred, and knew that Plaintiffs were seeking court intervention to resolve the matter. Therefore, on the record before us, where the district court found and the record reflects nothing more than mere “acquiescence,” in my view we cannot find that Local 496 was acting as LIUNA’s agent when it discriminated against Plaintiffs.
B. Affirmative Duty Theory.
The district court found that because LIUNA had notice of discrimination complaints against Local 496 and because it was a party to the National Maintenance Agreement, LIUNA had an affirmative duty to oppose Local 496’s discriminatory conduct. In so holding, the district court relied upon Kaplan v. International Alliance of Theatrical & Stage Employees, 525 F.2d 1354 (9th Cir.1975), wherein the Ninth Circuit held that under Title VII, “[b]y making and [tacitly] enforcing” a collective bargaining agreement that perpetuates past discriminatory effects, an international labor organizations has an affirmative duty to take corrective steps to prevent the perpetuation of such discrimination by their affiliates. Id. at 1360. Central to Kaplan’s holding, however, is that the international union actually negotiated on behalf of the local the agreement containing the discriminatory referral procedure. See id. at 1359-60. In contrast, LIUNA was not involved either in the negotiation of the 1973 PLA, or in Local 496’s decision not to comply with the PLA’s requirements. Moreover, in Kap-lan, the provision perpetuating the discriminatory effects was part of the collective bargaining agreement itself. In the present case, Plaintiffs did not allege and the district court did not find that the PLA itself was discriminatory; it was the failure of Local 496 (by referring only members) and the employers (because of security at Perry) to properly implement the agreement that caused the discrimination.
Plaintiffs argue that requiring anything more than an international union’s knowledge of discrimination claims against a local affiliate to create an affirmative duty would be “in direct conflict with the broad remedial purposes of Title VII and Section 1981, as well as LIUNA’s duties under the National Maintenance Agreement.” In General Building Contractors Association, however, the Supreme Court explicitly stated
[T]he question is not whether the employers and associations are free to delegate their duty to abide by § 1981, for whatever duty the statute imposes, they are bound to adhere to it. The question is what duty does § 1981 impose. More precisely, does § 1981 impose a duty to refrain from intentionally denying blacks the right to contract on the same basis as whites or does it impose an affirmative obligation to ensure that blacks enjoy such a right? The language of the *429statute does not speak in terms of duties. It merely declares specific rights held by “[a]ll persons within the jurisdiction of the United States.” We are confident that the Thirty-ninth Congress meant to do no more than prohibit the employers and associations in these cases from intentionally depriving black workers of the rights enumerated in the statute, including the equal right to contract. It did not intend to make them the guarantors of the workers’ rights as against third parties who would infringe them.
458 U.S. at 396 (second emphasis added). Thus, at least under § 1981, in the absence of an agency relationship, an international union does not have an affirmative duty to ensure that its affiliates do not undertake discriminatory practices. See also Goodman v. Lukens Steel Co., 482 U.S. 656, 687-89, 107 S.Ct. 2617, 96 L.Ed.2d 572 (1987) (Powell, concurring);18 cf. Phelan v. Local 305 of United Ass’n of Journeymen, 973 F.2d 1050, 1061 (2d Cir.1992) (in context of union democracy claim, citing Carbon Fuel for the proposition that: “An international union has no independent duty to intervene in the affairs of its local chapters, even where the international has knowledge of the local’s unlawful acts.”). Because of the close relationship between Title VII and § 1981, I conclude that the same standard applies in the Title VII context. See Berger, 843 F.2d at 1429 (noting that an international’s liability for the actions of an affiliate must be based on something “more than the abstract and unbounded premise that the entities regulated by [Title VII and § 1981] have an ‘affirmative duty’ to end discrimination.”). The non-discrimination clause of the National Maintenance Agreement likewise does not require anything greater.
Moreover, as a practical matter, holding LIUNA responsible for Local 496’s actions, under the facts of this case, makes little sense. In finding an affirmative duty, the district court relied in part on the fact that in March 1985, LIUNA signed the National Maintenance Agreement. Before 1985, LIUNA was not a party to any Perry agreement. After May 1985, no class member even applied for union membership until 1990. During that period, no class member filed an EEOC grievance claiming additional discrimination. In addition, no class member ever requested that LIUNA investigate the charges of discrimination. As far as LIU-NA was aware, the institution of the October 1987 referral policy, which the court acknowledged appeared to be neutral on its face, corrected any discrimination problems that had appeared in the past.
For the foregoing reasons, I would hold that the district court erred in finding that LIUNA had an affirmative duty to oppose Local 496’s discriminatory practices.19
. LIUNA also claims that the district court erred in finding that discrimination occurred within the limitations period applicable to LIUNA and in holding LIUNA liable pursuant to a continuing violations theory. Plaintiffs also claim that the district court committed error in two of its damages rulings pertaining to LIUNA. Because I conclude that the district court erred in holding LIUNA liable, I would not address these issues.
. For ease of discussion, except where otherwise noted, we include Defendant Conrad in collective references to "Local 496.”
. The PLA was signed by representatives of Perry’s owners and of all of the trades, including Local 496. The provision in question, Art. VIII, provides in pertinent part:
Contractors performing construction work on the Perry Nuclear Power Plant project shall, in filling job vacancies, utilize the registration facilities and referral systems operated by the Local Unions in accordance with the provisions of applicable Federal and States laws.
The selection of applicants for work on this project shall be on a nondiscriminatory basis and shall not be based on, or in any way affected by, union membership, bylaws, rules, regulations, constitutional provisions, or any other aspect of union membership, policies, or requirements.
(emphasis added).
The PLA was revised in March 1983, but the provisions regarding hiring of laborers remained unchanged.
. In the settlement agreement, however, the parties stipulated that if he had been asked, Conrad would have testified that after the adoption of the 1987 referral rules, nonmembers who followed the policy "would” have been referred pursuant to the policy’s written terms.
. Despite the fact that class counsel had a copy of the 1987 referral rules, he did not discuss the notification requirement or hew referral rules with Colvin, Tomblin, or any other class member.
.In those interrogatories, LIUNA stated:
During the period from January 1, 1982, to the date of production, defendant LIUNA has not formally investigated a claim of racial discrimination against a local affiliate. Customarily, upon receipt of letters from mem*418bers which charge a local affiliate with racial discrimination or with some other misfeasance or nonfeasance, the General President refers the letters over to the appropriate regional manger. The regional manager, in turn, normally contacts all the parties involved in an attempt to resolve the matter informally.
. Local 496 also asserts that the district court improperly found that Conrad was "individually” liable under Title VII. For the record, we note that the court did not make such an explicit determination, instead stating that "Local 496 and Floyd Conrad have engaged in a pattern or practice of discrimination which constitutes an unlawful employment practice in violation of 42 U.S.C. §§ 2000e-2(a), (c), and 42 U.S.C. § 1981.” In any event, we agree with Local 496 that for the same reasons as stated in Wathen v. General Elec. Co., 115 F.3d 400 (6th Cir.1997), with respect to the liability of an "employer's” "agents,” Conrad cannot be held personally liable under Title VII as the agent of a "labor organization.” See id. at 403-06; cf. 42 U.S.C. § 2000e-2.
. The parties stipulated that as of April 1985, Local 496’s members resided in the following counties: Lake — 290 (54%); Ashtabula — 125 (23.5%); Cuyahoga — 49 (9.2%); Geauga — 33 (6.2%); Other — 34 (6.4%).
. Dr. Pendleton testified that because of the road systems, he used the population figures for all of Cuyahoga county; he acknowledged, however, that the majority of Cuyahoga County's black population resided in the county’s *421eastern half, and noted that the Perry Plant was within 20 miles from the east-side.
. Between 1983 and 1985, the standard deviations for the SOC ranged from -9.77 to - 10.36; the standard deviations for the GOC ranged from -9.88 to -10.46; the standard deviations for the total labor force ranged from -7.75 to -8.34. Any standard deviation number smaller than -2.00 (e.g. -2.5) constitutes a statistically significant underrepresen-tation, i.e., one where the probability that the resulting numbers occurred by chance is extremely small. See Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 309 n. 14, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977).
. Dr. Martin found that there were a total of 6,488 workers in the GOC from the four-county area that work in Lake County, and that 358, or 5.5%, are black. There are 1,195 workers in the SOC from the four-county area who work in Lake County, and that 52, or 4.35%, are black. The overall workforce in Lake County is 1% black.
. The court also noted that between 1975 and 1979, Local 496’s membership grew from approximately 100 to over 500 members (400 + %), but that during that time, black membership increased only from 10 or 12 members to 18 members (less than 100%). Between April 1982 and February 1984, 42 new members were initiated into Local 496, none of whom was black. Between 1980 and 1985, the union accepted 54 new members, one of whom was black.
. While the court stated that the expert reports would not "control,” it also acknowledged that it would consider the reports in concluding whether or not there was discrimination. During the trial, the court asked Dr. Pendleton questions indicating that it might be more appropriate to include only the east side of Cuyahoga County. As discussed below, this likely would result in a greater percentage of blacks in the relevant labor market.
. Local 496 asserts that instead of declaring "the four-county area” to be the relevant market, the court needed to identify further the interested and/or available applicants within those geographic boundaries. I disagree.
As an initial matter, I note that Local 496 spends a significant amount of time arguing that Dr. Pendleton’s findings are "clearly erroneous” and/or legally insufficient, and that its expert’s report was "superior.” Contrary to Local 496’s assertions, however, we review the court's findings, not the expert's opinion for clear error. Thus, I construe Local 496's arguments directed toward the district court’s findings.
Citing Chicago Miniature Lamp Worlcs, Local 496 argues that the court erroneously failed to factor commuting times into the relevant population determination. Chicago Miniature involved the hiring practices of a manufacturing company, located in an Hispanic and Asian neighborhood of Chicago, for entry-level positions requiring few skills and offering low pay that did not rise substantially over time. 947 F.2d at 294-95. The court stated: "The identification of a relevant labor market — the key issue in a class-based Title VII case — means not only identifying qualified potential applicants for the job at issue but also identifying interested potential applicants.” Id. at 302 (emphasis in original). The court went on to hold that because low paying jobs are more likely to be filled by those living closer to the job (because the cost of a possible Pk hour commute cannot be justified), consideration of commuting time was especially important to the case before it. Id. The court also noted that the boundaries were arbitrarily drawn such that they did not include some areas just over 5 miles from the plant, but did include other areas as far as 20 miles away, id. at 302 n. 8, and that the district court had failed to consider that the company’s lack of an English-fluency requirement meant that it would receive a disproportionately large number of applications from non-English speaking persons. Id. at 302-03.
Chicago Miniature itself recognized that "a statistical model [in a discrimination case] need not be completely specified and some arguably relevant variables can be omitted in certain cases.” Id. at 300. While it might have been helpful for the court to have explained its reasons for concluding that commuting time did not affect its relevant market definition, such explanation was not mandated in this case. In sharp contrast to the low-wage situation in Chicago Miniature, the record reflects that the jobs at Perry offered more than double the average wage available in other positions. Obviously, higher wages make jobs with longer commutes much more attractive. Moreover, under its own argument, Local 496 must acknowledge that including all of Cuyahoga County, as opposed to just the east side, works in its favor.
Local 496 also asserts that the court would need somehow to factor into the calculation the fact that for proximity reasons, part of the relevant population residing in Cuyahoga County might join the two other local laborers unions instead of joining Local 496 and working at Perry. (Local 310 has approximately 1100 members, and Local 860 has approximately 850 members.) Local 496 has failed to cite, and I cannot find, any support for the assertion that to satisfy Wards Cove, the relevant population figures must factor in a deduction for persons with other job prospects. At least in the context of this case, I reject this contention. See Newark Branch, NAACP v. Town of Harrison, 940 F.2d 792, 800 (3d Cir.1991) (in assessing the district court’s definition of the relevant labor market, noting: "Nothing in Wards Cove or other cases interpreting Title VII so much as suggests that *424available 'outside' job opportunities have any relevance whatever to the Title VII liability of a particular employer.”).
. The district court proceeded to find that the union's asserted business justification was "clearly a pretext,” and that in any event, there was a less restrictive alternative. On appeal, Local 496 has not asserted that this finding was in error, instead addressing only the requirements of Plaintiffs’ prima facie case.
. The Supreme Court has previously observed that Congress clearly limited an international union's liability for acts of its affiliates to situations where a common-law agency relationship existed. See Carbon Fuel Co. v. United Mine Workers, 444 U.S. 212, 216-18, 100 S.Ct. 410, 62 L.Ed.2d 394 (1979). Although Carbon Fuel involved the breach of a collective bargaining agreement by a local affiliate, the Court has also required the existence of an agency relationship to impose vicarious liability for discriminatory practices. See General Bldg. Contractors Ass’n, Inc. v. Pennsylvania, 458 U.S. 375, 391-95, 102 S.Ct. 3141, 73 L.Ed.2d 835 (1982) (discussing 42 U.S.C.A. § 1981 liability of trade associations and employers for local union's discriminatory practices).
In Berger, relying in part on Carbon Fuel and General Building Contractors, the D.C. Circuit explicitly held that common-law agency principles apply to unions and their internationals in the civil rights context. 843 F.2d at 1427-29. We think that Berger aptly ties *427together the existing case law and sets forth the correct approach.
I am aware that in Myers v. Gilman Paper Corp., 544 F.2d 837 (5th Cir.1977), the Fifth Circuit stated that to hold an international union liable for the discrimination of a local affiliate, "[t]here must be a 'sufficient connection’ between the labor organization and the discriminatory practice.” Id. at 851. I agree with the D.C. Circuit, however, that Myers's actual application of the "sufficient connection” test does not support using in the area of Title VII and § 1981 an agency standard less stringent than that required by common-law agency principles. See Berger, 843 F.2d at 1428. In Myers, the international union was liable because of the "close relationship” with its local under which the international provided "advisors” who helped negotiate the discriminatory provision and required that the local submit its contract to the international for approval. 544 F.2d at 851. In other words, the international "participated in” and "approved of” the discriminatory practice. Moreover, Myers predated both Carbon Fuel and General Building Contractors; thus, to the extent it permitted a lesser connection between the international union and its affiliate, Myers has been superceded.
. In Carbon Fuel, the Court described the necessary agency relationship as requiring that the international union "instigated, supported, ratified, or encouraged” the local affiliate’s conduct in question. 444 U.S. at 218. Somewhat differently, the Court in General Building Contractors stated: " 'Agency is the fiduciary relation which results from the manifestation of consent by one person to another that the other shall act on his behalf and subject to his control, and consent by the other so to act.' ” 458 U.S. at 392 (quoting Restatement (Second) of Agency § 1 (1958)). The Court particularly emphasized the idea of “control,” noting that the “power to oppose” union discrimination is not tantamount to a "right to control” the union. Id. at 393. The Berger court’s characterization of agency adequately accounts for these various definitions.
. In Goodman, the Supreme Court declined to reach the question of whether a trade union has an affirmative duty to oppose the racial discrimination of others (because the unions' practice of refusing to file any and all grievances presented by a black person on the ground that the employer would resent such grievances was an actionable violation in and of itself), but Justices Powell, Scalia, and O’Connor expressed their skepticism that such an affirmative duty could be imposed. 482 U.S. at 687-89.
Plaintiffs seek to distinguish the concurring opinion in Goodman, noting that it states that an affirmative duty to oppose “employer” discrimination would work a disruption in the basic policies of labor laws, whereas here, the discrimination is by a "local union,” over which the international has a greater measure of control. I do not find this distinction to be of any import: federal law also protects the autonomy of local unions. Cf. 29 U.S.C.A. §§ 461-66 (West 1985 & Supp.1998) (regulating the suspension of local union autonomy).
. Although not acknowledged by the district court, the evidence shows that in the 1984-85 time period, LIUNA, through Regional Repre-*430senlative Tom Arconti, did make some effort to investigate discrimination claims.