| DOL Home > OALJ > OFCCP > U.S. Dept. of Treasury v. Harris Trust & Savings Bank, 78-OFC-2 (ALJ Dec. 22, 1986) |
USDOL/OALJ Reporter
U.S. Dept. of Treasury v. Harris Trust & Savings Bank, 78-OFC-2
(ALJ Dec. 22, 1986)
U.S. Department of Labor
Office of Administrative Law Judges
1111 20th Street, N.W.
Washington, D.C. 20036
[Page 2]RECOMMENDED DECISION ON REMAND:
LIABILITY
I
PRELIMINARY STATEMENT
[Page 3]
[Page 4]II
[Page 5]III
A
[Page 6]B
[Page 7]
[Page 8]
[Page 9]IV
HARRIS' STATISTICAL REBUTTAL
[Page 10]
[Page 11]B.
[Page 12]
[Page 13]
[Page 14]Dr. Harry Roberts
[Page 15]
[Page 16]
[Page 17]V
[Page 18]
[Page 19]
[Page 20]B.
[Page 21]C.
[Page 22]
[Page 23]VI
[Page 24]A.
[Page 25]
[Page 26]
[Page 27]2. Initial Salary
[Page 28]
[Page 29]
[Page 30]3. The Preponderance of All Evidence
[Page 31]B.
[Page 32]
[Page 33]
[Page 34]C.
[Page 35]
[Page 36]D.
[Page 37]VII
[Page 39]VIII
[Page 40]
APPENDIX
STATISTICAL TECHNIQUES
[ENDNOTES]
Office of Administrative Law Judges
1111 20th Street, N.W.
Washington, D.C. 20036
[Date issued: December 22, 1986]
Case No. 78-OFCCP-2
In the Matter of
United States Department
of Labor,
Plaintiff,
v.
Harris Trust and Savings
Bank,
Defendant,
Women Employed,
Limited Intervenor.
Appearances:
- George R. Salem
- James D. Henry
- Debra A. Millenson
- Richard L. Gilman
- Diane A. Heim
- For the Plaintiff
- Theophil C. Kammholz
- John J. Cassidy, Jr.
- Thomas G. Abram
- For the Defendant
- Marcia D. Greenberger
- Suzanne E. Meeker
[Page 2]
- Peggy Hillman
- For the Limited Intervenor
Before: NAHUM LITT
Chief Administrative Law Judge
This proceeding arises under Executive Order 11246, as amended, and the regulations promulgated pursuant to its authority at 41 C.F.R. Chapter 60. The Order and regulations proscribe employment discrimination by Government contractors based on race, color, religion, sex or national origin. Section 202(a) of the Order also requires Federal contractors to take affirmative action to ensure that employees and job applicants are not discriminated against and are treated fairly with regard to all terms and conditions of employment.1
On December 7, 1977 the Department of the Treasury filed a complaint charging Harris Trust and Savings Bank of Chicago, Illinois ("Harris" or "the Bank") with violations of the Executive Order and the regulations2 and an administrative law judge was designated to preside over the case. The Complaint alleged that Harris, as a Federal contractor, had discriminated through its employment practices against affected classes of women and minorities. The Complaint further alleged that Harris had failed to (1) provide relief for members of the affected classes who continued to suffer the effects of discrimination and (2) to take the affirmative action directed by the Executive Order.
Extensive pretrial discovery was conducted by the parties. During the protracted proceeding a question arose concerning the discoverability of two statistical reports prepared for the Bank. The presiding judge, Judge Rhea M. Burrow, ordered the studies produced. Harris vigorously maintained that the studies were privileged and refused to comply with the order. When the case was heard in August and September of 1979, the Government's case included both statistical and testimonial evidence. However, as a sanction for non-compliance with his discovery order, the Bank was not allowed to present any affirmative statistical evidence in its defense.3 Consequently the Bank's case-in-chief was essentially limited to the testimony of its witnesses and supporting documentary exhibits.
The Recommended Decision and Order was issued on January 20, 1981. The Decision contained comprehensive findings of fact regarding the Bank's employment
[Page 3]
practices and concluded that, on the merits, the Government had established a prima facie case of race and sex discrimination which Harris had failed to rebut. On the merits and as a further sanction for failure to comply with his discovery ruling, Judge Burrow determined that Harris should be debarred - i.e., prohibited for a specified time from benefiting from any government contracts. Additionally, because of the substantive violations, he also determined that the Bank should be required to pay $12,167,515 in back pay to members of the affected classes4 and to award lost seniority and promotions.
All parties filed exceptions to the Recommended Decision. Harris' exceptions were the most extensive, setting out 363 specific exceptions and spanning 172 pages. In his Decision and Order, issued May 17, 1983, Secretary of Labor Raymond J. Donovan focused solely on the exclusion of Harris' affirmative statistical evidence and Harris' assertions that the reports were protected from discovery by the work product privilege and the Federal Rules of Civil Procedure. While finding that the two statistical studies which the Bank had refused to produce were not privileged under the work product doctrine, he determined that the reports were protected from discovery by Fed. R. Civ. P. 26(b)(4)(B) which insulates the work of an expert prepared in anticipation of litigation when the expert will not testify at trial. Thus, the Bank's refusal to provide discovery of the two reports was justified, the exclusion of evidence was inappropriate and the proceeding was remanded to allow Harris to present its statistical defense.
The parties engaged in additional discovery and procedural maneuvering after the case was remanded. During this period, the case was reassigned to the Chief Judge due to Judge Burrow's retirement. In response to various motions, an Order was issued on February 26, 1985 which delineated the scope of the hearing on remand. Based on the language of the Secretary's Order it was determined that the proceeding had been remanded solely for the very narrow purpose of allowing the Defendant to present its statistical evidence. The Order described the procedural posture of the case as follows:
- The status of the case is that the Government has rested; Harris attempted to rebut the Government's case but was not permitted to put on its case; Harris may now present its statistical case; and then the Government may attempt to rebut Harris' case. Feb. 26, 1985 Order, at 4.
It was also concluded that the earlier ruling allowing the Government to impose back pay liability retroactive to the period of the 1974 compliance review and other relief back to the effective date of the Executive Order was not disturbed by the Secretary's Decision.
The additional evidentiary hearing flowing from the Secretary's Order was held in November and December of 1985. Final oral arguments were made on May 29, 1986 and the case is ripe for disposition.5
[Page 4]
BURDEN OF PROOF
This case arises under Executive Order 11246 and the regulations promulgated pursuant to its authority. The Order and regulations prohibit discrimination by Government contractors against employees and applicants for employment. 41 C.F.R. §60-1.4(a). While not necessarily binding upon administrative tribunals in enforcement proceedings under the Order, cases interpreting Title VII of the Civil Rights Act of 1964, 42 U.S.C. §2000e et seq., do provide guidance in examining alleged violations.
The Government's Complaint charged Harris with engaging in various employment practices which discriminated against women and minorities and with failing to take affirmative action to eradicate the present effects of past discrimination. These allegations were subsequently refined to charge Harris with having affected classes of women and minorities in its employ and with discriminating against them in initial job placement, salaries and promotions (79 Tr. 125-31). This case, therefore, is analagous to a pattern or practice action prosecuted under Title VII and cases decided under that provision will be looked to for guidance where appropriate.
The gravamen of the Department's allegations is that Harris engaged in a pattern or practice of disparate treatment of women and minorities.6 In a disparate treatment action the plaintiff must first prove a prima facie case by establishing that intentional discrimination was the defendant's standard operating procedure. Int'l Bhd. of Teamsters v. U.S., 431 U.S. 324, 336 (1977). Proof of isolated or sporadic discriminatory acts is insufficient. Ibid. The quantum of proof necessary to make a prima facie showing of discrimination is evidence sufficient to raise the inference that employment decisions were predicated on an illegal criterion. Teamsters, 431 U.S. at. 358; Segar v. Smith, 738 F.2d 1249, 1267 (D.C. Cir. 1984), cert. denied, 105 S.Ct. 2357 (1985). In establishing a prima facie case, the plaintiff may use statistical or testimonial evidence, or a combination thereof. Trout v. Lehman, 702 F.2d 1094, 1101 (D.C. Cir. 1983), vac. on other grnds, 465 U.S. 1056 (1984); McKenzie v. Sawyer, 684 F.2d 62, 71 (D.C. Cir. 1982). When a prima facie case is thus established and the inference of discrimination raised, the defendant may then rebut the plantiff's evidence.
The defendant's burden on rebuttal is a burden of production. The defendant must come forward with enough proof to convince the trier of fact that the plaintiff has not proven discrimination by a preponderance of the evidence. However, the ultimate burden of persuasion remains on the plaintiff. Coates v. Johnson, 756 F.2d 524, 531-32 (7th Cir. 1985); Segar, 738 F.2d at 1268-69, 1284; Roman v. ESB, Inc., 550 F.2d 1343, 1350 (4th Cir. 1976). In a recent decision, the Supreme Court reaffirmed this formula as the correct legal standard. Bazemore v. Friday, 106 S. Ct. 3000 (1986). The Court described the defendant's rebuttal burden in a Title VII pattern or practice case as follows:
[Page 5]
- [O]ur decision in United Postal Service Board of Governors v. Aikens, 460 U.S. 711 (1983), although not decided in the context of a pattern and practice case, makes clear that if the defendants have not succeeded in having a case dismissed on the ground that plaintiffs have failed to establish a prima facie case, and have responded to the plaintiffs' proof by offering evidence of their own, the factfinder then must decide whether the plaintiffs have demonstrated a pattern or practice of discrimination by a preponderance of the evidence. This is because the only issue to be decided at that point is whether the plaintiffs have actually proved discrimination. 106 S.Ct. at 3008.
Of course, the defendant's rebuttal must be tailored to defeat the prima facie case.7 The crucial determination, however, remains whether the plaintiff has proved, by a preponderance of the evidence, that discrimination was the employer's standard practice.
THE INITIAL RECOMMENDED DECISION AND ORDER
This case is on remand for the limited purpose of allowing the Bank to finish presenting its case-in-chief by offering its statistical defense. The Government may then rebut Harris' statistical evidence. The scope of the Secretary's remand is narrow. Therefore, the procedural rulings found in the earlier Decision are unaffected unless specifically addressed herein or in the Secretary's Decision.8 Moreover, the earlier Findings of Fact and Conclusions of Law regarding the Plaintiff's successful proof of its prima facie case, based principally on the testimonial evidence of affected individuals, are also unaffected. This stage of the trial is, therefore, a rebuttal and surrebuttal of statistical evidence and this new evidence must be viewed in this context as a response and counterresponse to the Plaintiff's prima facie case. From a logical perspective it must be determined whether the statistical fencing of the parties has altered the earlier evidentiary findings since these findings were not disturbed by the Secretary. The ultimate standard of proof remains, however, whether the Department has proved its case by a preponderance of all evidence, and this requires an analysis of all the earlier findings in light of the statistical rebuttals.
Applicable Time Periods for Liability and Relief
In the initial Decision it was concluded that the Title VII statute of limitations does not apply in enforcement proceedings under Executive Order 11246 (BD 63).9 Thus the period of potential liability was found to extend back to the effective date of the Order: October 24, 1965 for minorities and October 14, 1968 as applied to women (BD 60-63; COL 2). However, it was concluded that the time periods for potential back pay relief was more restricted. It was found
[Page 6]
that, generally, an award of back pay can date back to the effective date of the Executive Order, excluding time periods covered by an approved compliance review (BD 84). Because Harris was covered by an approved compliance review until September 1974, potential back pay liability commenced at that point.
These rulings were not disturbed or affected by the Secretary's remand. They also describe an important distinction: because potential liability extends back to the effective date of the Order, evidence of discrimination during that time frame is probative on the ultimate issues in this proceeding. The earlier Decision only limits the scope of the back pay remedy. Therefore, the Plaintiff need not establish a continuing violation in order to introduce evidence of discrimination dating back to the effective dates of the Order.10
It should be noted that throughout this hearing on remand Harris has maintained that the language of footnote 63 to the earlier Decision concludes that it was not in violation of the Order between 1971 and 1974. This contention is erroneous. Taken in context, the footnote merely restates the Judge's conclusion that the back pay remedy is unavailable during that period because of the existence of an approved compliance review. To hold otherwise in the face of the specific Findings and Conclusions made regarding discriminatory practices during that time frame would be anomalous.
Plaintiff's Prima Facie Case
Before undertaking an examination of Harris' rebuttal evidence, it would be useful to summarize the Government's prima facie showing so that the Defendant's case can be placed in the proper perspective.
The evidence constituting the prima facie case was both statistical and testimonial. However the earlier Decision relied most heavily on the testimony of individual witnesses concerning their experiences at the Bank. It was explicitly concluded that, even without the report of the Plaintiff's econometrician, a prima facie case was established.11 (COL 43).
An important component of the Plaintiff's evidence was the testimony of five past and present Harris employees: Janet Shumacher, Martha Mull Page, Joyce Perry Arthur, Joan McIlroy and Frances Huritz. All five of the witnesses are women.12 The importance of the evidence offered by these women is shown by the fact that 96 Findings of Fact regarding their employment histories were made in the earlier Decision (FOF 69-164). It was found that the Bank had discriminated against them and that their experiences exemplified the types of barriers erected by Harris to hinder advancement of women (COL 35).
[Page 7]
Joan McIlroy testified that she was employed by the Bank from January 1974 until October 1977 (FOF 122; 79 Tr. 1211, 1216). She sought responsibility, made sales and received exemplary supervisory ratings (FOF 124, 127-33; 79 Tr. 1221-25, 1309-13). Although she was eventually promoted to officer, she was first required to acquire $25,000,000 worth of business for the Bank in one year (79 Tr. 1237). No other Harris employee had ever been required to achieve such a goal as a promotion criterion (FOF 136).
Janet Schumacher, Martha Mull Page and Joyce Perry Arthur were also hired after they completed college. All three were initially placed in jobs that were essentially clerical with no career path to officer. (FOF 69, 74-76, 92-94, 109; 79 Tr. 808-11, 834-35, 1364-68, 3313-3320). Although their career patterns and supervisory evaluations varied, in each case the evidence was found to show that a disparity in pay and promotions existed between the witnesses and similarly situated males.
Ms. Schumacher was hired in 1971. Despite very favorable supervisory ratings she was still not an officer in 1977. (FOF 8185; Pl, Ex. 75-82). In 1977 her job grade was lower than the nine male college graduates also hired in 1971. Her salary was lower than seven of the males' salaries. (FOF 91). Ms. Page was hired in 1970 immediately after completing college (FOF 92). Although she experienced performance problems in 1974, these were apparently corrected within a few months (FOF 97). Ms. Page was paid less than a white male in the same position, and she left Harris in 1977 to accept employment at another bank. She was an officer of that institution within three months. (FOF 105-107). Ms. Arthur was also hired by Harris shortly after graduating from college in 1974. Despite satisfactory "plus" performance ratings and indications of promotion potential, in 1977 Ms. Arthur's job grade and salary were lower than all white males with degrees hired in 1974. (FOF 109-11, 118-20).
Frances Huritz was employed by Harris from 1960 until 1975. Although she did not hold a college degree, she did take college level courses in accounting and economics. (FOF 140-41; 79 Tr. 1044). She received outstanding supervisory evaluations (FOF 144, 149, 158; 79 Tr. 1068-72, 1088) but was told that her promotion potential was diminished because she was perceived as a difficult person (FOF 152; 79 Tr. 1147-50). Ms. Huritz generated in excess of $2,000,000 in new deposits and $2,000,000 in new loans from 1969-1973 (FOF 145; 79 Tr. 1058-59). However she was dropped from a training program for failing a course (FOF 161; 79 Tr. 1094-1102). Ms. Huritz left Harris in 1975 and accepted a position as the manager and vice-president of another Illinois bank at a substantially higher salary (FOF 162-63; 79 Tr. 1044-46). Judge Burrow weighed the evidence concerning Ms. Huritz, including that evidence impacting negatively upon her credibility, and found that Harris discriminatorily blocked her from entering the upper echolons of management (FOF 164).
Mr. Calvin Stowell, manager of Harris' Administrative Employment Section, testified regarding the Bank's treatment of of minorities. His testimony was generally favorable toward Harris. However, he testified that sometime between 1975 and 1977 he was told by the head of Harris' Investment Department that he did not want minorities assigned to the Department because they might leave for better opportunities (79 Tr. 737-39, 743). Mr. Stowell
[Page 8]
admitted to meeting similar resistance from a division within the Investment Department (79 Tr. 743-45). The Bank's equity research research section and the Economic Research Department also resisted accepting minorities (FOF 21).
A myriad of other employment practices also displayed discrimination. Separate men's and women's training programs were maintained until at least 1971 (FOF 14), women college graduates were excluded from some of the Defendant's Departments until 1971 (FOF 15), and female college graduates were historically placed in secretarial and clerical positions (FOF 22). Thus, although statistical evidence is an important and hotly contested component of the government's case, absent any statistical evidence the record is replete with incident after incident of employment actices displaying discrimination. A prima facie case of discrimination has been shown without reliance on statistical evidence.
The Government also presented extensive statistical evidence at the first hearing, including both numerical comparisons and econometric studies. A comparison of the sex and race of the highest and lowest paid groups of Harris employees, for instance, showed that in 1974 there were 296 Bank employees earning between $21,000 and $150,000.13 This 296 included only 8 women and 6 minorities. (FOF 52-56). Comparisons of the highest and lowest paid employees at various educational levels were found to yield similar results. For example, as of March 31, 1977 the 22 lowest paid employees with high school diplomas were all females and minorities. In contrast the 29 highest paid employees holding only high school diplomas were all white males. (FOF 58). Plaintiff's other comparisions based on Personnel Data System (PDS) data (1) evaluated job grade placement of initial hires with college degrees from 1969-1974 (FOF 44-51), (2) looked at the percentage of female and minority employees earning over $15,000 per year in the Banking, Trust and Investment Departments (FOF 57), (3) compared the percentage of employees given the highest supervisory ratings by race and sex (FOF 59-60), (4) compared the number and percentage of employees promoted from non-officer to officer positions and the average time necessary to achieve officer status (FOF 61-64). and compared (5) the average salary of officers, managers and professionals by race and sex (FOF 65-68). The earlier record was found to show that white males dominate the Bank's heirarchy while women and minorities are concentrated in the lowest grades (COL 26). Even among employees with the same level of education, white males tend to obtain higher supervisory evaluations, become officers more quickly and earn a greater salary than minorities or females (COL 27-29).
The other component of the Plaintiff's statistical evidence was the report and testimony of its econometrician, Dr. Stephen Michelson.14 Dr. Michelson limited his study to full- time Harris employees. (79 Tr. 215). Most of his analysis was done based on the 1974 PDS tape (79 Tr. 219). However, the entire period 1974-77 was studied (79 Tr. 233). He determined that substantial disparities existed in salary between men and women and whites and minorities. (FOF 166, 173; Pl. Ex. 22 p. 32-37, Pl. Ex. 22A Charts V-1, V-2). The disparities in pay were examined in light of educational level and seniority. But even when the different levels of seniority and education were considered, Dr. Michelson's comparisons revealed unexplained
[Page 9]
differences in salary between white males and the protected groups. (FOF 167-170; 79 Tr. 233-249; Pl. Ex. 22A, Charts V-1 through V-20).
The technique of multiple regression was then used in an attempt to explain the disparities.15 By using multiple regression Dr. Michelson testified that he was able to calculate the average annual salary disparity in dollars based on race and sex for the period 1974-1977 (79 Tr. 273; P1. Ex. 22 Table 1). After adjusting for level of education, educational discipline, experience and seniority, Dr. ichelson's calculated differences in pay were found to show that the Bank had discriminated against women and minorities (FOF 171-175).
Based on this evidentiary showing it was concluded, as is concluded here, that the Government's initial statistical presentation is reliable. But, even if no reliance is placed on the Michelson report, as discussed, supra, the Plaintiff has proven a prima facie case of discrimination against women and minorities on the basis of direct testimony and documentary evidence (BD 80, COL 43).
The following portions of this Decision, Sections IV and V, discuss the statistical evidence presented at the 1985 hearing. These Sections summarize the testimony of the 6 expert witnesses and their reports. Its inclusion was necessary at this point to permit an orderly discussion of how the statistical rebuttals impacted upon the evidence already in the record. The 1985 evidence is revisited later and forms the statistical basis for the Findings and Conclusions. The length of this summary is unavoidable.
A.
Finis Welch
Harris's statistical rebuttal evidence principally consisted of the testimony and reports of Doctors Finis Welch, George Neumann and Harry Roberts. The first of the Defendant's expert witnesses to testify was Dr. Welch. He studied initial placement as measured by salary at hire and salary growth as measured by annual increases in salary. The first data base used by Dr. Welch was "interviewer files." These files are summaries of each employee's current status at the end of the year, beginning with 1972. The interviewer files include information such as job title, rate of pay and date of most recent raise. (85 Tr. 47). The second data base was the computerized records of all active employees as of March 31, 1977. This subset is referred to as the "Roberts Cohert" (95R-2 at 2). The last data source was the actual employee personnel folders. This final source was used to check anomalous observations and to compare with the information contained in the computerized records. (85 Tr. 48-49). The statistical techniques used by Dr. Welch to analyze the data were least squares
[Page 10]
regression and probit analyses (85 Tr. 53).
Dr. Welch began his study by analyzing salary growth from 1973 through 1977. This was done by comparing the beginning and end-of-year salaries for all full-time employees who were with the Bank at the start and at the finish of each year studied. (85 R-2 at 9). Dr. Welch estimated the differential in average annual percentage growth in salary between race/sex groups. He performed regressions adjusted for job grade, professional or clerical status, education, potential prior experience,16 experience at Harris, and time as an officer. Estimates of differentials in salary growth were computed for minority females, minority males and white females as compared to white males. For the five years studied, only two statistically significant differentials existed indicating that the average salary growth of a protected group was less than that of white males. (85 R-2 Table 4). Consequently, Dr. Welch testified that his analyses did not show significant differentials in salary growth between white males and the other groups (85 Tr. 75-77).
The analysis of starting salaries also began with an examination of the end-of-year data base. Dr. Welch reported that white males were generally more likely to have acquired experience useful to the Bank and more likely to have a business related education than women or minorities. (85R-2 Chap. 3). This conclusion was reached by performing a regression comparing the raw starting salaries for men, women and minorities. This regression revealed large dispariteis at statistically significant levels. (85R-2 Table 10). In order to explain the disparities Dr. Welch performed additional regressions controlling for level of education and educational discipline while holding potential prior experience at zero (85R-2 Table 11). When so adjusted, the comparison did not indicate any significant disparities adversely affecting women or minorities. Thus, Dr. Welch concluded that the observed disparities in starting salary were attributable to pre-Harris job experience and not indicative of discrimination.
Dr. Welch then analyzed starting salaries by experience level and initial placement into clerical and professional jobs. He characterized one subset to be studied as the "low experience group." This group included hirees who had been out of school less than one year and who had less than one year of potential prior experience. A total of 1,445 personnel hired by Harris between 1973 and March 31, 1977 were included in this group.17 (85R-2 at 40-41). Again, unadjusted regressions revealed numerically and statistically significant disparities in starting salaries for all protected groups. However, regressions adjusted for educational level and discipline, trainee status and potential prior experience reduced the differentials in starting salaries. The differential for minority males remained statistically significant at two percent below the average starting salary of white males. (85R-2 Table 15). Dr. Welch considered the two percent shortfall numerically small and characterized all of his estimates concerning the low experience group as imprecise due to omitted productivity variables and coding errors.
The next area examined by Dr. Welch was the placement of new hires into either clerical or professional jobs. He testified that a new employee's placement by Harris as either clerical or professional was largely a function of the job held prior to coming to the Bank.
[Page 11]
(85R-2 at 47-48; 85 Tr. 377-78). Dr. Welch then analyzed starting salaries of 312 of 2,396 employees hired between 1973 and 1977 who were not classified as low experience, who were initially placed in clerical positions and for whom prior job information was available. Unadjusted regression estimates indicated substantial starting salary shortfalls for experienced clerical minority females and minority males, but not for white females. Only the differential for minority females was statistically significant. When variables for educational level and discipline, type of prior experience, length of prior experience, trainee status, time elapsed since most advanced level of education and time out of the work force were included in the regressions, the salary differentials were greatly reduced. However, the shortfall for high experience minority females in clerical positions remained high at 5.94 percent and statistically significant. (85R-2 Tables 20 and 21).
Dr. Welch again discounted the reliability of his regressions. He opined that the lack of certain data such as quality of education, student performance, last job title before coming to Harris and "amount of experience in total" limited the usefulness of his data (85R-2 at 58). Additionally, because he felt 9 of the 312 observations were anomalous, they were deleted from the regressions after inspection of 8 of the personal files indicated that the computerized data was inaccurate. These deletions eliminated the statistical significance of the salary shortfall for high experience minority females placed initially in clerical positions. The last analysis performed Dr. Welch examined the starting salaries of persons hired into professional positions who were not studied with the low experience group. Of the 184 observations, only 18 had been in clerical jobs immediately prior to being hired by Harris (85R-2 Table 23). The unadjusted regressions indicated gross salary disparities for all three protected groups, but that the disparity for minority males was not statistically significant. The adjusted regressions greatly reduced the salary differentials and rendered all of the disparities statistically insignificant. (85R-2 Table 24 and 25).
Based on the information contained in his report, it was Dr. Welch's opinion that the data did not indicate systematic discrimination by Harris against women and minorities in either salary growth or initial placement (85 Tr. 207).
George R. Neumann
Harris' second econometrician was George R. Neumann, a professor of economics at the University of Iowa. He examined promotions to officer status using the end of year data tapes for 1972-79 as his data base. (85 Tr. 434). Dr. Neumann considered the important questions for his study to be the length of time to becoming an officer and the probability of becoming an officer. However, since not all employees are qualified or considered for promotion to officer, he felt that it was first necessary to define the pool of employees deemed eligible for officer status. Based on the year end data, Dr. Neumann found that in all but two cases promotions to officer came from employees in salary grades 14 or above.18 Therefore, he defined the eligible
[Page 12]
pool as all non-officer employees in salary grades 14 or higher (85 Tr. 441-42).
Because his data base began at year end 1972, Dr. Neumann was unable to determine exactly how long employees already had been in the pool prior to that year.19 Therefore, he decided to study how long it took eligible employees to become officers using three different measures of time. Dr. Neumann designated the first measure as "Y officer A" (YOFFA). This measure computed the number of months to officer status by subtracting the date hired from the date promoted to officer. (85 Tr. 446-47). Dr. Neumann performed regressions using this model for each year studied and for the entire period, pooling all officer promotions.
The yearly YOFFA regressions indicated large disparties in promotions when controlled only for sex and race. When the model was adjusted for educational level and discipline, performance evaluation ratings, future development ratings and department where the employees were assigned, the practical significance of the disparities was greatly reduced. However, as the table below shows, the disparities based on sex remained practically and statistically significant for 1973, 1974 and 1976. For example, this model shows that in 1976 it took females an average of 30 months longer than males to be promoted to an officer position. The disparity is conventionally considered to be statistically significant because the absolute value of the t-statistic is greater than 2.
Yearly YOFFA Regressions Adjusted for All Variables (85R 16) Race Coefficient T-Statistic Sex Coefficient T-Statistic 1973 -29.30375 -1.219 53.09218 2. 517 1974 2.541638 0.085 43.92708 3.744 1975 no minority promotions -6.582274 -0.473 to officer 1976 5.563698 0.291 30.49098 2.271 1977 -.2266171 -0.009 17.33508 1.136 1978 13.15293 0.841 .7653683 0.056 1979 -28.20889 -1.465 -6.451372 -0.478
Dr. Neumann declined to draw any inferences from these regressions because of the small sample size (85 Tr. 468-69). Consequently, he pooled the regressions for the entire period. The pooled YOFFA regressions that were fully adjusted did not yield any statistically significant disparities attributable to either race or sex. (85R-15).
The second measurement used by Dr. Neumann was designated YOFFB. This model measures the actual time spent in the eligible pool, but only for employees hired after Decemb er 31, 1972. (85 Tr. 473-75). Employees who were already in the eligible pool on that date were also included in the model. However, because the year end data tapes did not show when they were promoted into the pool, employees already in grade 14 or above on December 31, 1972 were credited with only one year of eligibility. (85 Tr. 495-96). Using this model, Dr.
[Page 13]
Neumann ran the same regressions adjusting for the same variables that were used in the YOFFA regression. The only statistically and numerically significant results yielded by the yearly regressions were for the sex variable. The regression indicated that in 1978 females were promoted to officer 10 months earlier than males and in 1979 they were promoted 13 months faster. (85R-18). The fully adjusted pooled regressions indicated that females were promoted approximately 6 months faster than their male counterparts. (85R-17). Thus, the YOFFB regression did not reflect any discrimination against a protected group.
The third series of regressions performed by Dr. Neumann used a model designated YOFFC. The YOFFC model measured the date employees achieved officer status minus the date hired. For employees hired before December 31, 1972 that date was used as a substitute for the date of hire. (85 Tr. 508). This model did not consider time in the eligible pool. None of the year-end regressions fully adjusted for department, performance and potential ratings, and education generated t-statistics greater than 2. Therefore no statistically significant differentials in time to officer based on race or sex were indicated using this model. (85R-19). The same result was reached using the pooled YOFFC regressions.
Dr. Neumann's regression analyses only examined the amount of time elapsed until officer status was achieved. They did not consider employees in the eligible pool who left Harris or simply did not become officers. Therefore Dr. Neumann studied the prob- ability of becoming an officer using probit analyses. The same eligible pool was used in these probits.20
A series of probits was run for each year with the number of variables increased in each run. The fully adjusted probit for each year controlled for much the same variables as the regressions: age, sex, level of education, business disciplines, years of employment, years in the eligible pool, department and supervisory evaluations. The sex and race coefficients as well as the respective t-statistics for each fully adjusted year end probit were consolidated in 85R-27A, reproduced below. The table reflects two negative sex coefficients: 1973 and 1977. Both negative coefficients indicate that during those years females were less likely than males to be promoted to officer. Both are also statistically significant. There are also two negative race coefficients with t-statistics greater than 2: 1974 and 1977. The race and sex coefficients for the remaining years are either positive, statistically insignificant, or both.
It was Dr. Neumann's opinion that his probit analyses supported the proposition that there was no sex discrimination in promotion during the period studied. The results of his probits with regard to race appeared to be in conflict with the results of his time to officer to regressions. Therefore he testified that there was "something" that needed to be looked at. (85 Tr573-75). He did not testify that there were no indications of race discrimination.
[Page 14]
Sex Race
Coef. T.Stat. Coef. T.Stat.
1973 -0.765 -2.751 -0.189 -0.536
1974 0.265 1.105 -1.422 -2.090
1975 0.451 2.064 - -
1976 0.081 0.388 -0.261 -0.922
1977 -0.475 -2.354 -0.628 -2.294
1978 0.447 2.431 0.182 0.784
1979 0.381 1.946 -0.038 -0.146
The final portion of Dr. Neumann's studies dealt with movement of employees into the eligible pool of job grades 14 and above. He divided Harris' job or salary grades into five Job Grades for purposes of study (85 Tr. 576).21 He then examined the flow of employees into his Job Grade 4, the eligible pool for promotion to officer. Dr. Neumann started by calculating the percentage of females and minorities in each job group for each year studied. By performing probit analyses he calculated an expected number of women and minorities that should be promoted into the eligible pool. The expected number of protected group promotions was then compared to the actual number. Dr. Neumann found no statistically significant disparity in the actual number of minorities promoted as compared to the expected number. However there was a statistically significant difference between the expected and actual number of women promoted into the eligible pool in 1978 (85 Tr. 576-585; 85R-28). Based on his analyses, Dr. Neumann was of the opinion that race and sex were not barriers for promotion into the pool of employees eligible for promotion to officer.
Harris' third expert witness was Dr. Harry Roberts. The report Dr. Roberts prepared for the 1979 hearing was excluded by Judge Burrow. It was that exclusion which ultimately led to the Secretary's remand. However, interestingly, his 1979 report was not relied on by Harris in this remand proceeding. Dr. Roberts decided that it was necessary to perform a new study because he wanted to collect additional data22 and to study this data using statistical methodology he considered appropriate at the time of the hearing on remand. (85 Tr. 753).
[Page 15]
coupled with his inability to divide the cohort into more precisely defined job categories distorted the results of his study in favor of an inference of discrimination.
The first area examined by Dr. Roberts was salary advancement during the 1973-77 time frame (85R-31A Chap. 3). Pre-1973 hires were treated in these regressions as if they had been hired on January 1, 1973. Then the data base was subdivided by the job groups previously discussed (clerical, professional and computer specialist) and by period when the employee was hired by the Bank.23 Explanatory variables included in the regressions were length of pre-Harris work experience, age, level of education and seniority. The report also states that variables controlling for prior professional jobs, prior employment as a computer specialist, prior financial experience and months out of the work force before employment at Harris "were tried and occasionally contributed significantly to the fit" (85R-31A Chap. 3 at 2). The salary advancement regressions yielded the following coefficients and t-statistics:24
White Female Minority Male Minority Female
Coefficient Coefficient Coefficient
Cohort Job Group WFIND t-statistics MMIND t-statistics MFIND t-statistics
pre-1964 Clerical 0.0229 No Data No Data
1.295
Professional 0.0655 No Data -.0329
2.800 -.340
60-64 Clerical 0.1207 No Data 0.1513
3.871 2.821
Professional 0.0708 0.0057
3.199 0.270 No Data
65-68 Clerical 0.0357 -.0394 0.0226
2.113 -.796 1.053
Professional -.0289 -.0394 0.1061
-.859 -.796 0.979
69-72 Clerical -.0042 -.0178 -.0158
-.254 -.861 .655
Professional 0.1026 0.0555 No Data
2.034 1.179
pre-1973 Computer -.0232 0.0006 -.0275
-.604 0.010 -.299
73-1977 Clerical -.0134 -.0202 -.0326
-2.426 -2.262 -5.069
Professional -.0183 -.0421 -.0464
-1.464 -2.336 -2.032-
Computer -.0304 0.0556 -.0317
-1.483 2.077 -.832
[Page 16]
The study states that the coefficients for each race/sex group may generally be compared by multiplying the coefficient by 100 in order to obtain the approximate percentage of comparison. Minus signs before coefficients indicate a lower than average rate of advancement (85R-31A Chap. 3 at 3). Thus the 1973-77 Clerical Minority Female Coefficient would be interpreted to mean that, after controlling for variables in the equation, on the average the salaries of minority females advanced approximately 3.2% more slowly than white males. Because the absolute value of the corresponding t-statistic (-5.069) is greater than the absolute value of 2, the coefficient is conventionally considered to be statistically significant (85 Tr. 764-65).
In his discussion of these results Dr. Roberts noted that 12 of the 30 t-statistics were greater than the absolute value of 2. He considered this to be an excessive variation and attributed it to either the failure of the data base to include necessary variables or random differences in "situations" that affected salary, but were unrelated to policy. Therefore, Dr. Roberts' report stated that the t-values and race/sex coefficients should be considered as a pattern and not individually. After performing the calculations necessary to aggregate the results of his regressions, Dr. Roberts interpreted his salary advancement study as showing parity between races and sexes in this area. Beyond the statistical nuances of his regressions, Dr. Roberts also felt that any disparities were practically insignificant.
The second area studied by Dr. Roberts was initial placement of employees hired after December 31, 1972. This was accomplished by examining two possible forms of discrimination: initial salary discrimination and discrimination in initial job placement.
Dr. Roberts began with initial salary. The estimated coefficients of the initial salary regression for the 1973-77 cohort are as follows:
Job Group White Females Minority Males Minority Females Clerical 0.0247 -0.0417 -0.0221 Professional -0.0840 -0.0797 -0.1604 Computer -0.0421 -0.0556 -0.0534
As the study states, 8 of the 9 coefficients are negative. They indicate initial salary disparities ranging from 2 to 16 percent to the disadvantage of the protected groups. Three of the negative coefficients are statistically significant. (85R-31A Chap. 4 at 6). Dr. Roberts attributed these results to job heterogeneity and not discrimination. To illustrate the reason for this belief he performed a regression without distinguishing between job groups. The resulting coefficients for all three protected groups were negative indicating salary shortfalls between 3 and 10 percent. All were statistically significant. Therefore the report concluded that job heterogeneity was responsible for any indicated disparities.
[Page 17]
The examination of initial placement was performed by shunting regressions. The report states these regressions were run using all variable job qualifications, instead of just education. This type of regression predicts salary assuming all employees are white males. All shunting coefficients yielded by the regressions were interpreted by Dr. Roberts to indicate that the three protected groups were less qualified than white males. However, as in the case of the initial salary disparities, the qualifications shortfalls were attributed at least in part to job heterogeneity. (85R-31A Chapter 4 at 9-11). Thus Dr. Roberts' report stated that the shunting studies were also difficult to reconcile with a finding of systematic discrimination. Dr. Roberts reported other factors that impacted upon the interpretation of his regressions. He was concerned that the data base did not include any information on actual job performance. To illustrate the potential impact of this omitted variable he reran the initial salary and shunting regressions for one job group for which prior salary data was available. Using prior salary as a substitute for job performance data, the regressions indicated reduced salary disparities while increasing the qualifications shortfalls of the protected groups. Thus, Dr. Roberts concluded that if this type of data had been available for all job groups, the indicia of discrimination would have been reduced.25 Moreover, he reported that when some job groups were disaggregated into smaller job groups by identifying trainees, the coefficients and t-statistics moved closer to zero further reducing the likelihood of discrimination.
Finally, Dr. Roberts studied the pre-1973 cohorts (85R-31B). He performed advancement and placement regressions for the 1965-72 time period. However he noted that because the quality of data for these earlier time frames was lower than for the 1973-77 cohort, it was necessary to exercise greater care in interpreting results. It was Dr. Roberts' opinion that the results of these regressions were generally in harmony with the results for the 1973-77 cohort. Salary and qualifications shortfalls for females and minorities were indicated by the placement regressions. Disparities in advancement of about 3 percent also existed in the 1969-72 cohorts. However, Dr. Roberts was again of the opinion the shortfalls reflected the effects of job heterogeneity as well as random variation. He did not believe that the statistical disparities supported an inference of discrimination.
THE DEPARTMENT OF LABOR'S
SURREBUTTAL
A.
Dr. Mark R. Killingsworth
The Plantiff's first witness was Dr. Mark Killingsworth. His reports and testimony were offered to rebut the statistical studies performed for Harris by Drs. Neumann and Welch (85 Tr. 1071). His testimony was initially directed at Dr. Welch.
Dr. Killingsworth testified first regarding Dr. Welch's studies of initial
[Page 18]
salaries. Dr. Killingsworth stated that one problem impacting upon the value of Welch's studies was his categorization of the prior employment of high experience employees as either professional or clerical. Dr. Killingsworth felt that many of the categorizations of specific employees were erroneous. He noted that all statistical analyses proceeding from the point of this categorization accepted it as correct. (85 Tr. 1073-74). Therefore, Dr. Killingsworth examined the relative qualifications of the people Dr. Welch categorized as either clerical or professional employees using the technique of probit analysis.
His probit equation included variables for race, sex, citizenship, level of education, years of work experience, time out of work force, and type of pre-Harris job. (85G-38 at 13). Dr. Welch had also studied clerical/professional placement using probits. However Dr. Killingsworth was of the opinion that his studies were more meaningful because of the manner in which he treated the variable for job held immediately prior to being hired by the Bank. (85 Tr. 1077-78). He felt that his prior employment variables were more accurate than Dr. Welch's. For example, Dr. Killingsworth included additional variables for prior jobs involving high level sales and professional technical experience. Dr. Welch did not. He also felt that Dr. Welch's data codings contained numerous inconsistencies.
Dr. Killingsworth performed five probits examining initial placement (85 Tr. 1081). The first probit indicated that white females, minority females, and minority males were all more likely to be initially placed in a clerical position than similarly qualified white males. However, only the t-statistic for the minority male coefficient was greater than 2. (85 Tr. 1082-85). Dr. Killingsworth's additional probits adjusted the female and minority variables by combining them in different ways. It was his opinion that the practical implication of his probits was that discernible differences existed between the way protected groups were initially placed, as compared to white males. (85 Tr. 1089-90).
The witness then examined Dr. Welch's initial salary regressions using the Roberts cohort data: all active employees hired between 1973 and March 31, 1977. Dr. Killingsworth studied initial salary of new clerical and professional employees separately, as Dr. Welch had done. He also studied the initial salaries of the two groups combined, in order to evaluate the effects of initial assignment on salary. (85 Tr. 1102-04). Dr. Killingsworth testified that he used the same prior job variables that he had used in his probits, and that his prior job categories were more specific than those used by Dr. Welch (85 Tr. 1107-09).
Dr. Killingsworth's combined clerical and professional starting salary regressions yielded the following results:
differential T-statistic inority male -0.107704 -2.759 inority female -0.200769 -6.010 White female -0.106032 -3.840
[Page 19]
(85G-38 at 3). The differentials were interpreted to indicate a statistically significant shortfall of approximately 20 percent for minority females. The minority male and white female differentials were both approximately 11 percent and both were also statistically significant (85 Tr. 1130-31). When the regressions were run separately for the clerical and professional categories, the only statistically significant shortfalls were for white female professionals and minority female clericals. (85 Tr. 1133-35; 85G-41).
When asked what conclusions could be drawn from the combined and separate regressions, Dr. Killingsworth testified that there were some indications of discrimination within the clerical and professional groups. He opined that the results of the combined study indicated that the large initial salary shortfalls for protected groups were due at least in part to initial job assignment. Thus he felt that the regression's were indicative of discrimination (85 Tr. 1135-37).
Dr. Killingsworth was also critical of Dr. Welch's salary growth studies because salary growth was conditioned upon initial job placement (85 Tr. 1157-60). Therefore, in rebuttal, Dr. Killingsworth conducted his own analyses of salary growth. His first analysis included an additional variable for salary at the beginning of the year as a measure of initial placement. The inclusion of that additional variable caused the regressions to yield results indicating salary growth disparities, some statistically significant, with an adverse impact on women and minorities (85 Tr. 1161-66). While Dr. Killingsworth was unwilling to draw much significance from this regression, he was of the opinion that if Dr. Welch's study was to be used as a predictor of salary growth, his additional variable should be used.
The second salary growth study also included variables that Dr. Welch had not used. Dr. Killingsworth added variables to account for the interaction of pre-Harris job experience with experience as the Bank, and with race/sex group. (85 Tr. 116869). The pooled regression results controlled for these additional variables showed disparities in rates of salary growth between 1.2 and 2.8 percent for members of the protected groups with no prior experience. All were statistically significant at conventional levels. (85 Tr. 1169-74; 85G-43 at 5). Dr. Killingsworth also testified that for employees with some Harris experience, female and minority salaries increased at a slower rate than for white males. However, he also testified that the disparity among experienced employee was not practically significant. (85 Tr. 1175-77).
Dr. Neumann's studies of promotion to officers were also attacked by Dr. Killingsworth. It was his opinion that Dr. Neumann's studies were too conditional upon the assumptions and structure used. Specifically, Dr. Killingsworth was critical of Dr. Neumann's analysis of movement into the designated eligible pool because it did not include all personnel actually eligible for promotion. (85 Tr. 1191-94). He testified that one subgroup of employees was not categorized by Harris's numbered salary grades, and therefore not studied by Dr. Neumann. However, members of this group were eligible for promotion to officer. Although Dr. Killingsworth noted that this omitted group was relatively small, he felt the results of Dr. Neumann's studies were somewhat distorted because of the omission. He performed an alternate study examining promotions between all of Dr. Neumann's designated job groups. Dr.
[Page 20]
Killingsworth calculated the number of expected promotions between all job groups for white females, minority males and minority females. His expected number of promotions for the protected race/sex groups was significantly larger in the higher salary grades than Dr. Newman's. (85G-44, 45). Dr. Killingsworth attributed the difference to his cumulation of the failures to promote over the years. (85 Tr. 1194-1205). He achieved this cumulation by using the entire workforce as the eligible pool. He testified that this allowed his studies to include the effects of failures to promote.
Logistic regressions (logits) were then performed to determine the probability of being promoted to officer (85 Tr. 1215-17). Dr. Killingsworth's logits indicated statistically significant disparities in the probabilities of becomming officers for women and minorities. (85G-46; 85 Tr. 1217-1220). Dr. Neumann's results using his eligible pool indicated less statistical significance.
Dr. Neumann's time to officer regressions were also criticized by Dr. Killingsworth. He declined to credit these studies with any probative value because he felt that the amount of time that it took to become an officer was simply not a good measure of Harris' promotion process (85 Tr. 1225-27). Dr. Killingsworth's principal criticism was that those regressions only examined employees who actually achieved officer status (85 Tr. 1229). Therefore those persons who never became officers were excluded from Dr. Neumann's study.
Dr. Franklin N. Fisher
Dr. Fisher, a professor of economics at the Massachusetts Institute of Technology, was the Department's second witness. His report compared the relative merits of reverse and direct regression analysis in the context of employment discrimination. (85G-50). The technique of reverse regression was developed by Dr. Roberts, one of Harris' experts, as a method for correcting measurement error in productivity variables (85 Tr. 1513).26 However, it was Dr. Fisher's opinion that reverse regression overadjusts for any measurement errors thereby underestimating the presence of discrimination.
The report stated that the underlying assumption of reverse regression was that a correlation exists between measurement error and measured productivity, but not between measurement error and true productivity. However, Dr. Fisher felt that this assumption was erroneous, and that in any event, reverse regression tended to underestimate the presence of discrimination by overestimating the effect of qualifications.
The omitted variable question was also addressed by Dr. Fisher. His report stated that omitted variables are a component of the error term in every regression equation. He considered the critical determination to be whether an omitted variable is correlated with any of the independent variables. (85G-50 at 20-24). If a complete correlation exists between the omitted
[Page 21]
variable and whatever measure of productivity is chosen for the equation, the omission does not have an effect on the race/sex coefficients. Dr. Fisher offered an illustration using years of education as a variable. If the omitted productivity variable was perfectly correlated with years of education, all of the omitted productivity information would be accounted for by the education variable. The omitted variable would have no effect on the race/sex coefficient in that case. Therefore, in determining whether the omission of a productivity variable has any effect, it is first necessary to conclude if, and to what extent, the omitted variable is correlated to included variables.
Dr. Bruce Levin
Dr. Bruce Levin also testified on behalf of the Department. He initially addressed the suitability of reverse regression as a tool for examining salary disparities and initial placement in this case. He stated in his report that reverse regression equations study and draw inferences from qualifications indices conditioned on salary, rather than conditioning salary on qualifications. (85G-52 at 3). Thus, salary is an independent variable in a reverse regression equation. Dr. Levin considered this an inappropriate method study because it measures qualifications, not unexplained disparities.
Regarding omitted variables and underadjustment bias, Dr. Levin testified that these phenomena are not present to the same degree, if at all, in every case. If an omitted variable affects both groups studied in the same manner, no bias occurs. Moreover, even if underadjustment bias is present, its magnitude depends on the extent to which qualifications variables or other measurements also account for any omitted variables. If other variables account for some or all of the omitted variable the bias is eliminated or reduced. He further testified that omitted variables can result in overadjustment bias instead of underadjustment bias. (85 Tr. 1576-79). Dr. Levin did not believe Harris' evidence indicated that if data on the quality of either education or work experience had been available it would have had an effect in this case.
The witness attacked Dr. Roberts' studies in several other areas. He testified that, among other things, because Dr. Roberts had only included employees still with Harris as of March 31, 1977 in his cohort, the sample was not random and left many employees unstudied. (85 Tr. 1613-14). Dr. Levin also characterized Dr. Roberts' shunting studies as a reverse regression examining initial placement. Therefore, the concerns he expressed about that technique applied to the shunting studies as well. (85 Tr. 1616). It was also his opinion that Dr. Roberts double adjusted his initial salary regressions and shunting studies by subtracting the mean difference in qualifications from the race/sex coefficients. (85 Tr. 1623-27). Dr. Levin testified that controlling regressions for job qualifications, such as education and experience, adjusts the race/sex coefficients once. Any additional "netting out" is an overcorrection.
The Department presented a series of exhibits in which Dr. Levin interpreted the results of Dr. Roberts' 1985 regressions using the latter's source computer output
[Page 22]
(85 Tr. 1629-38; 85G -54 through 85G-61A). Although he opined that Dr. Roberts' regressions should not be credited, Dr. Levin testified that his interpretation of Roberts' computer output showed that if the analyses were credible, discrimination was indicated. Several statistically significant disparities in placement and salary were found to exist for the protected race/sex groups from 1965-72. For the period 1973-77, Dr. Levin's interpretations of Dr. Roberts' regression output revealed significant disparities in salary and placement for women and minorities during several of the years studied. Dr. Levin was clearly of the opinion that the computer output of Harris' own expert indicated discrimination.
For his own study Dr. Levin employed "urn-model" regressions. He testifeid that he preferred the urn model to classical regression in some situations because regressions were initially developed to make estimates of unknown quantities where causal mechanism were well understood. A well-defined random sampling procedure is a key element of this type of study. In this case the cause of actual raw number disparities is the key determination. Therefore the issue to be resolved, causation, is not the type of question suited to analysis by classical regression. Moreover, the studies performed were entirely observational because data was obtained from the actual population under study. Due to several problems he perceived in applying classical regression to observational studies, Dr. Levin preferred urn model analysis. (85 Tr. 1595 - 1601).
The urn model is a form of direct regression which assumes that race and sex have no significance in making employment decisions. The technique divides each employee's salary into two parts: explained and unexplained. The explained component is obtained from an adjusted formula based on the qualifications required for the job in question. The difference between the explained component and actual salary is the residual, or unexplained component. (85 Tr. 1639-43). The average residual computed for each protected group is compared to the residual for white males, and then tested for statistical significance.
The first area studied by Dr. Levin using the urn model was initial salary.27 His analyses indicated statistically signifi- cant shortfalls for minority professional employees in the 197377 time period, minority computer specialists in the 1965-77 time periods, and minority clerical employees during the entire 1965-77 time periods. (85 Tr. 1669-72; 85G-65, 85G-65A). The effect of sex on initial salary was also examined. Dr. Levin testified that his results indicated statistically significant disparities between male and female professional employees from 1968-77. The output concerning computer specialists was not compelling, with no statistically significant disparities reflected. The only statistically significant disparities among clerical employees occurred in the 1968-72 time period. (85 Tr. 1675-79; 85G-66, 85G-66A).
Salary advancement was also examined by Dr. Levin. His urn model analysis indicated statistically significant shortfalls in salary advancement for minorities in several subgroups. These disparities were observed in the subgroups of minority professionals from 1973-77, and minority clericals from 1965-72 and from 197377. (85 Tr. 1679-85; 85G-69, 85G-69A). No significant disparities were found for computer specialists. Only the female clerical subgroups
[Page 23]
displayed significant shortfalls when the effect of sex on salary advancement was studied. (85 Tr. 1685-87; 85G-70, 85G71).
FINDINGS OF FACT AND CONCLUSION OF LAW
This remand proceeding was ordered so that Harris could present its statistical evidence. Despite the parties' rhetoric and mathematical jargon, the jousting over statistical analysis can be reduced to principles which are not as conceptually difficult as it might appear. Expressed in simplest terms the Plaintiff's goal was to include every negative observation in the data base, while Harris, on the other hand, wished to emasculate the data base as much as possible by removing any damaging data observations. Both sides also manipulated the selection of variables in the statistical equations in such a way as to benefit them
These maneuverings translated into the Government including any female or minority employee, no matter how marginally qualified, in its data base, and Harris excluding any employee, no matter how well qualified, from its studies. The Department also studied the Harris work force in the largest slices possible to increase the statistical significance of its studies. The Bank disaggregated its employees to the maximum extent possible in an attempt to destroy any significant statistical output. Both sides selected or omitted variables for their regression equations which would skew the results of the studies in their favor. It is in the nature of its exclusions, disaggregations and choice of variables, however, where Harris fails. The Bank's explanations of its manipulations have so many subjective elements that they undercut the value of the results.
At this point it would be appropriate to set out some of the practical considerations as well as legal principles of general application which have guided the evaluation of this statistical evidence. First and foremost, the use and limits of statistical evidence in adjudicative procedures must be clearly understood. Statistics may be used as circumstantial evidence of intentional discrimination. EEOC v. Fed. Reserve Bank of Richmond, 698 F.2d 633, 645 (4th Cir. 1983), rev'd on other ground sub nom, 467 U. S. 867 (1984). Statistics are not irrefutable. Their usefulness depends on the facts and circumstances of the individual case. Teamsters, 431 U.S. at 339-40. Furthermore, it must be remembered that, generally, what statistics actually do is generate an estimate of the likelihood that any race/sex disparities occurred by chance. EEOC v. Sears Roebuck & Co., 628 F. Supp. 1264 (N.D. Ill. 1986). That court noted that although the Supreme Court has not adopted any specific standard of statistical significance, statisticians conventionally consider statistics to be significant at two or three standard deviations. Rather than adopting a stringent criterion for statistical significance in this case, standard deviations (t-statistics) greater than the absolute value of 2 will be considered significant. However, depending on the assumptions, supporting facts and variables underpinning each analysis, that general rule may be adjusted if deemed appropriate.
[Page 24]
The parties continuously attacked each other's statistics as suffering from coding errors, mischaracterizations of employees, and incomplete data. Harris constantly contended that adjustment bias and omitted variables permeated the statistical evidence. As previously stated, the ommitted variable problem exists in every statistical model. Adjustment bias, either over or under, is also a constantly lurking demon. Where such deficiencies exist they can reduce or completely eliminate the probative value of a statistical study. Due to the frequency with which these issues arose, however, they have been explicitly addressed only when found to have significant impact on the statistics in question.
Harris also attacked some of the Government's regressions for lacking predicative power as reflected in R2 values. This statistic represents the amount of any indicated disparity accounted for by the variables, with 1.0 being perfect prediction. Harris appeared to argue that R2 values must reach a level of at least .9 before a regression equation is valid. However R2 values that high are an indication of tainted variables. D. Baldus and J. Cole, Statistical Proof of Discrimination §8.22 (1980 & Supp. 1985). The R2 values for the Department's statistics indicate acceptable levels of predicative power. In final analysis the statistical evidence, despite its shortcomings, has an appropriate role in resolution of this proceeding.
Initial Placement and Salary
1. Initial Placement
A large quantity of statistical evidence was presented at the 1985 hearing on the issue of initial placement and salary. The Government's evidence was more persuasive, and it establishes that Harris did discriminate against women and minorities in initial job placement and starting salary.
The parties agree that Harris' work force is bifurcated into clerical and professional employees. Approximately 60% of the Bank's employees are in clerical type positions, and, because of the high turnover rate among clericals, approximately 80% of all new employees are placed in non-professional jobs. (85 Tr. 14850). The essence of the initial placement issue is the determination whether to place a new hire in a clerical or a professional position. It is from this initial decision that all else proceeds.
Dr. Killingsworth, Plaintiff's witness, studied the probability of a newly hired employee being placed in a clerical job as opposed to a professional position. The data base used was the 496 personnel hired between 1973 and 1977 with more than one year of prior experience - Dr. Welch's high experience group. Dr. Killingsworth's probit, which included variables for minority males, minority females and white females, showed that all three protected groups were more likely to be placed in a clerical job than a similarly qualified white male. Although the estimates for the other two groups approached conventional statistical significance,
[Page 25]
only the minority female t-statistic was greater than 2. Dr. Killingsworth also combined the three protected group variables in one probit. The resulting combined protected group estimate showed that women and minorities were significantly more likely to be assigned to a clerical job than similarly situated White males. Dr. Killingsworth's opinion was that this combined study was the most meaningful and that discrimination was clearly indicated.
This placement study was vigorously, but unsuccessfully, attacked by Harris. Among other things, the Bank criticized the decision to study all high-experience hires together, rather than in clerical and professional subgroups, asserted that the choice of variables was inaccurate and inadequate, and argued that the number of employees actually involved was so small that the results bad no practical significance.
First, Dr. Killingsworth's decision to group the professional and clerical hires together in one study was clearly correct for this particular issue. The key determination is whether the distinction itself was based upon discriminatory criteria. It is bard to visualize how the question could have been properly examined without a simultaneous comparison of the employees directly affected. Regressing each subgroup individually would have assumed Harris' initial employment decisions were correct, and included this assumption in the probits. Dr. Killingsworth's choice of variables must also be sustained. As all parties readily admit, no set of variables perfectly accounts for all factors entering into an employment decision. However, his inclusion of variables for nine different levels of education, eight different categories of pre-Harris occupations', and several levels of prior experience represent a reasonable and accurate conforming of theory to the reality of an initial placement decision.28 Finally, the fact that a relatively small number of women and minorities were predicted by Dr. Killingworth to be placed in professional positions, but were not, does not negatively impact on his results. The subgroup of employees included in this particular data base was small: approximately 490 high experience new hires. Therefore the fact that the actual number of affected employees is also relatively small is no basis for attack.
Dr. Welch modified Dr. Killingsworth's probit analysis by deleting the 43 employees who were mispredicted by the probits. Harris maintains that these deletions remove any inference of discrimination from the probits. The majority of the 43 employees who were deleted by Dr. Welch in his replication of the probits were women or minorities. It is certainly understandable that their deletion would reduce the indicia of discrimination. However, exclusion of these observations from the probit would be erroneous. The Bank maintains that they should be deleted because Dr. Welch felt that their placement upon hire was consistent with their prior occupation reflected on the employment application. Dr. Killingsworth included eight pre-Harris occupation variables in his probit equation as well an prior experience variables. These variables adjusted the results of the probit for pre-Harris employment. Therefore excluding these persons from the data base would be an improper double correction.
Dr. Welch also calculated the probability of his high experience new hires being placed in clerical positions during the 1973-1977 time period. His report states that white
[Page 26]
males are generally more qualified for professional positions than women and minorities, and went on to state that the initial placement of an experienced employee by Harris was normally a function of prior employment. However, only cursory attention was payed by him to his probability calculations in his report. 'The report states that the estimated differentials from his probits on this issue were not readily interpretable, and were not reported. Although the race/sex coefficients were not listed, the standard deviation of the estimated differential (t-statistic) for each protected group, none of which was greater than two, were listed in the report. (85R-2 Table 18). The implication is that the coefficients were favorable to Harris but were not noted because they were not significant. But on cross-examination Dr. Welch conceded that ..the race/sex estimates did indicate that women and minorities were more-likely to be initially assigned to clerical jobs than professional positions, given the same coded qualifications. (85 Tr. 286). Because none of the differentials was statistically significant these probits are of little value. However the selective presentation of partially favorable results while unfavorable output was omitted casts further doubt on the value of Dr. Welch's probits.
The Bank argues that Dr. Robert's shunting regressions prove that Harris did not discriminate in initial job placement. Dr. Roberts noted that, on the average, women and minorities have less education than white males. Therefore, he maintained that any disparities are the result of differences in qualifications. The actual shunting regressions yielded coefficients that illustrated the relative qualifications of the protected groups as compared to white males. The Department avers that these shunting studies are reverse regressions and should be given little or no weight.
Dr. Roberts' report indicates in Chapter 6 that reverse regression was not used in his shunting study. However, language in other parts of the report reveals otherwise. As previously noted, reverse regression reverses the dependent variable (usually sex or race) and the independent variables (usually qualifications) in the regression equation. This is the technique incorporated by Dr. Roberts in his shunting studies. His report states:
- The placement shunting study uses the results of the placement salary studies of Section 2, which estimate the way the Bank appears to weight job qualifications in determination of salary. Each placement salary regression given an index, called FITADJ (for "fitted-adjusted") that weights all the individual qualifications variables used in the regression. The index shows how salary would be predicted under the assumption that all employees are white males. Therefore, qualifications are weighted by factors that serve to predict salary regardless of race or sex. (Emphasis in original).
(85R-31A, Chapter 4 at 9). Therefore the shunting study is in fact a reverse regression. The fundamental flaw in the rational of reverse regression requires that the technique be rejected. Dr. Roberts' report states, for example, that it "is better to compare qualifications for given salaries than to compare salaries for given qualifications" (85R-31A, Chap. 6 at 3). However this assumption is faulty. If there has been discrimination in initial placement or initial salary, in either case the independent variable (salary) is tainted. The use of so-called reverse regression, therefore, necessarily accepts Harris' determinations regarding initial placement or
[Page 27]
salary. The use of reverse regression methodology would conceal the existence of race or sex discrimination. Finkelstein, The Judicial Reception of Multiple Regression Studies in Race and Sex Discrimination Cases, 80 Columbia L. Rev. 737, 748-749 (1980). Conclusions flowing from use of this methodology can not be credited as a reliable tool in analyzing the issues in this proceeding.
The shunting study also has another flaw. The initial salary study accepts the Banks initial placement decision, which has a direct impact on initial salary, and then computes the FITADJ index. The shunting study then "uses" this index in estimation of initial placement. This is clearly an improper method of analysis. In a roundabout way the shunting study accepts the very decision it is supposed to examine. The Bank's initial salary determinations are also at issue in this proceeding. Their use as a basis to study initial placement is erroneous.
The question of discrimination in initial placement was also examined using starting salary as a measure of initial placement. Harris' rebuttal on this issue was provided by Drs. Welch and Roberts, and the Department's by Drs. Killingsworth and Levin.
Dr. Welch, Harris' first expert, limited his analysis to the 1973-March 31, 1977 time period. It was his opinion that the disparities in starting salary existed because white males were more likely to have the types of prior job experience and educational background for which Harris was willing to pay larger salaries for. Pre-Harris job experience was considered the most important of the two factors. Dr. Welch felt, however, that more could "be done" on the issue, and undertook a more detailed examination. He divided his data base into three distinct groups: low-experience hires, high-experience clerical hires, and high-experience profes- sional hires.29 The Government asserts that this division improperly stratifies the study thereby reducing the accuracy of the analysis. Therefore the Department would afford no weight to Dr. Welch's report. This position cannot prevail. The Bank's expert was certainly permitted to perform a reasonable stratification of the work force. Without some differentation in the data base the regressions would be unfairly weighted in favor of the Department. Although the stratification does raise some concerns expressed below, the three groups delineated by Dr. Welch are an acceptable division for purposes of that regression. However, the results of Dr. Welch's fully adjusted regressions do not rebut the prima facie case. Indeed his results are indicative of some race/sex discrimination.
Dr. Welch's low-experience group regression, fully adjusted for educational level and discipline and potential experience, indicated a statistically significant starting salary shortfall of 2% for minority males. He felt that a 2% disparity was not practically significant. However, even a 2% differential is important in real terms when it is considered that the shortfall is perpetuated and compounded in each subsequent salary adjustment. The high-experience clerical regression also indicated a statistically significant shortfall of 5.94% for minority females. Although statistically significant disparities existed for only these two groups out of the three
[Page 28]
regressions, their presence is still important. The differentials continued to exist even after the regression equations were fully adjusted to suit Dr. Welch, the Bank's expert. Moreover, Dr. Welch's regressions incorporated certain personnel determinations that were possibly tainted by Harris' employment practices. For example, the regressions accept Harris' decision whether a new hire was placed in a clerical or a professional position. While this alone does not completely discredit the regressions, it does affect their weight.
Dr. Roberts' initial salary regressions are even less helpful to Harris. Unlike Dr. Welch, he divided the Bank's new hires for the 1973-77 time frame into three subgroups: professional, clerical and computer specialist. Also unlike Dr. Welch, he studied the 1965-72 period. However, for this earlier time period he only used clerical and professional subgroups in the regressions.
The regression results contained in Chapter 4 of Dr. Roberts report strongly suggest the presence of systematic discrimination at the Bank from 1973 through March 1977. As previously discussed, Dr. Roberts' initial salary regressions showed shortfalls in initial salary ranging from 2 to 16 percent for women and minorities in every job group except white female clericals. The estimated disparities for minority male clericals, white female professionals and minority female professionals were statistically significant.30 Therefore Harris' own evidence lends support to the Government's case.
It is not surprising that the Bank argues that Dr. Roberts' report should be given little or no weight on this issue. His caveats regarding job heterogeneity, underadjustment bias, and omitted variables notwithstanding, Dr. Roberts' report was offered into evidence by the Bank. It cannot repudiate the output generated by its own witness.
In an attempt to achieve more homogeneous groups of employees Dr. Roberts further disaggregated the workforce by creating separate subgroups of general office trainees from the clerical group and professional trainees from the professional group. The resulting regression estimates generated for these subgroups indicated small initial salary disparities, none of which was statistically significant. Harris argues that these regressions illustrate the desirability of disaggregating the data base. This argument is not persuasive. It is generally accepted that disaggregation of the data base reduces the statistical significance of multiple regressions. See Baldus and Cole, Supra, §9.221. While the division of the data base into professional, clerical, and computer specialist subgroups is a reasonable fitting of statistical theory to reality, the further disaggregation is an attempt to unnecessarily divide the data base in order to dilute the significance of each of the regressions. Dr. Roberts' initial salary regressions were run in order to study all initial hires from 1973-77. The trainee regressions examined only the two trainee subgroups. The effect of this refinement on the remainder of the clerical and professional hires was not reported. In effect, Dr. Roberts disaggregated the clerical and professional groups into trainee and non-trainee subgroups, but only regressed the trainees. As an illustration of the problem of job group heterogeneity the trainee regressions only address a portion of the question. Therefore these results are afforded little probative value.
[Page 29]
The initial salary regressions performed for the periods 1965-68 and 1969-72 reveal dramatic disparities adversely affecting professional and clerical women and minorities. With the exception of white female professional hires from 1969-72, the results of the regressions were all statistically significant. Dr. Roberts attempted to minimize the value of this study because it was done hurriedly and because the "quality of data" was lower than for later time periods. The regressions not only fail to rebut the prima facie case, they support it, and his comments do not significantly diminish the impact of his regressions.
In his testimony on behalf of the Government, Dr. Killingsworth expressed the opinion that starting salary should be studied by combining newly hired professional and clerical employees in one regression. These combined regressions yielded large significant disparities for all three protected groups. It is recognized that the less stratified the data is, the likelihood of statistical significance is increased. This concern adversely affects the weight attributed to these regressions. However he also addressed the subgroups separately, as Harris' experts had done. His regressions, which considered the clerical and professional subgroups individually, yielded two statistically significant shortfalls in starting salary: white female professionals and minority female clericals. These regressions, which included the experience and education variables necessary for a proper examination of the issue, will be afforded substantial weight. Dr. Killingsworth's testimony clearly shows initial salary discrimination specifically against the two subgroups discussed above.
Dr. Levin's analysis of starting salary was performed for the Plaintiff using his urn model, which both parties agree is a modification of direct regression. There is nothing inherently unreliable or speculative in the technique, and indeed the asserted reasons for its use support its application here. Harris' objections regarding omitted variables and underadjustment bias notwithstanding, no attack was mounted against the urn methodology which would justify its rejection. Dr. Levin's results were evaluated and weighed in the same manner as the other output derived from least squares regressions. The caveats and concerns surrounding that form of evidence, including underadjustment bias, were considered and weighed with respect to the urn model, just as they were all other types of statistical evidence.
The evidence generated by Dr. Levin's urn model corroborates the earlier evidence establishing systematic discrimination. This output revealed statistically significant disparities in initial salary disadvantaging women and minorities. Dr. Levin examined the same subgroups and time periods studied by Dr. Roberts, using the same variables. (85R Tr. 1662). Starting with race, the revealed statistically significant shortfalls by subgroup and time period were as follows: 5% shortfall for minority professional hires from 1973-77, 3% disparity for minority clerical hires without finance company experience from 1973-77, 5% shortfall for minority clerical hires from 1965-72. (85G-65). No significant starting salary disparities were indicated for computer specialists. When the time periods were combined for each subgroup, discrimination against minorities was even more strongly indicated. Likewise, the urn model output studying disparities based on sex using the Roberts cohort and variables indicated significant starting salary disparities of almost 7% for female professionals from 1968-72 and female clericals from 1968-72.
[Page 30]
More than just an analysis of statistical evidence is necessary# however, in this proceeding. The ultimate standard of liability is a preponderance of all evidence.31 An evaluation of the totality of evidence establishes that Harris discriminated against women and minorities in initial placement and salaries. The combined weight of the statistical and anecdotal evidence in this case is overwhelming.
The testimony of the affected female employees is most persuasive. The case of Janet Schumacher is illustrative of the type of convincing evidence offered by the Government. Hired as a college graduate, she was placed in the College Training Program. Subsequently, she became an Assistant to Investment Account Supervisor, along with three other females with bachelor's degrees. The position was primarily clerical with no career path to officer, and was different treatment than that received by similarly qualified males. An officer of the Bank testified that through 1976 some women college graduates were hired into the secretarial program at the Bank without being advised of the existence of a training program for college graduates. Similar experiences were related by Ms. Page, Ms. Arthur and Ms. McIlroy. Harris attempted to justify this practice with testimony that this was done when the college graduate program was full. (79 Tr. 1946-51). However the picture that emerges shows this justification to be merely pretextual. The evidence establishes that women with degrees were hired as secretaries without being advised of the other program. Moreover, those women that did participate in the college graduate program were placed in less desirable positions. Harris' evidence to the contrary is not persuasive.32
Calvin Stowell, an Assistant Vice President, testified that between 1975 and 1977 when he was with the Bank's Administrative Employment Section he was told not to assign minorities to Harris' Investment Department (79 Tr. 737-39). Although the stated reason for this policy was that minorities were likely to leave for other jobs more quickly, the practice was clearly discriminatory. Ms. Sorenson also indicated that this exclusionary policy extended to women (79 Tr. 762-64). John Stephens, an Executive Vice President of Harris, also acknowledged that it was the general practice not to assign women to certain areas such as commercial lending and municipal bond sales. This practice continued unabated until the late 1960's or early 1970's. (79 Tr. 656-58). Race and sex discrimination were part of Harris' standard operating procedure. Rarely in a case of this nature, with management witnesses expected to show great deference to their employer, are such "smoking guns" evident. The testimony of Harris' own officials combined with the testimony of the affected individuals shows that Harris' entire employment was infused with discrimination.
Other historical evidence is corroborative. Separate training programs for men and women were maintained until at least 1971.33 Harris maintains there was no evidence of less favorable treatment as a result of the dual system. However the Plaintiff's statistical studies established initial salary and placement discrimination: a logical result of the dual system. The Government's statistical evidence has
[Page 31]
proven race and sex discrimination in initial placement and starting salary. It is true that none of the statistical studies indicate statistically significant disparities for all protected groups, at every level, for every year. However, every study is indicative of some discrimination. When these statistics are buttressed by the compelling testimony presented at the 1979 hearing, the preponderance of the evidence convincingly establishes that Harris discriminated against both women and minorities.
Salary Advancement
The evidence on the issue of salary growth also proves by a preponderance of the evidence that Harris treated women and minorities differently than white males. Again, Harris' own evidence supports this conclusion.
Dr. Roberts studied the issue using direct regression. He included variables for length of pre-Harris experience, age, level of education and seniority in order to mirror the salary adjustment decision as much as possible. His data base was all full-time employees as of arch 31, 1977. He treated any employee hired before January 1, 1973 as if he or she had been hired on that date. It should be noted that this arbitrary decision accepts all pay raise decisions occurring prior to that date as correct. Therefore the race/sex coefficients could be improperly deflated. With this caveat in mind, Dr. Roberts' regressions strongly indicate the presence of discrimination in the 1973-77 time frame. The regressions using the pre-1973 cohorts (persons hired be-fore 1973) reveal either negative coefficients that are not statis-tically significant, or coefficients that are significant indicating that the protected groups are treated more favorably than white males. Because of the possibility of erroneous deflation of estimates, little weight can be afforded these regressions.
The regressions studying employees hired after January 1, 1973 do not suffer from this inherent deficiency. As the Government points out, they offer a complete study of salary advancement in this cohort. The results of these regressions are quite different. Statistically significant disparities in salary growth ranging from one to four and one-half percent were revealed for every professional or clerical group studied, except white female professionals.34 Dr. Roberts asserted that the disparities were practically insignificant. However this position is simply wrong. The disparities in question were shortfalls in percentage increases in pay. Compounded over time, even a one or two percent shortfall in the rate of salary growth is highly significant. It was also argued that some small sample sizes for certain job groups rendered the output suspect. The facts speak for themselves: these regressions still strongly indicate race and sex discrimination. Indeed the most statistically significant (t-statistic of -5.069) indication of discrimination (3.26% shortfall) was generated from the second largest sample size (211 minority female clericals) in the 1973-77 cohort. Thus, Dr. Roberts regressions generated substantial evidence of discrimination during this time period.
Dr. Welch's study of the issue focused exclusively on the 197377 time
[Page 32]
period. His regressions were adjusted for job grade, professional or clerical status, education, potential prior experience, experience at Harris, and time as officer. Only two significant shortfalls were generated for protected groups. The weight of his regressions, however, is greatly reduced because many of the variables are endogeneous. They accept as correct and incorporate employment decisions made by the Bank, some of which are not at issue in this proceeding. Job grade, professional versus clerical status, and time as an officer are all completely controlled by Harris, and must be viewed, at least to an extent, as "tainted" or "suspect" variables. Consequently, Dr. Welch's regressions are of little help to Harris. The Government's witness, Dr. Killingsworth, attacked Dr. Welch's salary growth regressions because of the endogeneous variable problem. He performed his own regressions adding his own variables to compensate for the "problem". Although Dr. Killingsworth declined to draw any independent inferences from his own regressions, the presentation further highlights the extent to which Dr. Welch's salary growth regressions are tainted by use of subjective or nongermaine variables.
The evidence presented by Dr. Levin also indicates race and sex discrimination. Using Dr. Roberts' regression variables and Dr. Roberts' cohort, Dr. Levin's fully adjusted analysis yielded the following practically and statistically significant shortfalls in salary growth: 1.8 for minority clericals from 1973-77, 2.7% for minority clericals from 1965-72, 3% for minority professionals from 1973-77, and 1.1% for female clericals from 1973-77. Again, it must be noted that this is an action alleging discrimination against affected classes of women and minorities. While Dr. Levin's output does not reflect discrimination against every protected group at every job level, based on the above disparities, a strong inference can be drawn that the Bank engaged in a pattern of discrimination against women and minorities.
Dr. Roberts 1979 report, a portion of which dealt with "miniregressions" (85G-19b, chaps. 6 and 6A), was offerd into evidence by the Government.35 The miniregressions were traditional or direct regressions performed on highly stratified job groups of Harris' new hires. These employees were mostly employees hired into clerical jobs, the college training program, and professional positions. The independent variables included years of education, age, seniority, amount of prior experience, several previous job variables, accumulated absences, salary and certain statistical variables. Miniregressions were selected as the method of study in order to achieve the most homogeneous job groups possible. In chapter 6 of his report, Dr. Roberts compared the salary increases of various protected job groups to white males. He expressed his findings as a percentage of the white male growth rate, and uniformly found the protected race/sex groups were not discriminated against. Although most of his percentage comparisons were in the high ninety percent range indicating only a few percentage points of difference, the government argues that the actual output summaries set out in chapter 6A show that Dr. Roberts' results support its case. Its position is botttomed on the fact that disparities in salary growth as small as one percent may be practically significant because the effects are cummulative. Even accepting Dr. Roberts' expression of the shortfalls, the statistically significant disparities cited by the Government do weigh in favor of the Department's position.
[Page 33]
The report offered by Dr. Michelson at the 1979 hearing also indicates that the Bank discriminated against women and minorities. (Pl. Ex. 22). Harris quite correctly points out that he did not specifically separate the placement and advancement questions in his study or stratify the work force into job groups. While this objection reduces somewhat the probative value of the report, it does not destroy it. As shown below, Dr. Michelson's report and testimony do contribute to the conclusion that Harris discriminated in salary growth.
Dr. Michelson's regression equation provided for the variables he deemed necessary to study employee compensation: level of education, major school subject, time at Harris, and potential prior experience. Admittedly these variables do not represent the universe of factors that make up an individual employment decision because there are omitted variables absent from every regression analysis.36 Even though he did not actually interview Harris' managers, he testified that he had the benefit of the information obtained in discovery regarding the Bank's employment practices. The amount of such evidence is extensive. It cannot be said that his model bears no resemblance to reality and are not a reasonable solution to qualifying the Bank's employment decisions, as Harris asserts. The results of his compensation study strongly indicate a discriminatory system at the Bank, and while the weight given the evidence may be reduced as a refinement of the variables is made, his study still contributes to the Plaintiff's case.
Despite the amount of evidence admitted on the salary advancement issue, again it is the testimony of individuals involved that is not only persuasive, but adds substance to the discrimination described by some of Harris' officials. When combined with the statistical evidence its weight is overwhelming. Even without statistical evidence, the Government proved that Harris' employment decisions impacting on salary growth were discriminatory and that the Bank engaged in a pattern of salary discrimination.
Initially, it is obvious that the conclusions made regarding Harris' initial placement discrimination contributed to smaller pay raises for women and minorities. To the extent that initial placement determines initial salary, and all increases proceed from that point, the Bank discriminated against women and minorities in salary advancement. Even if Harris had not discriminated in the rate of salary increases, the initial placement discrimination resulted in some protected group members being unfairly disadvantaged in every pay check.
Nancy Sorenson, administrator of Harris' employment and staffing division, testified that until about 1971 college women were paid less than college men (79 Tr. 755). Ms. Sorenson testified that this practice did not continue after 1971. However her memorandum to Janet Schumacher's file renders that statement literally incredible. The memo, dated November 29, 1973, clearly stated that some college women hired in June of that year had been given "nominal" pay raises because they had been hired at a lower salary than men. (Pl. Ex. 71). Therefore as late as 1973, disparities in initial salary existed, and the disparities were carried forward with each pay raise because any percentage increase was based on an unjustifiably lower starting salary.
[Page 34]
Nor can the question of salary growth be addressed without reference to non-officer promotions, because promotions between salary grades are a component of salary advancement. The shortfalls in salary growth also indicate discrimination in this area. The testimony concerning the five affected women establishes that discrimination was Harris' normal practice.
Janet Schumacher's experience serves as a clear example of how the Bank's system operated to unfairly disadvantage the protected groups. She was kept in job grades 10 and 11 for approximately 5 years while a similarly educated and experienced male counterpart, David Sturdy, was awarded a grade 14 soon after completion of the training program. Although he and s. Schumacher had been in the same training program, by 1975 he was a grade 17 and became an officer in September of that year. At that time, Ms. Schumacher was still a grade 11. Throughout this hearing Harris has stressed the importance of a business-related college major, prior finance company experience, and advanced degrees in its employment decisions. Neither r. Sturdy nor Ms. Schumacher had these qualifications. They were clearly similarly qualified. The difference is that Ms. Schumacher was a victim of discrimination. The other witnesses cases are similar.
Ms. Arthur, a minority and a woman, possessed a college degree. She was hired in 1974 but was not informed about the college graduate program because it was filled. She was placed in a grade 10 clerical position formerly held by Ms. Schmacher.37 Ms. Arthur requested and got another position at a grade 11 the next year, and a nominal increase in pay. Her supervisor was aware that she should have received a larger salary increase, but this was never corrected. Moreover, Ms. Arthur was never told of the procedure for placing her name on the "Aspiration List" for promotion. When she finally became aware of the procedure she availed herself of it and in 1977 was placed in a grade 13 position with a career path to officer. Notwithstanding the Bank's witnesses to the contrary, the cases of Ms. Arthur and the other affected individuals establish that Harris did indeed discriminate in salary advancement.
Promotion to Officer
The evidence regarding this question proves that Harris discriminated against women.38 Dr. Neuman's time to officer regressions, the YOFFB and YOFFC model are so flawed that they have no probative value. The models arbitrarily assign either seniority or time in eligible pool values for employees hired before December 31, 1972 without reference to the actual date involved. This has the effect of slanting the data in favor of Harris so much that the regressions based on these models are without value. While not fatally flawed, the YOFFA model, which measured the months elapsed betwen date of hire and being promoted to officer, is entitled to little weight. Employees in the eligible pool who did not become officers or who left Harris were not considered by the
[Page 35]
model. Therefore the model implicitly assumes that all qualified persons in the eligible pool will be promoted to officer. This assumption is erroneous, especially in light of the victims' testimony. Even with these deficiencies, which tilted the studies in Harris' favor, the single year YOFFA regressions still showed significant discrimination against women in officer promotions for 3 different years. In another example of Harris aggregating the data when it suited its purposes, Dr. Neumann then pooled the yearly regressions. All disparities disappeared. However, the time to officer regressions lend very little support to the Bank.
The transition flow analysis is similarly flawed. Initially it should again be noted that this study also accepts the Bank's initial placement decision as correct. Furthermore, Dr. Neumann only examined the transition flow from his Job Grade 3 (Harris' salary grades 10-13) into the eligible pool, his Job Grade 4 (Harris' salary grades 14-21). The analysis does not even attempt to consider the possible existence of discrimination in the lower grades. Because the effects of any such discrimination would be carried forward, the failure to study the lower grades renders the transition flow analysis a mere partial picture entitled to little weight.
Dr. Neumann also performed probit studies to consider the probability of being promoted to officer. While his choice of variables was generally adequate, the variable accounting for the number of years in the eligible pool is problematic. The variable is endogenous and incorporates the effects of Harris' promotion and placement decisions. Clearly, the variable could conceal the existence of any disparities. Additionally, employees in the eligible pool prior to December 31, 1972 were arbitrarily considered to have entered the pool on that date. This has the effect of further slanting the probits in Harris' favor. In spite of these problems, the probits still indicated discrimination against women in 1973 and 1977 and minorities in 1974 and 1977.
The Government's statistics are also deficient. Dr. Killingsworth's use of all employees, as his eligible pool in calculating expected officer promotions is was too broad. Although some attempt to consider the cumulative effect of the Bank's failures to promote was justified, Dr. Killingsworth calculations were unreasonably slanted in favor of the government. In sum, the statistical evidence offered by both parties at the 1985 hearing on this issue is of minimal probative value.
The testimonial evidence, however, convincingly establishes that Harris discriminated against women in promotions to officer. While this pattern or practice proceeding has been divided into three main issues for ease of analysis and presentation, it must be remembered that it is Harris' entire system that has been attacked, defended and scrutinized. Each of the employment practices at issue impacts upon the others. In many ways their effects cannot be separated. Harris' initial placement and non-officer promotion discrimination are factors in its officer promotion system. They are a component of the pattern. Thus its initial placement and promotion discrimination are the beginning of its discrimination in promotions to officer. The continual placement of women such as Ms. Schumacher, Ms. Page and Ms. Arthur in dead end jobs with no route of advancement to officer was discrimantory. The practice was clearly an impediment placed squarely in the path of women that unfairly delayed or prevented promotions to officer. It is particularly noteworthy that two women, Ms. Huritz and Ms. Page, left Harris,
[Page 36]
took positions with other banks, and were made officers shortly after being hired.
The fact that Joan McIlroy was required to obtain $25,000,000 in new business as a precondition to becoming an officer, a demand made on no male employee, can only be described as discriminatory. This is especially true in light of her excellent supervisory ratings and the fact that no male employee was encumbered with such an onerous burden. The entire incident is a clear example of sex discrimination. This inference is heightened by the testimony of Robert Russell. He was a VicePresident at Harris and Ms. McIlroy's supervisor. He testified
that in 1976, even after she met the $25,000,000 criterion, the Bank's officials tried to convince her to direct her career path into an administrative area. Mr. Russell's supervisor, Mr. Gooch, was able to get Ms. McIlroy's male counterpart promoted easily. However, Mr. Gooch did not approve her promotion. When Mr. Russell complained about her treatment, Mr. Gooch told him to keep his mouth shut for the good of his career. (79 Tr. 1318-30).39 It is clear that Ms. McIlroy was discriminated against and when her supervisor attempted to secure the promotion promised to her he was confronted with threats. The annecdotal testimony establishes that Harris discriminated against women in its officer promotion.
With regard to minority officer promotions, the Government has not met its burden of persuasion. It was unable to produce one employee to testify regarding race discrimination or offer any testimony regarding discriminatory practices such as that provided by r. Stowell on the initial placement issue. Absent such testimony the Department has not proven discrimination against minorities in officer promotions.
Other Issues
All of the evidence of record, from both the 1979 and the 1985 hearings, was considered and weighed in reaching the conclusions discussed herein. As the evidence and legal arguments were considered, however, certain issues remained that require discussion at this point.
This proceeding dealt with the Bank's pattern and practices in dealing with its female and minority employees. Although the analysis was presented under three general headings, the issues are inextricably linked. Thus discrimination in initial salary or placement carries over into pay raises and promotions, and is compounded. Although no statistical output found significant disparities for every year studied, the persuasive results uniformly reflected some discrimination. When considered with the testimony of witnesses the picture that emerged showed a pattern of discrimination against women and minorities. It was Harris' system that was attacked and found flawed.
Considerable evidence was introduced regarding the Bank's Affirmative Action Program (AAP) and certain specific programs designed to aid women and minorities. An
[Page 37]
effective AAP is indeed the antithesis of a pattern or practice of discrimination. Coser v. Moore, 739 F.2d 746, 751 (2d Cir. 1984). However the evidence establishes that Harris' AAP was not effective. Testimony of past and present officers of the Bank establishes that protected groups were excluded from at least some divisions, that women college graduates were paid less than male college graduates and that a manager's career was threatened when he tried secure a promised promotion for a woman. The statistical presentations, raw numerical data and testimony of victims are corroborative. Harris' actions speak louder than its AAP.
Harris also maintained that a job preference study conducted by Dr. Thomas Ziemba indicated that many members of the affected classes were initially assigned in accordance with their wishes. The study, 85G-19C, was originally one component of Dr. Roberts' 1979 study. It was identified as evidence by the Department along with the rest of his 1979 report, but the Government withdrew its sponsorship of the document when Dr. Roberts testified that he did not write the preference study. (85 Tr. 995-97). The study was then moved into evidence by Harris.40 Due to several flaws, however, the study was of little value. Data was available only for the periods 1960-64 and 1975-77. Moreover, the data that was available for the 1975-77 time-frame was approximately 30% part-time employees who are not the subject of this proceeding. A substantial number of the workers studied either requested jobs that could not be classified, did not list a job preference, or applied for any position. (85 Tr. 1002-09; 85G-19C at 56). Therefore this crude preference study falls far short of the more extensive and precise surveys conducted by the employer in the Sears case. See Sears, 628 F. Supp. 1308-1312. The results Dr. Ziemba's study were given little weight.
Women Employed's Motion For Attorney's Fees
By motion filed October 4, 1985, Limited Intervenor, Women Employed (WE) requested an award of attorney's fees and costs in-curred by responding to the Motion of Harris for reconsideration of an Order dated February 26, 1985. The order denied a previous motion by Harris to limit the actionable period in the above-captioned case. The motion for reconsideration was denied by order dated August 9, 1985.
WE contends that Harris' motion for reconsideration was in contravention with Rule 11 of the Federal Rules of Civil Procedure.41 Rule 11 requires the imposition of appropriate sanctions against parties who file unwarranted pleadings or motions. It provides the following in pertinent part:
- The signature of an attorney or party constitutes a certificate by him that he has read
the pleading, motion, or other paper; that to the best of his knowledge, information, and belief
formed after reasonable injury it is well grounded in fact and is warranted by existing law or a
good faith argument for extension, modification, or reversal of existing law, and that it is not inter-
posed for any improper purpose, such as to harass or to cause unnecessary delay or needless
[Page 38]increase in the cost of litigation.
- * * * If a pleading, motion or other paper is signed in violation of this rule, the court, upon motion or upon its own initiative, shall import upon the person who signed it, a represented party, or both, an appropriate sanction, which may include an order to pay to the other party or parties the amount of the reasonable expenses incurred because of the filing of the pleading, motion, or other paper, including a reasonable attorneys' fee.
In response, Harris argues that its motion "in no way violated the letter or spirit of Rule 11; that it was "'not interposed for any improper purpose' ... [and] was a totally good faith attempt to obtain a clarification of the February 26, 1985 Order."
A party's subjective good faith no longer suffices to relieve it from responsibility for filing oppressive or frivolous motions.42 Since its amendment in 1983, Rule 11 imposes an affirmative duty upon attorneys to conduct a reasonable inquiry into the viability of motions before signing them. Davis v. Veslan Enterp., 765 F.2d 494 (5th Cir. 1985); Eastway Construction Corp. v. City of New York, 762 F.2d 243 (2d Cir. 1985). The United States Court of Appeals for the Second Circuit maintained in Eastway that "Rule 11 is clearly phrased as a directive. Accordingly, where strictures of the rules have been transgressed, it is incumbent upon the district court to gashion proper sanctions." Id. at 254, n.7.
Harris' tactics in this matter leave much to be desired. It attempted to relitigate part of an issue decided in the initial Decision and restated in the February 26, 1985 Order. It filed its motion for reconsideration and clarification five months after the initial order and only two working days before the prehearing conference. Indeed the timing of the motion by itself indicates that at least, in part, it may have been a harassing tactic.
However, WE's motion for attorney fees must be denied. Despite the Bank's questionable tactics, the August 9, 1985 Order did marginally clarify the actionable period. The February 26 order noted at page 5 that Harris appeared to argue that evidence concerning employment practices occurring more than two years prior to the 1974 compliance review could not be introduced. The Order then stated on page 6 that liability could be imposed back to the effective dates of the Executive Order, but that the back pay remedy would only date back to the date of the approved compliance review. Considerd in the context of the earlier 1981 Decision, the delineation of this justiciable period was considered apparent. However, the wording of the August 9, 1985 order was more explicit. It made absolutely clear that the actionable period extended back to the effective dates of Executive Order 11246. Given the contentious nature of the parties, it cannot fairly be concluded that Harris' motion was not warranted even if untimely and not free from an inference that relitigation of already decided issues may have been a motivating factor. Despite several justifying factors militating towards granting WE's motion, WE's motion for attorney fees will be denied.
[Page 39]
Remedy and Request for Interlocutory Decision
Pursuant to a stipulation between the parties this proceeding was bifurcated into a liability phase and a remedy phase. In its Post-Hearing brief the Department requested that the scope of any individual remedy be delineated in this Stage I Decision. Harris, on the other hand, argues that any remedy determination is premature and contrary to the terms of the stipulation.
The stipulation dated October 5, 1983 provides in pertinent part that:
- Pursuant to the established policy of the Solicitor of Labor, the Plaintiff and Defendant request that the Administrative Law Judge bifurcate the proceedings and that on remand the issues be limited to issues of liability and whether Defendant is in violation of Executive Order 11246, as amended, and to appropriate sanctions and relief other than individual relief.
This general language does not indicate specific agreement regarding the stage at which any remedy would be delineated. The question is further complicated by the Government's request that this Recommended Decision be deemed interlocutory and be certified to the Secretary for resolution prior to commencement of trial on specific remedies. The Department cites no regulation promulgated pursuant to the Executive order authorizing such a procedure. It argues that an interlocutory appeal of the liaibility issue is the most expeditious way to proceed and cites the Recommended Decision in OFCCP v. Honeywell, 77-OFCCP-3 (April 26, 1982), a case in which this procedure was used, as support. However, as the Government acknowledges, the liability determination in that case is still before the Secretary and there has been no ratification by the Secretary of this procedure. Many of the liability issues here were before the Secretary during the first appeal and they were not reached - an omission which argues strongly against granting this request here. While there might be some economy of resources, it is by no means certain and any benefits outweigh the need to move this proceeding forward. The request for premature certification of the liability issue will be denied.
By February 2, 1987, the parties will submit briefs detailing their positions on all procedural and substantive issues to be resolved in the remedy phase. At a minimum, the parties should address delineation of class43 and individual remedies, burdens of proof, proposed schedules for hearings, and additional discovery if any.
[Page 40]
complete, no Recommended Order will be issued..
NAHUM LITT
Chief Judge
Dated:
Washington, D.C.
Dec 22, 1986
STATISTICAL TECHNIQUES
1. Multiple Regression
The statistical technique used most often in this case is multiple regression. Simply put, in the employment discrimination context, it is a mathematical procedure which describes the affect of several factors on an employment decision.
Regression analysis begins with the hypothesis that there is a relationship between certain factors (independent variables) and an employment decision (the dependent variable). For example, it may be believed that race, years of education and years of prior experience have an affect on initial salary. Initial salary is the dependent variable becasue it is believed to be determined, at least in part, by the independent variables. Choice of variables is the first step in analysis. If the choice of variables is improper, the underlying assumptions of the model are, therefore, incorrect and the study is of little probative value.
The actual data for each employee studied is fed into the computer which is programmed to perform the computations. These computations yield numerical estimates, called coefficients, for each independent variable. Each coefficient is a numerical estimate of the affect of that independent variable on the dependent variable, all other independent variables being held constant. Thus, the race/sex coefficients potentially have the ability to provide an estimate of the impact of race and sex on the employment decisions at issue.
2. Statistical Significance or "T" Factor
The estimation of the affect of a variable is one component of regression analysis. The evidentiary inference to be drawn, if any, is another. Assuming the choice of variables is adequate, the statistical significance of the estimates must be determined. This is accomplished by reference to the "t-statistic." In this proceeding, t-statistics greater than the absolute value of 2 (either greater than 2 or less than -2) were deemed significant. What this actually means is that the probability of a particular coefficient (estimate) occurring by chance is less than 5 percent.
3. "Fit" or "Fittage"; "R2" statistics
Another statistical test used in evaluating regressions is the R2 statistic. It is an overall indication of the extent to which any indicated disparities are explained by the independent variables. It is said. that R2 measures the Of it" of a regression. However the test is Very problematic. A high R2 does not necessarily mean that the regression fits the data well. The R2 is almost always increased by including more variables, even if the variables are unnecessary. Vuyanich v. Republic Nat. Bank of Dallas, 505 F.Supp. 224, 273 (N.D. Tex. 1980), vac. on other grounds, 723 F.2d 1195 (5th Cir.), cert. denied, 105 S.Ct. 567 (1984). Moreover, a high R2 may indicate tainted or employer controlled variables and cannot be blindly accepted.
4. Logistic (Logit) and Probability (Probit) Analyses
Logits and Probits are very similar techniques for studying the likelihood of an event occurring. For example, in this case the techniques were sometimes used for studying the probability of individuals being promoted. In both techniques the dependent variable is assigned one of two possible values: 0 or I. Those persons who were promoted are accounted for by the value 1. Those who were not are accounted for by 0. Otherwise, the actual computation is like multiple regression.
The primary difference between the two techniques is the form of the equations. A Probit, like a regression, is linear, while a logit equation is in the form of a fraction. In either case, the equations are designed to assess the relationship between the probability of an event (promotion) occurring and the independent explanatory variables. As in regression, the variables represent an attempt to quantify the factors which are thought to influence the event being studied. Therefore selection of the proper variables is just as important and serves the same purpose as in multiple regression. The function of the t-statistic is the same as in regression, and the test for significance is also the same: the t-statistic should normally be greater than the absolute value of 2.
1The following abbreviations will be used herein. 79 Tr.__ and 83 Tr. __ will refer to pages from the transcripts of the 1979 and 1985 hearings, respectively. The Plaintiff's and Defendant's Exhibits from the 1985 hearing will be indicated by the abbreviations 85 G__ and 85 R__ , respectively. Exhibits from the 1979 hearing will be referred to as Pl. Ex. __ and Df. Ex.__.
2At the time the Complaint was filed responsibility for enforcement of the Executive Order upon banks and financial institutions rested with the Department of :he Treasury. This responsibility was transferred to the Department of Labor in 1978, and the Secretary substituted the Department of Labor as Plaintiff in his Decision and Order.
3Harris' statisticians were allowed to testify for the limmited purpose of impeaching the Government's expert witnesses (79 Tr. 2791-92, 2891-92).
4The period of potential back pay liability was found to extend back to May 1972. However, on the facts of the case, it was recommended that entitlement to back pay commence at the beginning of the period covered by the 1974 compliance review.
5Throughout the course of this remand proceeding the parties continuously requested extensive postponement of procedural dates, such as filing deadlines, briefing schedules and hearing dates. In the face of these requests, normally agreed to by all parties, there appeared to be no public interest justification for requiring a more expeditious resolution of this case. Although much time was devoted to settlement negotiations, the fact that this remand has taken in excess of three years to conclude is largely due to delays requested by the parties.
6The Government also alleged that Harris' performance evaluation system was discriminatory under the disparate impact theory. In order to prove discrimination under this theory the OFCCP must show that some facially neutral employment practice has a disproportionately adverse impact on minorities, and that the practice is not significantly related to successful job performance. Griggs v. Duke Power Co., 401 U.S. 424 (1971).
7The employer can rebut a prima facie case of discrimination by showing that the alleged disparity in treatment or impact does not exist. Normally, this involves an attempt by the defendant to show that the plaintiff's statistical evidence is inaccurate or insignificant. The defendant can also rebut the plaintiff's case by showing that any disparity is not the result of intentional discrimination. This defense attacks the inference of intentional discrimination drawn from the plaintiff's evidence by offering a legitimate, nondiscriminatory explanation for any disparity. Segar, 738 F.2d at 1268-69.
8Prior to analyzing any evidence, Judge Burrow addressed the Bank's affirmative defenses. He rejected Harris' contention that an award of back pay was unauthorized under the Order and, in the alternative, unconstitutional. The argument that the Secretary's regulations were outside the scope of the Order's authority was also rejected. It was noted that jurisdiction to resolve these questions was reserved for the Judicial branch. These rulings are not within the scope of the remand and are not affected by this Decision on remand.
9 The numbered Findings of Fact and Conclusions of Law in the January 30, 1981 Recommended Decision and Order will be referred to as FOF__ and COL__ , respectively. References to specific pages of Judge Burrow's Decision will be cited as BD__ .
10The continuing violations theory allows alleged victims of discrimination to avoid statutes of limitations by showing that systematic discrimination operated against them within the limitations period, even if some or all of the events giving rise to the discrimination are time barred. United Air Lines, Inc. v. Evans, 431 U.S. 553, 558 (1977); Williams v. Owens-Illinois, Inc., 665 F.2d 918, 924 (9th Cir.) cert. denied, 459 U.S. 971 (1982). In pattern or practice cases plaintiffs must show the maintenance of a discriminatory system both before and during the limitations period. McKenzie, 684 F.2d at 72.
In the instant case, the Plaintiff charged the Bank with having affected classes of women and minorities in its employ, encompassing systematic discrimination in initial placement, salaries and promotions. Out of an abundance of caution and a desire to see this proceeding concluded, the evidence will be examined, infra, in order to determine whether it is sufficient to establish a continuing violation, had it been necessary for the Department to do so.
11In reaching that conclusion the Judge had an opportunity to observe the demeanor of the witnesses. Consequently, even if a total reexamination of this evidence were required as Harris suggests, the earlier credibility determinations made at the 1979 hearing would, perforce, have to be afforded great deference in this remand proceeding.
12Ms. Arthur is also black. However she refused to attribute any discrimination inflicted upon her to race.
13The raw data for these comparisons was taken from the Personnel Data Systems (PDS) computer tape provided by Harris.
14Pl. Ex. 22
15A brief explanation of the statistical techniques used may be found in the Appendix.
16Potential prior experience is the amount of time elapsed between The completion of schooling and date of hire at Harris (85 Tr. 74). It was used as a substitute for data on relevance, quality and continuity of job experience (85R-2 at 18).
17This represents 60.3 percent of the 2,396 personnel hired during that period.
18There are 21 salary grades, in addition to officer positions at Harris (BD at 16).
19Dr. Neuman testified that he was unable to reconstruct the eligible pool prior to 1972 because the earlier computer records were unreadable and there was no "good way" of obtaining the information from the actual personnel files (85 Tr. 443).
20Evaluation of the results of probit analyses is similar to the evaluation of regressions (85 Tr. 518). T-statistics greater than 2 are conventionally viewed as significant.
21The job grades used at Harris correspond to Dr. Neumann's Job Grades as follows:
Harris job/salary grades Dr. Neuman's Job Grades salary grades 1-5 Group 1 salary grades 6-10 Group 2 salary grades 10-13 Group 3 salary grades 14-21 Group 4
22Dr. Roberts supplemented the data base by conducting a limited sampling of employees hired between 1964 and 1977, but termi- nated prior to March 31, 1977. He determined from the limited sample that a more extensive study of terminees was unnecessary. (85R-31A Chap. 2).
23The periods of hire were divided into five entering cohorts:
Cohort 1 Pre-1960
Cohort 2 1960-64
Cohort 3 1965-68
Cohort 4 1969-72
Cohort 5 1973-77
24These results are obtained from Tables 1 and 2 to Chapter 3.
25However on cross-examination Dr. Roberts testified that he could not predict whether inclusion of the omitted variables would cause his results to change in one direction or the other. He conceded that all data bases suffer from some omitted variables of varying potential importance. (85 Tr. 893).
26The technique of reverse regression alters the standard regression equation by making productivity the dependent variable. The typical dependent variable, such as salary, appears in the equation as an independent variable. This reversal of variables gives rise to the term "reverse regression." (85G-50 at 12-13).
27Computer output generated from the urn model is interpreted in much the same way as standard regression output. The data column labeled "Delta R Bar" is the average difference in the unexplained component of salary for the protected group as compared to white males. Although it does have a slightly different meaning, the z-statistic is interpreted in the same manner as a t-statistic. (85 Tr. 1667). Z statistics greater than 2 were considered statistically significant by Dr. Levin.
28Harris also attacked Dr. Killingsworth's use of census occupation codes in categorizing prior employment. No categori- zation is perfectly accurate. However, use of these occupation codes was a reasonable and credible means of organizing the prior job classification.
29Low-experience hires are newly hired employees with no more than one year of potential prior experience. High-experience clericals are those hired after 1972 who are not low-experience hires, but who were not placed in professional positions. The high-experience professional hires are those who are not low- experience, but were placed in professional positions.
30The t-statistic for white female clericals was also greater than 2, but the corresponding coefficient did not indicate a salary shortfall.
31The question of proving a prima facie case is not at issue in this hearing because it is outside the scope of the remand. If it were, however, the same conclusion reached by Judge Burrow would result. Indeed, without any statistical evidence, it would still be found that a prima facie case of discrimination had been proved based solely on anecdotal evidence.
32Harris' witnesses included officers such as John Stephens, Nancy Sorenson and Melody Camp. They essentially testified that the Bank's employment practices were not discriminatory. These witnesses were also offered as specific proof that women and minorities could and did advance to officer status with Harris. However, the fact that some specific individuals were able to overcome the obstacles errected by Harris does not alter the conclusions reached herein.
33It was previously noted that the Department of Labor is not required to establish a continuing violation in this proceeding. However, had it been necessary the Government would have succeeded. With respect to Title VII cases, the Supreme Court has recently stated:
- A pattern or practice that would have constituted a violation of Title VII, but for the fact that the statute had not yet become effective, became a violation upon Title VII's effective date, and to the extent an employer continued to engage in that act or practice, he is liable under that statute. While recovery may not be permitted for pre-1972 acts of discrimination, to the extent that this discrimination was perpetuated after 1972, liability may be imposed. Bazemore v. Friday, 106 S.Ct. at 3006
Plaintiff has proven systematic discrimination that was perpetuated at least until 1977. The record includes numerous specific discriminatory practices, such as intentional failure to assign women and minorities to certain areas within the bank, as well as promotion and salary discrimination. Although some of the violations began prior to the effective date of the Executive Order, they were perpetuated in systematic discrimination even after the Government's Complaint was filed.
The continuing violations issue, however, has also proceeded on another level. Harris argues that in order to introduce evidence of discrimination between the effective dates of the Order and two years prior to the 1974 compliance review, continuing violations must again be established. The earlier Decision correctly holds otherwise. But again, if such a showing had been necessary, the systematic discrimination practiced by Harris from the effective dates forward is more than sufficient to establish a pattern of continuing violations during that period. The historical pattern of discrimination discussed in the initial Decision along with the testimony of bank employees, including officials and the 5 victims, proves a discriminatory system existed throughout the entire period. With the statistical presentations, the evidence is overwhelming.
34As elsewhere, throughout the presentations of statistical evidence the parties continually twisted the analyses to suit their needs. However, one interesting twist on this type of manipulation was provided by Dr. Roberts. Since the inception of this proceeding he has maintained that one of the major problems with the statistical studies is the inability to sufficiently disaggregate job groups. Apparently, faced with the results of his own salary growth regressions, he determined that the output indicated excessive variation due to ommitted variables. He felt that the way to compensate for the problem in this instance was to consider the coefficients as a pattern, and so he aggregated the results. (85R-31A, chap. 3).
35This is a bizzare twist of fate in that it is the same evidence that was excluded at the 1979 hearing. It was that exclusion which necessitated this hearing on remand.
36The story is told about how detailed records were kept between 1900 and 1982 of the amount of krill estimated to have been eaten by all Antarctic mammals. A statistical whiz, with unlimited use of free computer time, compared these observations with both the gross national product of Lithuania in 1985 and the sale of liters of wine in Andorra in 1986. He found several direct correlations. He concluded that he could show that krill eaten was an absolute predictor for all sorts of phenomena if given appropriate access to a free computer. It is also told that he received large fees in many court cases by testifying about how krill eaten in Antarctic was the missing variable in the statistical analysis of one party or another in merger and discrimination matters. Luckily, no such presentation was made in this case and this "omitted" variable was not addressed.
37One is compelled to conclude that as far as women and minorities were concerned the college graduate training program was of little value. Ms. Arthur was not in that program, yet she was placed in the same job as Ms. Schumacher who was. If the position did not require a college degree Ms. Schmacher should never have been placed there. If it did require a degree, Ms. Arthur's placement in the position undercuts the Bank's position that she could not be expected to advance as quickly as a person in that program. The inconsistency is obvious.
38In its briefs the Department repeatedly asserted that the earlier Decision found Harris' performance evaluation system to be discriminatory under the disparate impact theory. This is plainly wrong. The portions of that Decision cited by the Government do nothing more than make findings concerning raw disparities in supervisory ratings based on race and sex. This clearly falls far short of the legal conclusion Plaintiff would confer upon that language.
Moreover, the evaluation system is not subject to attack under that theory. Harris' performance evaluation is not the type of facially neutral policy, such as a diploma requirement, that the Supreme Court has implicitly required for application of the theory. The Bank's evaluation process is highly subjective and cannot be characterized as facially neutral. Although it has not made an unequivocal ruling on the question, the Court has implied that the disparate impact theory should only be applied to such neutral policies. Sears, 628 F.Supp. at 1282. This proceeding concerns disparate treatment of women and minorities and must decided under that theory.
39The true flavor of the testimony of witnesses, in a case of this magnitude, is often lost in the hundreds of pages of legal argument. Any review of the record in this proceeding should begin with a reading of the testimony of Mr. Russell, a Bank Vice-President, which details the discrimination evident in the Bank's promotion system.
40However the Job Preference Study remained marked and identified as 85G-19C.
41The Federal Rules of Civil Procedure apply in this proceeding to the extent that the Rules of Practice and Procedure for Administrative Hearing do not. 20 C.F.R. §18.1(a); 41 C.F.R. §6030.1.
42Although some courts continue to couch the standard of Rule 11 in terms of bad faith, see, e.g., McLaughlin v. Bradlee, Corp., 602 F. Supp. 1412 (D.D.C. 1985); Aynesworth v. Beach Aircraft Corp., 604 F. Supp. 630 (W.D. Tex. 1985); Syufy Enterp. v. American Multicinema , 602 F. Supp. 1466 (N.D. Cal. 1983), it is clear that the amended Rule 11 requires a more stringent standard, that is objective rather than subjective. Davis v. Veslan Enterp., 765 F.2d 494 (5th Cir. 1985); Eastway Construction v. City of New York, 762 F.2d 243
43The Government dropped its request for debarment in this remand proceeding. Therefore that issue need not be addressed.
