Full opinion text
AMENDED AND CORRECTED OPINION ROBERT L. CARTER, District Judge. Plaintiffs City of New York and New York State Division of Human Rights ask the court to hold defendant Local 28 of the Sheet Metal Workers International Association (“Local 28”) in contempt of court on the basis of numerous alleged violations of this court’s Order and Judgment entered in 1975 (“0 & J”) and the Amended Affirmative Action Program and Order (“AAAPO”) entered by this court in 1983. In the alternative, they seek modification of the AAAPO. Plaintiff Equal Employment Opportunity Commission is not a party to this motion. The parties also ask the court to approve several stipulations into which they have entered. This court has continuing jurisdiction over this ease pursuant to 28 U.S.C. § 1331 because the ease was originally brought under Title VII of the 1964 Civil Rights Act, 42 U.S.C. § 2000e-6(a). It also has jurisdiction pursuant to the court’s inherent powers to enforce its orders and to punish violators for contempt. Roadway Express, Inc. v. Piper, 447 U.S. 752, 764-765, 100 S.Ct. 2455, 2463-2464, 65 L.Ed.2d 488 (1980); 18 U.S.C.A. § 401 (1966). I. Background Local 28 is a union representing sheet metal workers who work for contractors in the New York metropolitan area. The Local 28 Joint Apprenticeship Committee (“JAC”) is a management-labor committee which runs an apprenticeship program to teach sheet metal skills to entrants into the union. The Sheet Metal and Air Conditioning Contractors Association of New York City, Inc. and the Sheet Metal and Air Conditioning Contractors National Association of Long Island, Inc. (“Contractors’ Associations”) are trade associations of building contractors which employ sheet metal workers in New York City and Long Island. The Contractors’ Associations were not named in the original complaint but have since been joined by the court for the purposes of obtaining complete relief. The background of this case has been thoroughly documented in previous opinions of this court, the Second Circuit and the Supreme Court, see Local 28 of Sheet Metal Workers’ Int’l Ass’n v. EEOC, 478 U.S. 421, 106 S.Ct. 3019, 92 L.Ed.2d 344 (1986); EEOC v. Local 638 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, 753 F.2d 1172 (2d Cir.1985), aff'd, 478 U.S. 421, 106 S.Ct. 3019, 92 L.Ed.2d 344 (1986); EEOC v. Local 638 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, 674 F.Supp. 91 (S.D.N.Y.1987) (Carter, J.), but for purposes of clarity this opinion will provide a summary of the ease. This case dates back to 1964 when the state of New York commenced proceedings in state court against Local 28 in response to the union’s refusal to admit blacks as members and apprentices. In 1971, the United States charged Local 28, the JAC, and several other unions and contractors’ associations with violating Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq. The Local 28 case was severed, the EEOC was soon substituted for the United States, and the New York City Commission on Human Rights (“City”) and New York State Division on Human Rights (“State”) intervened as plaintiffs. Judge Werker of this court held a trial and determined that Local 28 and the JAC had violated Title VII and New York law by denying nonwhites access to employment opportunities in the sheet metal trades. EEOC v. Local 638, 401 F.Supp. 467 (S.D.N.Y.1975) (Werker, J.), aff'd in part, 532 F.2d 821 (2d Cir.1976). He then entered the O & J, which enjoined Local 28 and the JAC from discriminating against nonwhites and enjoined a number of specific practices which he found to be discriminatory. He also ordered the union to achieve 29% nonwhite membership by July 1, 1981, awarded back pay to nonwhites excluded from union membership, appointed an administrator (“Administrator”), and ordered the parties to design new recruitment and admission procedures to achieve the membership goal. The Administrator accordingly proposed an affirmative action program, which the court adopted. The Second Circuit affirmed Judge Werker’s finding of liability and most of the remedies that he had ordered, modifying only part of his remedy regarding the apprenticeship program. EEOC v. Local 638 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, 532 F.2d 821 (2d Cir.1976). On remand, Judge Werker adopted a Revised Affirmative Action Plan and Order (“RAAPO”) which, among other things, extended the 29% membership goal for one year, and the Second Circuit affirmed this plan. EEOC v. Local 638 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, 565 F.2d 31 (2d Cir.1977). In 1982, the City and State commenced a series of contempt proceedings strikingly similar to the current proceeding, alleging that Local 28 and the JAC had failed to meet the membership goal because they had violated a number different provisions of the O & J and RAAPO. Judge Werker issued a civil contempt order, finding that Local 28 and the JAC had “failed to comply with RAAPO ... almost from its date of entry,” and imposed a fine of $150,000, which he ordered Local 28 and the JAC to deposit in a fund to be used to increase nonwhite membership in the union. EEOC v. Local 638 ... Local 28 of Sheet Metal Workers Int’l Ass’n, No. 71 Civ. 2877, 1982 WL 445, *4 (S.D.N.Y. Aug. 16, 1982) (Werker, J.). In 1983 the City brought a second contempt proceeding on the basis of the failure of the union and the JAC to maintain and submit records in violation of the O & J, the RAAPO, and several administrative orders. The Administrator found contempt, and Judge Werker adopted his findings, ordering the union and the JAC to finance a computerized recordkeeping system. He also entered an Amended Affirmative Action Plan and Order (“AAAPO”) which, among other measures, increased the nonwhite membership goal to 29.23% and extended the deadline for meeting the goal to August 31, 1987. In 1987 this deadline was again extended to August 31, 1992, pursuant to a stipulation between the parties. The Second Circuit affirmed most of Judge Werker’s contempt findings, upheld the contempt remedies he had ordered, and affirmed the AAAPO with modifications to the apprenticeship program, Local 28, 753 F.2d at 1172, and this opinion was upheld by the Supreme Court. Local 28, 478 U.S. at 482-83, 106 S.Ct. at 3053-54. When Judge Werker passed away, I inherited the case. In the instant contempt motion, the plaintiffs allege that the union has violated the AAAPO and O & J by fading to achieve the membership goal, ensure equal work opportunities for its white and nonwhite members, and keep accurate and complete records. Specifically, the plaintiffs claim that the union has inflated its census reports to make it appear as if it were closer to achieving the membership goal than it actually is, has permitted and contributed to a disparity between the hours worked by white and nonwhite members, and has adopted a reinitiation policy having a disparate impact on minorities. The plaintiffs seek a wide range of remedies for the alleged contempt. To remedy the hours disparity, they ask the court to order the union to operate a hiring hall which will refer its members to jobs and to implement a work sharing system in which jour-neypersons would return to the hiring hall after working for a given period of time. They also seek back pay for nonwhite jour-neypersons whose hours fall two or more standard deviations below the norm. To remedy the failure to meet the nonwhite membership goal, plaintiffs ask the court to calculate the membership goal separately for journeypersons, to set up interim goals, and to invalidate the reinitiation policy. To remedy the inaccurate and incomplete record-keeping, the plaintiffs request the court to order the union to implement and pay for a new recordkeeping system. Finally, the plaintiffs seek attorneys’ fees, court costs, and reimbursement of experts’ fees. In response to this contempt motion, the parties have entered into several stipulations which they request the court to approve. In the first stipulation, the union agrees to institute a hiring hall and work share program similar to that requested by the plaintiffs in their contempt motion, and the plaintiffs agree to withdraw that request if the stipulation is approved. See Appendix A, Proposed Stip. & Order Regarding Hiring Practices (“hiring hall stipulation”). In the second stipulation, the union agrees to modify its reinitiation policy, and the plaintiffs agree that as soon as the unión has done so they will withdraw their request for invalidation of the current reinitiation policy. See Appendix B, Proposed Stip. & Order Regarding Reini-tiation to the Union (“reinitiation policy stipulation”). In the third stipulation, the union agrees to institute and pay for a new computerized recordkeeping system, and the plaintiffs agree to withdraw their request for a new recordkeeping system. I have signed the recordkeeping stipulation — the other two will be discussed in this opinion. II. Liability for Contempt A Contempt Standards In order for the court to find contempt on the part of the union for not complying with a court order, the order must be “clear and unambiguous, the proof of noncompliance [must be],... ‘clear and convincing,’” and the union must not have “been reasonably diligent and energetic in attempting to accomplish what was ordered.” Local 28, 753 F.2d at 1178 (quoting Powell v. Ward, 643 F.2d 924, 931 (2d Cir.1981), cert. denied, 454 U.S. 832, 102 S.Ct. 131, 70 L.Ed.2d 111 (1981)). It is not necessary that the union willfully disobey the court’s order for there to be contempt. McComb v. Jacksonville Paper Co., 336 U.S. 187, 191, 69 S.Ct. 497, 499-500, 93 L.Ed. 599 (1949); Local 28, 753 F.2d at 1178. Parties are bound by a court order until the court modifies the order or releases them from it, and defendants who act without first asking the court to clarify the order “act[] at their own peril.” McComb, 336 U.S. at 192, 69 S.Ct. at 500; see also Maness v. Meyers, 419 U.S. 449, 95 S.Ct. 584, 42 L.Ed.2d 574 (1975). On several prior occasions this court has been forced to remind the union that it is not free to disregard court orders simply because it does not believe that it is bound by them. See, e.g., EEOC v. Local 688 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, No. 71 Civ. 2877, Mem.Decision by Special Master David Raff, at A-131—A-132 n. 5 (S.D.N.Y. May 18, 1983) (Werker, J.) (“Administrator’s Decision”). The union continues to insist that it may decide when to violate the court’s orders, evincing its blatant disrespect for the court and for the principles underlying those orders. See infra p. 653. A contempt proceeding does not provide an opportunity for a party to relitigate the factual basis of a court order. United States v. Rylander, 460 U.S. 752, 756, 103 S.Ct. 1548, 1552, 75 L.Ed.2d 521, reh’g denied, 462 U.S. 1112, 103 S.Ct. 2466, 77 L.Ed.2d 1342 (1983). Consequently, the court will not consider claims that its orders were unreasonable when they were issued, but the union may defend against the contempt citation by showing that circumstances have arisen that make compliance with the court’s order presently impossible and that were unforeseen when the order was issued. Id. at 757, 103 S.Ct. at 1552-53; see also Badgley v. Santacroce, 800 F.2d 33, 36 (2d Cir.1986), cert. denied, 479 U.S. 1067, 107 S.Ct. 955, 93 L.Ed.2d 1003 (1987); United States v. Chase Manhattan Bank, N.A., 590 F.Supp. 1160 (S.D.N.Y.1984) (Goettel, J.) (refusing to permit impossibility defense where none of the facts had changed). Defendants have the burden of production in such eases, and this burden may be difficult to meet, especially “where the defendants have a long history of delay and the plaintiffs’ needs are urgent.” Badgley, 800 F.2d at 36; see also EEOC v. Local 580, Int’l Ass’n of Bridge, Structural & Ornamental Ironworkers, 669 F.Supp. 606, 611 (S.D.N.Y.1987) (Carter, J.), aff'd, 925 F.2d 588 (2d Cir.1991). The burden is “certainly not less where the obligations in question were accepted in a decree entered on consent.” Aspira of N.Y., Inc. v. Board of Educ. of N.Y., 423 F.Supp. 647, 654 (S.D.N.Y.1976) (Frankel, J.). An impossibility defense will not lie where the defendant cannot demonstrate that he has made diligent efforts to comply with the order. “A person charged with civil contempt is entitled to notice of the allegations, the right to counsel, and a hearing at which the plaintiff bears the burden of proof and the defendant has an opportunity to present a defense.” United States v. Yonkers, 856 F.2d 444, 452 (2d Cir.1988). Under Local Civil Rule 43(b), a district court need not take oral evidence if the alleged contemnor does not demand a hearing. In this case, the union has not requested a hearing, and in light of the great volume of evidence submitted by the parties and the substantial amount of uncontested evidence, the court does not find it necessary to hold a hearing and will decide this motion on the parties’ submissions. B. Methodology Both plaintiffs and the union use statistical experts to present their evidence regarding the contempt motion, and each expert attacks the credibility and methodology of the other. The statistical evidence forms the basis of the plaintiffs’ case, so the court finds it necessary to assess the validity of the statistical studies. For over a decade, in his role as an expert for Local 28, Dr. Richard Buchanan' has used statistics in a misleading manner. At one point he used methods that were “so clearly misdirected as to lead to the conclusion that they were used in a conscious effort to result in depressed findings.” EEOC v. Local 638 ... Local 28 of Sheet Metal Workers’ Int’l Ass’n, No. 71 Civ. 2877, Memorandum and Order, slip op. at 4 (S.D.N.Y. Aug. 31, 1983) (Werker, J.). At another, he “lost his credibility when he attempted to use statistics to explain away every compliance problem that defendants had,” EEOC v. Local 638 ... Local 10, No. 71 Civ. 2877, Report and Recommendation of Special Master David Raff at 46 (S.D.N.Y. Jan. 19, 1993) (Carter, J.), and gave opinions that “defie[d] common sense,” id. at 49-50. In the report submitted by the union in response to the plaintiffs’ contempt motion, Dr. Buchanan continues his practice of attempting to mislead the court. In his haste to criticize the plaintiffs’ expert, Dr. Bernard Siskin, Dr. Buchanan often fails to familiarize himself with the data and to understand Dr. Siskin’s methods. For example, Dr. Buchanan claims that Dr. Siskin did not take into account the fact that the northern counties of New Jersey demerged from Local 28 in June 1991, and that as a consequence Dr. Siskin’s data for that year is tainted by the inclusion of people who were no longer Local 28 members. Dr. Buchanan ignores the fact that all of Dr. Siskin’s data ended in March 1991 precisely in order to avoid the difficulties posed by the demerger. Similarly, Dr. Buchanan notes that Dr. Siskin examined the numbers of hours that journeypersons worked in roofing and ventilation shops to determine whether they had special skills. Dr. Buchanan asserts that this analysis is invalid because these shops are typical, not special, disregarding Dr. Siskin’s explanation that he uses roof and ventilation work as a proxy for lack of special skills. Finally, Dr. Buchanan claims that in conducting regression analyses regarding the effect of various factors on the hours worked by journeyper-sons Dr. Siskin combined data from several different years and that the data should not have been combined because economic conditions during those years varied widely. Had Dr. Buchanan paid attention to Dr. Siskin’s results, he would have seen that Dr. Siskin did not run his regression analyses for the entire period but rather ran them for each year and then calculated the average disparity for the entire period, which he expressed both as a number and in units of standard deviation. When Dr. Buchanan cannot find the source of Dr. Siskin’s data he simply asserts, with no basis, that Dr. Siskin must have fabricated the data. For example, Dr. Siskin used a computer file that the union compiled which listed the hours worked by each member to assess the disparity between hours worked by white and nonwhite workers. Dr. Buchanan claims that there is no race data included in this file, but he seems to have looked at the wrong file. He also complains that in reconstructing the union’s census reports Dr. Siskin might have used data from the “APTFIL” hours file for periods of time for which the union failed to compile the data, ignoring the fact that Dr. Siskin states in his report that he used only the membership files to reconstruct the census and did not use data from the hours files. Other of Dr. Buchanan’s criticisms of Dr. Siskin stem from flaws in the data that the union provided to the plaintiffs. The plaintiffs seem to have satisfactorily compensated for this faulty data, and it does not seem to have undermined the integrity of their analy-ses. For example, Dr. Buchanan claims that Dr. Siskin was unaware of the fact that the union’s membership files contain data for people who entered the union as production and asbestos workers, who are not covered by the AAAPO, and who later switched to sheet metal work. Dr. Buchanan fails, however, to estimate how much error this oversight could have inserted into the plaintiffs’ statistics. In fact, Dr. Siskin requested the union to provide information regarding how many of these workers were included in the files. The union was unable to provide an exact figure, but it did estimate that throughout the period 1984-1991 there were no more than twenty-five such members and that the number had dwindled to one member by 1993. Thus, there were too few workers involved to skew the plaintiffs’ statistics. In addition to their statistical evidence, the plaintiffs also submit affidavits of nonwhite union members dated from December 1992 through June 1993 in support of their contempt motion, so that one can truly understand the impact that the contempt had on the lives of the individual journeypersons. These stories of unrealized dreams and personal pain are summarized in Appendix C in order to give the statistics more meaning and context. In their affidavits the journeyper-sons allege that white mechanics were given overtime opportunities, were allowed to work at positions which enabled them to learn skills, and were employed on a long-term basis while the minority journeypersons were denied all of these benefits. Jamie K. Nicastri, an associate in the firm of Edmund P. D’Elia, attorney for Local 28, seeks to discredit the “anecdotal” testimony of the minority journeypersons. (Nicastri Aff. in Opp. to Plaintiffs Mot. for Contempt ¶ 4.) Nicastri discounts their stories of employment discrimination and of lack of access to opportunities for more skilled and economically rewarding employment. Nicastri claims that the affiants have not shown that whites were given preferences over minorities, that the contractors engaged in discriminatory treatment and that the minority affi-ants had satisfactory attendance records. These arguments are unconvincing. The record shows that most of the poor attendance records occurred during an individual’s apprenticeship and that minority journeyper-sons were not denied work because they were unskilled. Furthermore, the union’s obligation under the consent decree requires that Local 28 investigate discrimination claims, especially in the face of the undisputed evidence that white journeypersons junior in work experience to some of the senior minority journeypersons are working steadily as mechanics while minority persons once they finish the apprentice program are not. The responses to the claims of the minority journeypersons are evidence that the union is not in compliance with the AAAPO. C. Violations 1. Recordkeeping Violations Pursuant to the 0 & J and AAAPO, the union is required to provide and maintain certain records in order to ensure its compliance with court orders. See O & J at ¶ 21(e); AAAPO at ¶¶ 4, 50-60. This court has concluded that “accurate reporting of membership is an essential part of the [Amended Affirmative Action] Program” and is necessary to ensure compliance with its mandates. EEOC v. Local 638, 421 F.Supp. 603, 606, 620 (S.D.N.Y.1975) (Werker, J.), modified, 532 F.2d 821 (2d Cir.1976); see also Local 28, 478 U.S. at 444 n. 23, 106 S.Ct. at 3033 n. 23 (“the detailed recordkeeping requirements ... w[ere] clearly designed to foster petitioners’ compliance”); Local 28, 1982 WL 445, at *3 (recordkeeping and reporting were “absolutely vital to the effective monitoring and implementation” of the affirmative action plan). However, Local 28 has a tradition of failing to provide and maintain accurate records as required by the O & J and the AAAPO, resulting in prior contempt findings by this court. See Local 28, 753 F.2d at 1182 (upholding 1982 and 1983 contempt findings on the basis of “union’s ... egregious noncompliance”); Administrator’s Decision at A-131, A-133 (“Local 28’s failure to provide complete, accurate and timely data” resulted in contempt finding). This court has noted that “[t]he record in both state and federal court against these defendants is replete with instances of their bad faith attempts to prevent or delay affirmative action.” Local 28, 478 U.S. at 476-77, 106 S.Ct. at 3050 (quoting Local 28, 401 F.Supp. at 488). Once again, Local 28 has been accused of failing to provide the requisite documentation, in contempt of the court’s orders. Neither this court nor either of the parties asserts that the O & J or the AAAPO is unclear as to their recordkeeping provisions. Pursuant to paragraph 58 of the AAAPO, “[d]efendants shall maintain all records and/or lists in the computer base which are necessary to produce all the reports required pursuant to the O & J and this Amended Program.” In addition, [a]ll records and lists required to be compiled by the O & J and this Amended Program shall be maintained until such time as the Court terminates this Amended Program and shall be made available for inspection and copying by plaintiffs and the Administrator.... Plaintiffs and the Administrator shall be permitted access to all computer tapes containing records or reports required by the O & J or tins Amended Program. AAAPO at ¶ 60. Despite the plan’s unequivocal mandate that the union compile and keep the requisite information, at least since 1984 Local 28’s maintenance of its records has been irresponsible and sloppy at best, and purposefully deceptive at worst. Census reports have neither been saved on computer nor retained on hard copy, and records of hours worked have not been maintained as required by paragraph 52 of the AAAPO. For example, the union has continually updated its computer database without saving any copies of the biennial census reports, and it could not provide hard copies of the census reports for March 1989 and March 1990. in addition, the union did not provide hours files from October 1988 through December 1988, as required by paragraph 52 of the AAAPO. Local 28 does not deny that it failed to keep these required records. Rather than explain-why it has failed to supply the court with the requisite records, the union gives a laundry list of documents that it did manage to submit. This is no defense. The union does not have the discretion to pick and choose which information it will surrender to the court and which it will withhold. The 0 & J and the AAAPO require that Local 28 submit all of the enumerated documents; anything less is noneompliance. In prior contempt findings against Local 28, this court has concluded that where recordkeeping falls short of what its orders demand there is contempt. For example, the court has concluded that [cjompliance with the quarterly reporting requirements is not satisfied by anything less than four complete reports each year. The defendants’ contention that two reports in 1979, two reports in 1980 and no reports in 1981 is sufficient does nothing more than evince their blatant disregard for their obligation to provide the appropriate parties to this suit with information pertinent to the enforcement of the 0 & J and RAAPO. Local 28, 1982 WL 445, at *4. The court has also found contempt where “names and racial or ethnic identities of [new union members] were not provided to the ... Administrator.” Administrator’s Decision at A-180, A-131. The relevant sections of the court’s orders do not require Local 28 to produce information that is not in its possession or under its control, rendering compliance with the court order factually impossible. See discussion supra pp. 649-650. The union does not claim that it is presently incapable of producing the documents in question, and furthermore, it is the union’s obligation to inform the court if it is unable to furnish the requisite records — it may not make an independent determination not to submit certain information. See Administrator’s Decision at A-137, A-140 (“a unilateral decision that various provisions cannot or will not be met will continue-to lead to findings of violation”). Not only has the union failed to maintain the records required of it, but the records that it has submitted to the court have been inaccurate. Pursuant to paragraph 51 of the AAAPO, twice a year Local 28 shall submit to' the plaintiffs and the Administrator both a geographical census, by former local, and a master census of its membership which shall include ... a) the total number of journeymen; b) the-total number of apprentices; c) the percentage of non-whites and d) the name, address, race/national origin, social security number and number of hours worked in the prior six (6) months by each Local 28 journeyman and apprentice. Paragraph 4 of the AAAPO states clearly that [f]or the purpose of measurement, total membership shall include: (a) all journeymen members; (b) all pensioners, reduced or limited members who, while receiving any retirement benefit ... have been employed as sheet metal workers by a Local 28 contractor within three (3) years prior to the date of the most recent membership census; (c) all members or participants in the Apprentice Program; and (d) all individuals who (i) have been offered admission to and membership in Local 28 but have ... deferr[ed] for no more than two years. The “defendants must insure, to the greatest degree possible, that the data submitted to the parties are true and correct.” Administrator’s Decision at A-132. Despite the plan’s plain language, the union’s compilation of census report data has been haphazard and full of errors. First, the reports included union members who had been terminated for non-payment of dues and were no longer considered members under Local 28’s own Constitution. Members who are suspended and then terminated for non-payment of dues “forfeit all rights, privileges, and benefits of membership.” Constitution and Ritual of the Sheet Metal Workers International Association, Art. 16, § 10. For example, Travers Everson, an African-American former member, was included in the March 1993 census report, despite the fact that his membership was terminated in 1990 for failure to pay dues as a result of unemployment. This inclusion of suspended and terminated members inflates the union’s nonwhite membership because nonwhites are suspended at a higher rate than their white counterparts. See discussion infra pp. 665-666. For example, the March 1991 hard copy list of membership submitted by the union as required by paragraph 51 of the AAAPO included 162 individuals whose membership had been terminated by March 1991, 38.9% of whom were nonwhite. Excluding these terminated individuals would bring the percentage of nonwhite joumeypersons from 16.44% to 15.52%, and the percentage of nonwhite membership in general from 21.40% to 20.76% for March 1991. Second, the census reports excluded a group of disproportionately white union members who were listed as active members in the union’s membership files. For example, the March 1991 hard copy list of membership submitted by the union excluded 497 individuals who were still considered active members and who had not retired within the past three years. Over 92% of the excluded individuals were white, and including these active members would reduce the percentage of nonwhite membership from 21.40% to 20.02% for March 1991. Third, the census reports contained members who either were not listed in the union’s membership flies or were considered inactive. For example, the March 1991 census hard copy list included nine individuals who did not appear in the union’s membership files, 77.8% of whom were nonwhite. Furthermore, the March 1991 hard copy list included thirty-eight joumeypersons whom the union’s membership files indicate were inactive, 15.8% of whom were nonwhite, although nonwhites comprised only 14% of all joumeypersons during that month. Both of these inaccuracies serve to inflate the nonwhite membership rate. Finally, the census reports counted some white members as nonwhite. For example, the March 1991 census hard copy list identified seventeen joumeypersons as nonwhite, although the membership files identified them as white. This court has concluded that misreporting the race of union members constitutes grounds for contempt. Local 28, 753 F.2d at 1182 (finding contempt in part because of “misdesignation of ... race”); see also Administrator’s Decision at A-137 (“Local 28’s submission of ... [a white union member] as a nonwhite was either the result of gross negligence or wanton disregard of this court’s orders,” warranting a finding of contempt). In sum, the union’s misreporting of census data, in violation of the AAAPO, has resulted in the inflation of its nonwhite union membership for each year from 1984-1991, and in the obfuscation of its abysmal lack of progress toward meeting the AAAPO’s 29.23% nonwhite membership goal. The defendant’s own expert, Dr. Buchanan, has confirmed that the union exaggerated its nonwhite membership for the years 1985, 1988 and 1989, and found that none of the union’s membership percentages from 1984 to 1991 submitted to the court was correct. The inaccuracies are not limited to the census reports but rather extend to other data submitted by the union to the plaintiffs. First, the racial identification of union members has been inconsistent among and within data sources. For example, joumeypersons were assigned different race-identifying codes in different files and were assigned different race-identifying codes in different weeks within a given file. Second, individuals who were identified in the hours files as employed were identified in the membership files as deceased, retired or suspended. Finally, some census lists could not be reconciled with membership lists covering the same period. For example, a computer file in December 1992, which was purported to be a comprehensive census list of members dating back to 1984, did not include members who appeared on previous membership lists. The union acknowledged the error and had to supply updated membership lists in May 1993. Discrepancies in March 1991 alone included individuals who were included in the census but listed as terminated in the membership files; active journeypersons included in the membership files but not in the census; and journeypersons included in the census but listed as inactive in the membership files. While the existence of a small number of errors would be understandable, the Second Circuit has concluded that numerous record-keeping mistakes constitute more than a de minimus violation of the affirmative action plan. Local 28, 753 F.2d at 1182. In the past, the sheer volume of mishaps and gross incompetence has led the court to be “convinced that ... [the union’s] violation ... reflected] defendants’ unwillingness to comply with ... [the affirmative action plan].” Id. This court is of a similar mind in the present case. Given the sum total of Local 28’s recordkeeping blunders, the union’s noncompliance is clear and convincing. The union has not been reasonably diligent and energetic in trying to meet its record-keeping responsibilities as defined by the O & J and the AAAPO. The union claims that white members were “inadvertently” omitted from membership lists and that others were deleted as a result of computer programming glitches. (Hilary B. Klein Decl. Concerning Union Record Keeping, Exh. C.) This type of excuse has been made unsuccessfully in the past. See Administrator’s Decision at A-133 (the Administrator “rejectfs] any argument that the ... [record keeping] problems were simply the result of human error”); see also Local 28, No. 71 Civ. 2877, slip op. (S.D.N.Y. Aug. 31, 1983) (Werker, J.) (affirming Administrator’s recommendations and May 18, 1983 conclusion that “[i]t is inconceivable that the officers of Local 28 did not understand the importance of accurate record keeping and that all submissions to the parties, the Administrator, and the court had to be as correct as possible”). The court does not take such “accidents” lightly and will not allow the union to keep using “oops” as a defense for noncomplianee. The union argues that “[g]iven the great number of [data] entries, clerical errors are not unusual.” (Wilton Aff. ¶ 10.) Certainly, a reasonable and random number of such mistakes is expected. However, where there is a pattern of undercounting whites and overcounting nonwhites every single year this argument becomes highly suspect. Furthermore, the union is responsible for creating and maintaining a system of checks and balances to ensure that its data is correct. Where the union has not set up a “formal system to verify the racial and ethnic composition of Local 28’s membership ... [and s]ueh verification that was done, was done on a totally haphazard basis,” this court has found contempt. Administrator’s Decision at A-133. Once again, “[t]he lack of any proper verification controls confirms ... that Local 28 has not acted in the affirmative manner contemplated by the court.” Id. Local 28’s defense that it has been responsive to inquiries by the plaintiffs and the Administrator about its records is equally unimpressive. If it were not for the union’s mismanagement of information to begin with, opposing counsel would not have had to constantly ask for clarification; Local 28 was simply fixing its own self-generated problems. The 0 & J and AAAPO do not permit the union to display a cooperative facade while simultaneously undermining the affirmative action program by providing inaccurate, shoddy, incomplete information to the court. The AAAPO specifically requires that the union take a proactive stand against discriminatory employment practices rather than adopt a passive, “band-aid,” reactive posture. The fact that the Administrator has not used the full force of his court-invested powers against the union does not warrant the union’s self-congratulation or its assumption that the Administrator approved of the union’s actions or lack thereof. The Administrator has indicated to the union on previous occasions that its recordkeeping was problematic, thus putting Local 28 on notice that it was in violation of the court’s orders. See Administrator’s Decision at A-130 (holding Local 28 in contempt where Administrator gave clear prior notice that names and racial data had. to be submitted because of a merger, yet union failed to provide this information). For example, upon discovery of a misdesignation of race error in 1990 the Administrator informed the union, “As you are aware, correct racial/ethnic identification of apprentices and members is critical to the validity of the record keeping system.... [Because this racial mis]identification has raised some question about the accuracy of the records, I am requesting that the ... union promptly review its records to determine their accuracy.” (Letter from Raff to D’Elia of 1/16/90, at 1.) The union acknowledged the problem, not surprisingly identifying it as a “computer clerical inadvertence,” (Letter from D’Elia to Raff of 1/26/90, at 1,) as well as other errors. (Letter from D’Elia to Raff of 2/2/90, at 3 admitting that “[b]y reason of [a computer conversion] oversight, the March 1989 census report was produced with inaccuracies” and conceding that it was “alerted to certain [other] factual inaccuracies” prior to the census report). Furthermore, the Administrator cannot independently verify certain data furnished to him; he can only recognize inconsistencies and discrepancies within the data. For example, the Administrator has no way of knowing by looking at the records whether a union member is white or nonwhite, is an apprentice or journeyperson, or is dead or alive. The fact that the Administrator did not take greater action against Local 28 may be attributable to his inherent inability to fully grasp the extent of Local 28’s violation, not to the union’s upstanding behavior. Moreover, “[i]t cannot be stated too strongly that the burden is not upon plaintiffs or the Administrator, but is upon defendants to act in a forthright and affirmative manner to comply with the letter and spirit of the court orders.” Administrator’s Decision at A-139. 2. Failure to Reach the Membership Goal a. The Census Figures In 1975, the court ordered the union to achieve 29% nonwhite membership by July 1, 1981. In 1976, in the RAAPO, the deadline was extended for one year. In 1983, in the AAAPO, the court adjusted the goal to 29.23% to take into account the fact that two unions had merged and as a result union members came from areas outside of New York City, and it extended the deadline to August 31, 1987. AAAPO ¶ 5. In 1987, pursuant to a stipulation between the parties, the court extended the deadline to August 31, 1992. The union’s own census figures show that it has once again fallen far short of the nonwhite membership goal. As of March 1992, the last census prior to the August deadline, the union reported that nonwhites constituted only 23.61% of its membership. One month after the deadline expired, nonwhite membership had reportedly decreased to 22.89%. The most recent data supplied to the court indicates that as of March 1993, nonwhite membership was 24.13%. In order to test the accuracy of the censuses that the union has submitted, both Dr. Siskin and Dr. Buchanan used the union’s membership files to create their own censuses. Dr. Siskin’s census covers every six-month period from September 1984 through March 1991, following the parameters set by the AAAPO, and for every period he studied Dr. Siskin found that the union’s census figures inflated the percentage of nonwhite members. See supra pp. 653-654. As already discussed, Dr. Buchanan attempts to destroy the credibility of Dr. Sis-kin’s data in several different ways that the court does not find convincing. See supra pp. 650-651. Additionally, Dr. Buchanan notes that pursuant to the AAAPO pensioners who have not worked during the three years prior to a census should be excluded, and he argues that Dr. Siskin could not have determined from the membership files who had worked during the three years prior to 1984, 1985 or 1986 because the files do not provide information for time periods prior to 1984. Dr. Siskin worked around this flaw in the data by assuming that people worked until they started receiving pensions and then excluding those pensioners who had been pensioned for three years or more. Dr. Buchanan also claims that people who had been suspended for over two months on the date of a census should not have been excluded because some of them might have been members during part of the six-month period prior to the census. A census is a snapshot of the people in the union on the date of the census, not a cumulative measure of everyone who was a member at any time during the six-month period prior to the census date, so it does not matter who was a member prior to the date of the census. In fact, Dr. Buchanan complied with the “snapshot” definition of a census when he performed his own census and excluded from it people who were suspended and not readmitted or who transferred out, withdrew or died before the census date, revealing that his criticism of Dr. Siskin for following the same practice is specious. Dr. Buchanan insists that the census he created shows that the union is even closer to reaching the membership goal than its own figures indicate, but his census contains a serious methodological error. Dr. Buchanan did not count anyone for whom the union did not assign an initiation date in its computer file of union members, resulting in the omission of an unacceptable proportion of union members. For example, he excluded 3,353 people from his June 1984 census, although the union only reported a total membership of 3,879 in September of that year. Dr. Buchanan’s explanation for excluding these individuals is that he was trying to exclude all members who were production and asbestos workers and thus were not covered by the AAAPO. Since there were no more than twenty-five such employees during the relevant period, see discussion supra p. 651, Dr. Buchanan unnecessarily excluded approximately 3,328 employees in June 1984. The people for whom initiation codes are not listed are long-term members who are disproportionately white, so the result is that Dr. Buchanan has managed to exclude a large portion of the union’s white employees, thus making the union’s percentage of nonwhite employees appear greater than it actually is. The plaintiffs have provided clear and convincing evidence that the union has failed to achieve the 29.23% membership goal, and although the union attacks the plaintiffs’ methodology, it does not dispute this conclusion. As of March 1991, the last census prior to the demerger, Dr. Siskin’s calculations show that the union’s nonwhite membership was only 18.89%, which was 10.34 percentage points below the membership goal. It is evident that as of that date the union had not made diligent efforts to comply with the court’s order, and the union has presented no evidence to show that it has taken steps since that time to speed the entry of nonwhites into the union. b. The Constitutional Argument The union revives its old argument that the membership goal violates Title VII and the Equal Protection Clause despite the fact that the goal has been upheld twice by the Second Circuit and once by a plurality of the Supreme Court. Specifically, the Second Circuit has found that the membership goal satisfies a two-prong test for the validity of a temporary, race-conscious affirmative action remedy. Local 28, 753 F.2d 1172. The first prong, which requires a “clear cut pattern of long-continued and egregious racial discrimination,” is satisfied because of the union’s “long continued and egregious racial discrimination” and because the union’s failure to meet the membership goal shows that the effects of the discrimination had not been eliminated. Id. at 1186 (cites omitted). The second prong, under which “the effect of reverse discrimination must not be ‘identifiable’, that is to say, concentrated upon a relatively small, ascertainable group of non-minority persons,” is satisfied because “the effects of the affirmative action remedies incorporated in the AAAPO will not unnecessarily trammel the rights of any readily ascertainable group of nonminority individuals.” Id. at 1187. The union’s persistent insistence on the uneonstituüonality of the membership goal in spite of repeated rulings affirming the goal reveals the union’s disrespect for the court and its orders. See EEOC v. Local 580, Int’l Ass’n of Bridge, Structural & Ornamental Ironworkers, No. 71 Civ. 2877, 1988 WL 131293, at *4 (S.D.N.Y. Dec. 16, 1988) (Carter, J.) (warning that “[cjounsel is put on notice that it will be dealt with sternly, even to the point of sanctions, for making arguments that have already been disposed of’). The union’s claim that the membership goal is unconstitutional and unenforceable is also surprising in light of the fact that the union stipulated to achieve the goal, leading the court to question whether the union signed the stipulation in good faith. The union does not assert that any of the factors underlying the Second Circuit and Supreme Court decisions upholding the membership goal have changed in a way that would require the court to revisit the issue of the constitutionality of the goal. On the contrary, almost twenty years after Judge Werker’s original decree in this case the union’s history of discrimination grows ever longer and more egregious and its continuing failure to meet the membership goal grows more distressing. The membership goal remains a necessary and permissible way to monitor the union’s progress in ending its discriminatory practices. c. Impossibility Defense The union asserts that it was not possible for it to achieve the membership goal by 1992, but it does not contend that circumstances have arisen that were unforeseen when the membership goal was set in the O & J in 1975, or when it was modified in 1976, 1983, and 1987 which now make compliance with the order impossible. See discussion supra pp. 649-650. The court notes that the union consented to the 1987 modification, which extended the deadline to August 31, 1992, and thus agreed that it would be able to meet the goal by that time. Nor has the union presented evidence that it has made diligent efforts to reach the membership goal. The union has not even moved the court to modify the goal based on the difficulty that it alleges that it has faced in meeting the goal. S.E.C. v. Musella, 818 F.Supp. 600, 602 (S.D.N.Y.1993) (Wood, J.) (finding lack of diligence in meeting duties under decree where defendant never requested modification or clarification). Consequently, it has no basis for raising an impossibility defense. Even if the union did have the right to raise an impossibility defense, however, its defense would fail. The union claims that the impossibility of the union’s meeting the membership goal is demonstrated by a study in which Dr. Buchanan calculated the rate at which the union could be expected to achieve the nonwhite membership goal given the 1984 levels of white and nonwhite journey-persons and apprentices, a yearly class of 100 apprentices entering the union, and 100 members leaving the union each year. Dr. Buchanan ran his analysis once using an average apprentice ratio of one nonwhite for every white and again using a ratio of three nonwhites for every white. He purports to show that based on the 3:1 ratio the union could achieve a nonwhite membership of only slightly over 22% by the end of 1991 and that based on a 1:1 ratio it could achieve just under a 21% ratio by that time. Although the union asserts that it would have been impossible for it to have achieved a higher nonwhite membership by 1991 than predicted by Dr. Buchanan’s models, the union in fact did achieve a higher nonwhite membership by that date, making it evident that the models are wildly inaccurate. As Dr. Buchanan notes, for every year from 1985 through 1990 the nonwhite membership of the union (as reported by the union) exceeded the maximum that Dr. Buchanan predicted that the union could achieve given a 3:1 nonwhite-to-white apprentice ratio. This result is particularly striking in light of the fact that the nonwhite to white apprentice ratio has never been as high as 3:1 and thus there have been fewer nonwhites entering the union than Dr. Buchanan assumed. Given that the model predicts that the result that actually occurred is impossible, the court finds that the union has not met its burden of production on this issue. d. Failure to Meet the Goal as Basis for a Finding of Contempt The union contends that the court may not base its contempt finding solely upon the union’s failure to reach the membership goal. The court need not reach this question, because there are numerous bases for the contempt finding, as the rest of the opinion makes abundantly clear. Furthermore, in interpreting the AAAPO courts have always used the membership goal as a “benchmark” to measure whether defendants are complying with the court’s orders. See Local 28, 478 U.S. at 478, 106 S.Ct. at 3051 (plurality op.). The court assumes that if the union and the JAC had not discriminated against nonwhites, the union would contain the same percentage of nonwhites as exists in the relevant labor force. The court views the union’s failure to meet the goal as a symptom of its refusal to stop discriminating against nonwhites in admission to the union and in the provision of equal job opportunities to journeypersons, and it is on this discrimination, not on the failure to meet the membership goal, that the contempt finding is based. It is important to recognize that the union’s failure' to meet the membership goal does not reveal the extent of the union’s failure to erase the effects of its past and ongoing discrimination because the census includes apprentices as well as journeyper-sons. Apprentices are predominantly nonwhite, so even if union membership as a whole were equal to the percentage of nonwhites in the relevant work force, journey-persons could still be disproportionately white. For example, even the union’s inflated figures show that in March 1991 the apprentices were 66.30% nonwhite, while jour-neypersons were only 16.44% nonwhite. The low proportion of nonwhite journeypersons is troubling because only journeypersons can take advantage of the employment opportunities which the AAAPO intended to guarantee. The apprenticeship program exists solely to create journeypersons, and people can be apprentices for only four years, during which time they earn between 30% and 70% of journeyperson wages. Consequently, the nonwhite membership percentage largely reflects the increasing number of nonwhites who spend four years in the apprenticeship program at a reduced wage, and it does not indicate that nonwhites are provided with the opportunity to have careers as sheet metal workers. The court notes that the membership goal is twenty years old and may no longer reflect the composition of the relevant labor pool. In June 1991 the northern counties of New Jersey were demerged from Local 28 and formed a separate local. The parties have not submitted information to the court regarding the number of union members affected and how this move will alter the racial composition of Local 28, but they agree that the membership goal must be recalculated. The 29.23% goal is based on the 1970 census, and the 1990 census indicates that the percentage of whites in the New York metropolitan area has decreased over the past decade. The union attempts to deflect attention from the fact that it has not attained the court-ordered membership goal by arguing that it cannot be held in contempt for failing to meet a membership goal that may no longer reflect the relevant labor pool. The union is indulging in mere speculation here, and its argument flies- in the face of the rule that parties are bound by court orders until the court modifies the order or releases them from it. The union does not argue that the order to achieve 29.23% nonwhite membership is somehow vague or unclear. Nor does the union contend that the changes in the larger population mean that the union’s nonwhite membership suddenly reflects that of the nonwhite population in the relevant labor pool and thus all vestiges of discrimination have been erased. On the contrary, it seems probable that the departure of largely white counties in New Jersey from the union and the increase in the minority population of New York City will mean that there is a greater disparity between the nonwhite membership of the union and that of the larger population than currently appears. The cases cited by the union in support of its contention that it cannot be held in contempt for failing to meet a membership goal that does not exactly reflect the relevant population are inapposite. This case concerns a contempt motion, in which the court focuses on whether the defendant has made a diligent effort to comply with the decree. It is the defendant’s compliance with the decree, not the appropriateness of the decree, that is at issue here. In contrast, Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977), and EEOC v. Local 14, Int’l Union of Operating Eng’rs, 553 F.2d 251 (2d Cir.1977), are Title VII cases in which precise measurement of the relevant population was necessary in order to determine whether the plaintiff had made out a prima facie case of discrimination. Acknowledging the difference between contempt and Title VII cases, the Second Circuit has ruled that the membership goal in this case need not reflect the relevant labor pool as precisely as in a Title VII case because “[i]n Hazelwood and in Local 14 ... the court was concerned with a statistical basis for a finding of discrimination. Here we have no such problem, discrimination having been established by direct evidence of longstanding practices.” Local 28, 565 F.2d at 36. In Mackin v. City of Boston, 969 F.2d 1273 (1st Cir.1992), cert. denied, — U.S. —, 113 S.Ct. 1043, 122 L.Ed.2d 352 (1993), which concerned a .motion to terminate a decree, it was also necessary for the court to be particularly precise in its measurements of the relevant population so that it could determine whether the purposes of the decree had been met. Furthermore, the language of the decree in Mackin is quite different from that in the AAAPO. The court in that case interpreted the language of the original order, which said that the decree would be terminated when Boston “achieve[d] a complement of minorities commensurate with the percentage of minorities within the community,” to mean that the decree would end when the percentage of minorities within the fire department equalled the current percentage of minorities in the community, not when the percentage of minorities within the department equalled the percentage of minorities in the community as of the date of entry of the decree. Mackin, 969 F.2d at 1276. In the AAAPO, on the other hand, “the [29.23] percent figure, as a numerical goal, is stated in absolute terms.” Patterson v. Newspaper & Mail Deliverers’ Union of N.Y. & Vicinity, 13 F.3d 33, 39 (2d Cir.1993), cert. denied, — U.S. —, 115 S.Ct. 58, 130 L.Ed.2d 16 (1994). 3. Disparity in Hours a. Plaintiffs’ Evidence Plaintiffs present their evidence regarding the disparity in hours worked between white and nonwhite journeypersons in the form of regression analyses, which are statistical tools commonly used by social scientists and increasingly accepted by the courts to determine the effect of a number of factors (“the independent variables”) on a particular phenomenon (“the dependent variable”). Ottaviani v. State Univ. of N.Y. at New Paltz, 875 F.2d 365, 366-7 (2d Cir.1989), cert. denied, 493 U.S. 1021, 110 S.Ct. 721, 107 L.Ed.2d 740 (1990). Dr. Siskin first developed a figure for nonwhite journeypersons’ “expected” hours worked by calculating the number of hours that they would have had to work for each year from 1984 to 1991 in order for their percentage of the total hours worked by union members to equal their percentage of membership in the union. He was then able to calculate the probability that the disparity would have occurred solely as the result of chance, which he expressed in terms of standard deviations. As the number of standard deviations increases, the probability that a disparity occurred by chance decreases. Id. at 371; D. Baldus & J. Cole, Statistical Proof of Discrimination 308 (1980). When a disparity is represented by two or more units of standard deviation, there is less than a five percent chance that the variation occurred randomly, and social scientists and courts commonly deem the disparity statistically significant. See Hazelwood Sch. Dist., 433 U.S. at 311 n. 17, 97 S.Ct. at 2743 n. 13 (“a fluctuation of more than two or three standard deviations would undercut the hypothesis that decisions were being made randomly with respect to race”); Coser v. Moore, 739 F.2d 746, 754 (2d Cir.1984). Dr. Siskin found, and Dr. Buchanan agrees, that the disparity between the actual and expected hours worked by white and nonwhite journeypersons was statistically significant for each year from 1984 through 1991. For every year from 1984 through 1991 the disparity was at least three standard deviations; in 1990 it was as high as 7.03 standard deviations; and for the entire period the average disparity was 14.83 standard deviations. The impact of the disparity on the lives of joumeypersons has been dramatic. During the period 1984 to 1991, nonwhite joumeypersons worked an average of 177 hours less than their white counterparts with the result that nonwhites earned an average of $23,332 each year, while whites earned an average of $28,192 each year. It is possible that the variation in hours between white and nonwhite joumeypersons can be accounted for by factors other than race, so the existence of a disparity alone does not provide evidence that impermissible discrimination has occurred. EEOC v. Joint Apprenticeship Comm. of Joint Indus. Bd. of Elec. Indus., 895 F.2d 86, 90 (2d Cir.1990). In order to ascertain the role that race played in creating the disparity, Dr. Siskin ran multiple regression analyses, in which the statistician identifies all “legitimate” (i.e. nondiscriminatory) factors that might have an effect on the dependent variable and which might create the observed disparity. Ottaviani, 875 F.2d at 367. The assumption is that once the effects of all possible nondiscriminatory factors have been accounted for, the remaining disparity can be attributed to the discrimination for which the statistician is testing — here, racial discrimination. Id. Dr. Siskin hypothesized that nonwhite workers may have less seniority or fewer skills than white workers, making them less desirable employees. Alternatively, nonwhites may live primarily in New York City, where there is less work. Dr. Siskin ran regression anal-yses controlling for seniority, special skills and residence, and he found that the disparity between expected and actual hours worked by nonwhite joumeypersons remained statistically significant for six of the eight years studied and that for the period 1984 to 1991 there remained a disparity of over nine units of standard deviation which can be attributed to the effects of racial discrimination. Another explanation for the discrepancy in hours could be that nonwhite joumeypersons choose to work fewer hours than white jour-neypersons. To test this hypothesis, Dr. Sis-kin assumed that anyone who did not work in a given year was not looking for work, and he excluded such people from his analyses. He found that a statistically significant disparity persisted for every year studied and for the entire period. To further ensure that a few individuals who were either not seeking work or who were unusually available for work did not skew the outcome, he ran the analysis excluding people with very high and very low hours. Once again he found that the disparity remained statistically significant for each year and throughout the entire period. There is no evidence that any nondiscriminatory factors other than those accounted for by Dr. Siskin created the disparity in hours worked by white and nonwhite journeypersons, see Sobel v. Yeshiva Univ., 839 F.2d 18, 34 (2d Cir.1988) (party challenging the validity of a multiple regression analysis on the basis that it omitted factors must identify those factors and demonstrate that they would weaken the showing made by the analysis), cert. denied, 490 U.S. 1105, 109 S.Ct. 3154, 104 L.Ed.2d 1018 (1989), and the