Full opinion text
MEMORANDUM OPINION AND ORDER PHILIP PRATT, District Judge. This opinion contains the court’s findings of fact and conclusions of law on the merits of these consolidated cases. For the reasons stated below, the court finds that based upon the evidence the plaintiffs are entitled to 1.5 million dollars as a monetary award. Under the terms of the consent judgment entered heretofore, the defendant Michigan Blue Cross/Blue Shield did not nor does not admit it engaged in discriminatory policies or practices. Further, the parties agreed that this court was not to make any determination that discrimination did, in fact, occur. Rather the court’s role was confined to the evaluation of the statistical evidence and whether that evidence, which comprises the trial record in this case, warranted an award of damages of no less than $1,500,000, and no more than $4,500,-000. Thus, it should be clearly understood that the findings below are not intended nor should they be construed to constitute a finding of discrimination, nor, on the other hand, a finding that there was no discrimination. I. BACKGROUND The above entitled consolidated class action cases charge that the defendant, Michigan Blue Cross/Blue Shield, Incorporated (“BCBSM”), a Michigan corporation, discriminated against women in initial placement and promotion practices between the years 1972 and 1982. The class is comprised of female employees in Management, Administrative and Technical (“MATs”) or “exempt” positions in various divisions of the defendant. As certified, the class consists of approximately 4,300 women. As is not unusual in class actions alleging discriminatory practices, progress was laborious and tedious. However, because of the unique settlement agreement reached by the parties and memorialized in a Consent Judgment entered by this court, only a brief recital of the procedural history need be recounted, and that only for the purposes of providing the context within which this memorandum is to be viewed. The cases were instituted in 1975 and alleged discriminatory practices to the disadvantage of women in every facet of defendant’s employment policies and practices from initial hiring through discharge— or, as frequently characterized, “across the board” discrimination. The jurisdictional bases were the Equal Pay Act, 29 U.S.C. § 206, and Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e, et seq. The “across the board” approach necessarily contemplated several thousand employees in both “exempt” and “non-exempt” positions. Lacking pre-suit, hard facts regarding the internal workings of the defendant with regard to employment policies and practices, as opposed to anecdotal information, pre-trial discovery was an arduous task. The problem for both sides, was exacerbated by not only the sheer size of the undertaking, but also because of some significant developments. For example, in 1975, Blue Cross and Blue Shield, which had been separate entities, merged into a single corporation, and there was an amalgamation of personnel, records and policies; the introduction of the Hay System of job classifications and salary ranges; the refinement of the “across the board” charges to discrete claims; the honing of the class and the identification of appropriate representatives; legal developments concerning the viability of wage claims under the Equal Pay Act vis-a-vis Title VII wage claims; the advent of computerization of personnel records with its attendant problems of accurate and complete processing, retrieval and, eventually, preparation for statistical analysis. Contributing most heavily to the complexity of the cases, particularly as it related to statistical proofs upon which the plaintiffs relied substantially, however, was the nature and structure of the defendant’s organization. The health care business is highly sophisticated and includes many diverse technical and professional disciplines. As the class evolved, it became readily apparent that the positions in the company did not constitute a fungible group. Job descriptions numbered in the many hundreds; skills, education and experience factors were not interchangeable; job descriptions ran the gamut from doctors, dentists, lawyers, accountants, actuaries, marketing specialists and media representatives through computer analysts, customer billing consultants and related technical fields. Thus, the frequently used approach of merely comparing raw average pay differentials, for example, would be inappropriate, if not impossible. Consequently, the proposed utilization of statistical evidence at trial required exhaustive research, translation, and preparation. In addition, of course, were the problems associated with the investigation, discovery, and preparation of anecdotal evidence with relation to specific acts or conduct of management from hundreds of employees. Eventually these obstacles were overcome and trial, after several adjournments, commenced on December 15, 1983. After the second day of what was anticipated to be a 12 to 16 week trial, the parties advised the court that settlement negotiations had reached an encouraging level. At the suggestion of the parties, conferences were held with the court and culminated in a proposed Consent Judgment. Trial was suspended and the requisite procedures under Federal Rules of Civil Procedure, Rule 23 to effectuate the settlement were conducted. On March 26, 1984, the court conducted a hearing concerning the propriety of the proposed agreement and found that the Consent Judgment should be approved and entered. On April 6, 1984, the court issued its Order Approving the Consent Judgment. The Consent Judgment can only be described as unique and a brief exposition will facilitate an understanding of the issues to be determined and the focus of this Opinion. Foremost, the Consent Judgment essentially assumes liability. It does not suggest, and the court does not mean to imply, that BCBSM admits to discriminatory policies and practices, but the consent decree transforms the issue of liability into a question of relief. Further, the Consent Judgment concerns itself with equitable relief. It recognizes and contains certain modifications which had been previously implemented by the defendant with respect to employment practices as they affect women and other minorities. It also included additional modifications in employment practices to assure related, non-discriminatory policies and practices. These provisions were final and needed no further judicial consideration. The unique feature of the Consent Judgment dealt with the issue of monetary award. The court previously described the pertinent provisions in this regard as follows: The issue of monetary award has been hotly disputed. The disagreement does not rise simply from the relative merits with which each party perceives the case, but also from the size of the class and the variety and number of the positions involved. In an effort to compromise, the parties have agreed to a floor and ceiling regarding an award of 1.5 million and 4.5 million, respectively. These figures represent the minimum and maximum of the linear range of awardable damages. The exact figure is to be established by the Court after a trial on the merits. Under the proposed judgment, trial will be limited to the submission of statistical evidence. At the conclusion of this presentation and any rebuttal the parties deem necessary, the Court will fix damages on the “strength” of plaintiffs’ statistical case. Thus, if plaintiffs make no showing of discrimination with the statistical evidence or if that showing is very weak, then plaintiffs will only be entitled to 1.5 million. Conversely, if the plaintiffs’ case is very “strong,” then the Court will award 4.5 million. In the event plaintiffs’ statistical case is neither “weak” nor “strong,” but falls somewhere in between, then the Court will assess the award appropriately between 1.5 and 4.5 million. Moreover, whatever plaintiffs’ recovery, the decree provides for the payment by defendant of reasonable fees and costs incurred by plaintiffs, thereby protecting the fiscal integrity of the award. Memorandum Opinion and Order Approving Consent Judgment, April 6, 1984, at page 6. Further, the Consent Judgment provided that: In determining whether a case” has been made out, or to what degree a case has been made out, the Court should take into consideration the totality of all evidence before the Court at Om conclusion of the statistical presentation which is within the guidelines set above. ... The Court is not to be governed solely by a statistical dollar analysis which may show dollar differentials in excess of the maximum of the range. Such showings are only intended to be evidence of the strength (or lack of strength) of the overall case and are not intended to have a dollar for dollar relationship to the amount of the award. On the other hand, the parties contemplate that the Court will take into account the period of time, if any, during which the Court concludes the plaintiffs have shown the existence of a “case” of discrimination, the nature of the discrimination, together with the strength of that showing. Consent Judgment, at pages 7 and 8. The effect of these provisions 'S to modify the usual order of proof and the customary step-by-step analysis suggested by the Supreme Court in such cases as McDonnell Douglas Corp. v. Green, 411 U S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973) and Texas Dept. of Community Affairs v. Burdine, 450 U.S. 248, 101 S.Ct. 1089, 67 L.Ed.2d 207 (1981), which direct, in brief, that the plaintiff must first establish a prima facie case, the defendant must show business necessity, and the plaintiff must establish that the asserted business necessity is pretextual. The ultimate burden, however, is always to remain with the plaintiff to prove the fact of discrimination. In accordance with this agreement, the court conducted the trial in a one week period, the parties submitted post-trial statistical rebuttals and briefs and the case is now ripe for decision. II. INTRODUCTION TO STATISTICAL EVIDENCE The plaintiff class is composed of nonclerical employees during 1972 to 1982 and restricted to the managerial ranks. As adverted to above, the “across the board” attack on alleged discriminatory employment practices eventually resolved itself into two issues: did the defendant discriminate against women in (i) initial placement in the MAT ranks, and (ii) promotions within the MAT ranks. Pursuant to the Consent Judgment, the parties presented statistical evidence in the following fashion. First, Barbara R. Bergmann, Ph.D., testified on behalf of the plaintiffs and submitted her studies and reports. Finis Welch, Ph.D., then testified for defendant and presented his studies and reports. Trial was recessed and thereafter Mark R. Killingsworth, Ph.D., submitted a written report on behalf of plaintiffs’ analyzing the statistical evidence presented by Dr. Welch; and Edward D. Rothman, Ph.D., for defendant, presented a written “Evaluation and Critique of Studies Presented by Dr. Bergmann and Dr. Welch.” These reports were followed by Dr. Bergman’s and Dr. Welch’s respective submission of written surrebuttal reports. Finally, the parties presented final arguments in post-trial briefs. Prefatory to the discussion of the statistical evidence, the following observations are in order. As mentioned, the court is not to apply strictly the traditional rules concerning the burden of proof pursuant to the Consent Judgment. The award in this case is to be set by the “strength” or “weakness” of the plaintiffs’ case and not by a showing of discrimination by the preponderance of the evidence. The “strength” of plaintiffs’ case, to the extent that plaintiffs introduced statistical evidence, has been substantially undermined because of the lack of worth of the studies and reports submitted by Dr. Bergmann on plaintiffs’ behalf. Virtually all of Dr. Bergmann’s reports contain a substantial number of data and specification errors and thus conclusions and results of these reports can be accorded little weight. As Dr. Rothman correctly emphasized in his critique, Dr. Bergmann made several computer processing errors in classifying and assigning information with such frequency as to discredit any findings produced from this data. Further, Dr. Bergmann systematically based her studies on the incorrect sample populations so as to bias her results. Additionally, almost all of Dr. Bergmann’s reports failed to provide adequate statistical models to evaluate the data meaningfully. See Valentino v. United States Postal Service, 674 F.2d 56 (D.C.Cir.1982). In short, Dr. Bergmann’s reports did not achieve the necessary standards of statistical reliability be considered helpful in the determination of the issues before the court. Plaintiffs implicitly recognize the severe deficiency of Dr. Bergmann’s analysis and reports in their final arguments. Not only do plaintiffs not attempt to rehabilitate Dr. Bergmann’s testimony from the considerable criticism of defendant’s experts, but the plaintiffs do not rely on one report or study of Dr. Bergmann in arguing that the evidence established discriminatory hiring and promotion practices of the defendant. See Plaintiffs’ Post-Trial Brief. Instead, plaintiffs rely upon the data and much of the analyses of Dr. Welch and Dr. Killings-worth to present Dr. Welch’s reports in a way favorable to plaintiffs. Consequently, the arguments of the parties are not based upon the studies of their respective experts but are grounded on their interpretation and, in the plaintiffs’ case, modifications of Dr. Welch’s studies. But for the defendant’s expert’s studies, there would be no competent evidence before the court to make any finding of discrimination. Hence, in weighing the “strength” of plaintiff’s case, the court is mindful of plaintiffs’ failure to produce competent, affirmative evidence showing discrimination. III. THE STATISTICAL EVIDENCE The central issues in this case are whether comparably qualified women and men MAT employees received different treatment in their initial assignments and salaries and in their chances for promotion. As is generally necessary in discrimination cases, the procedure is to compare and contrast the treatment of individuals with comparable qualifications. In a case based primarily on statistical evidence such as the instant case, the procedure is to present statistical models which best demonstrate the treatment of comparable individuals. Commonly: [t]o be legally sufficient these statistics must show a disparity of treatment, eliminate the most common nondiscriminatory explanations of the disparity, and thus permit the inference that, absent other explanation, the disparity more likely than not resulted from illegal discrimination ... ... A plaintiff’s statistical evidence must therefore focus on eliminating this nondiscriminatory explanation by showing disparities in treatment between individuals with comparable qualifications for the positions at issue. The best known statistical method and the method chosen by the experts in this case is statistical analysis by multiple linear regressions. Multiple linear regressions are mathemathical formulas which indicate how one variable is related to another. The object of this analysis is to determine the effect of sex, race, or age— the basis of the discrimination claim—on such aspects of employment as salaries, placements, and promotions. This analysis requires that comparable qualifications be accounted for so that the difference in salaries or promotions between persons of different races or sexes can be traceable to the variable of sex or race and not to dissimilar job qualifications and skills. Typically the independent variables will consist of factors which bear on productivity, such as educational level and prior experience. These variables are then analyzed by the use of computers through various formulas in relation to such independent variables as sex, age, or race and certain dependent variables such as wage and promotional rate. The object of this analysis is to determine whether after all relevant factors bearing upon productivity have been accounted for there remains a disparity in the treatment of employees based upon their race or sex. This disparity is normally shown in terms of different wage levels or rates of promotion. Thus, a critical aspect of this form of analysis is determining the variables which should be included in the regression. As stated in Segar v. Smith, 738 F.2d 1249, 1261 (D.C. Cir.1984) (citations omitted): The first step in multiple regression analysis is specification of the independent variable. The choice of proper explanatory variables determines the validity of the regression analysis. A coherent theory, devised prior to observation of the particular data, must be employed to select the relevant explanatory variables. ... When the proper variables have been selected, the multiple regression analysis is conducted, generally by a computer. In essence, the regression measures the impact of each explanatory variable upon the dependent variable [e.g., salary] by holding all other explanatory variables constant. The analysis yields figures demonstrating how much of an observed disparity in salaries can be traced to race, as opposed to any of the other potential explanatory variables [e.g., educational level]. The results of these analyses are numeric figures representing, in this case, percentages or dollars. The reliability of these figures is weighed by a measure known as the “T-Ratio.” The T-Ratio reflects the probability that the result obtained could have occurred by chance. Simply, the T-Ratio indicates how likely it is that the result of the regression was produced by chance. If the result is not likely the product of chance then it is concluded that it represents a true relationship between the measure of discrimination (e.g., wage rate) and an independent variable (e.g. sex or education). Generally, the higher the T-Ratio the less likely that the figures are based on chance. The experts in this case have followed the commonly recognized standard that a T-Ratio of approximately 2.0—indicating a 5% probability or a one in 20 probability that the observed result could have occurred by chance—or above indicates that the figure is statistically significant. In determining the variables to be used in the regressions in this case, the nature of the job positions and the composition of the personnel needs further explanation. The positions in the MAT ranks vary substantially. Low level MAT employees generally perform simple clerical work in processing claims, while upper level MAT positions require system analysts, actuaries, lawyers, doctors, and other professionals. Further, the nature of the defendant’s business requires a wide variety of specialties ranging from the medical profession to computer programing. The duties and skills necessary for many of the positions are not easily defined and are not readily comparable. The positions at issue here are quite diverse, necessitating individuals with specialized backgrounds. Hence, since MAT positions require and are composed of individuals of divergent backgrounds, to be meaningful multiple regressions must employ a considerable degree of sophistication in the selection of data and its field of variables. Not only are the skills and backgrounds required to perform MAT positions generally not fungible, there appears to be a sizeable difference in the backgrounds of male and female MAT employees. Dr. Welch presented studies concerning the composition of the qualifications of MAT personnel which the plaintiffs did not dispute. Women account for two-thirds of MAT employees, yet 43% of the women compared to 12.7% of the men were hired when they were 21 years or younger. Conversely, almost twice the percentage of men (36.8%) were hired over the age of thirty as compared to women over 30 (19.4%). Almost half (47.8%) of the women MAT employees did not have any education beyond high school, while only 14.1% of the men had no further education. As to bachelors degrees, only 12.2% of the women, compared to 53.8% of the men, obtained this degree. Moreover, 14.4% of the men hold advanced college degrees versus only 2.3% of the women. Dr. Welch reported: Sex differences in education levels are further amplified by examining fields of specialization. ****** The trade business and technical school group provides an example of the kinds of differences found throughout the table. Half of the men specialized in computer, math, and actuarial whereas half of the women specialized in secretarial and keypunching. If we add together the first three specialities, business economics, engineering, drafting, and computer, math, actuarial sciences, we see that proportionately more than twice as many men as women majored in these areas. They account for 61.9, 72.1, and 57.7 percent of the men who, respectively, had some college, a bachelors degree, or an advanced degree, but they account for only 22.7, 29.9, and 25.3 percent of the women at these levels. Dr. Welch’s November 7, 1983 Report, at page 7. Dr. Welch further reported that men and women also differ in their prior experience. Fifteen percent of the men previously worked as computer operators or programmers, while only 1.5% of the women previously performed such work. Further, Dr. Welch commented: ... 72.4 percent of the women entered these grades through promotion from lower grades [non-MAT positions]. Since promotion is most often one or two grade levels, these entries result in our first seeing the majority of women in the lower ends of the [MAT] grade range. In contrast, 27.8 percent of the men enter the reference range via promotion while 72.2 percent were hired directly into these grades. Dr. Welch’s Report of November 7, 1983, at page 7. Elsewhere, Dr. Welch notes that male MÁT hires (37.5%) are more likely than female MAT hires (11.0%) to have their longest job experience as an accountant or controller, programmer, or manager. Moreover, Dr. Welch’s study demonstrated that 61.5% of men compared to 22.5% of the women had their longest previous jobs in a specialized field relevant to the insurance industry. Among female MAT employees 28.4% of their prior career was predominantly clerical or secretarial while only 5.8% of the male MAT employees had such prior experience. It is against this background that the plaintiffs challenge the defendant’s policies and practices in the initial assignment and promotion of comparably qualified men and women. A. Initial Assignment Plaintiffs allege that women were discriminated against by defendant in their initial placement into the MAT ranks. This challenge does not concern employees who were in a non-MAT position then promoted to the MAT ranks, but only those who were initially hired into MAT levels. Since more than 70% of women MAT employees reach MAT positions through promotions from non-MAT jobs, this dispute concerns less than a third of all MAT female employees. As a benchmark or indicator of the treatment of men and women in initial assignment, the experts’ studies focused on the initial salary of new hires to determine if men and women of comparable qualifications received comparable salaries. Dr. Welch determined that the “raw differential” between men and women when the only independent variables are sex, the year of hire and full or part-time work, was 23.8% to the disadvantage of women. When Dr. Welch accounted for educational background through various independent variables the sex differential or coefficient decreased to 16.7% to the disadvantage of women. Dr. Welch ran a further regression which accounted for the type, length, and continuity of previous experience and found a sex differential of 5.9% to the disadvantage of women, with a corresponding T-Ratio of 2.8. Finally, Dr. Welch performed a study which contained an additional variable which purported to represent the job requested by the employee on his or her application for employment. This regression resulted in a sex coefficient which showed no advantage to men over women. Hence, if Dr. Welch’s regressions and analyses are accurate there is no showing of discriminatory conduct by the defendant in its initial hiring practices and policies. Plaintiffs concede the accuracy of Dr. Welch’s regressions which account for education and much of the regression which considers prior experience. Plaintiffs present two primary objections to the remainder of Dr. Welch’s studies. First, plaintiffs contend that Dr. Welch improperly implemented and used the “job requested” variable. Second, plaintiffs argue that Dr. Welch inappropriately used female interaction terms in the context of blank job request applications and leading career gaps. Plaintiffs assert that a proper study would not include the “job requested” variable or the female interaction terms. Plaintiffs calculate that if these two objections are factored into Dr. Welch’s study, then the sex coefficient or the disparity between comparably qualified men and women would be 7.5% to the disadvantage of women. Plaintiffs determined that the female interaction term used for the “leading gap,” a factor in Dr. Welch’s prior experience variable, accounts for 1.6% of the raw differential. Dr. Welch’s regression which accounted for education and pri- or experience found a sex differential of 5.9% to the disadvantage of women. If plaintiffs’ argument is credited then that figure would be 7.5%. Plaintiffs assert that the use of the “job requested” variable is inappropriate and should not be considered by the court; thus the correct sex coefficient to be derived from Dr. Welch’s regression which accounts for education and prior experience without a female interaction variable is the 7.5% figure. Plaintiffs argue, in the alternative, that if there is some appropriate use for the “job requested” variable, the female interaction term used within that variable is improper. That female interaction term, according to plaintiff, accounts for nearly 3.5% of the sex differential. Thus, if that term were excluded then there would be a sex coefficient of 3.5% even accepting all other calculations and variables used by Dr. Welch. The following table summarizes these figures: Sex Coefficient (negative signifies to the Regiession disadvantage of women) Raw Differential (“R.D.”) independent variables include sex, year of hut?, full or part-time —23.8% R.D. with education variable —16.7% R.D. with education and prior career without female interaction term variables —7.5% R.D. with education and prior carter witii female interaction term variables —5.9% R.D, with edi/cation, prior career and ‘‘job requested” without female interaction term variab'ies —3.5% R.D. with ed .cation, prior career and “job requested” with female interaction team variables 0.0% A detailed discussion of plaintiffs’ arguments follows. Plaintiffs’ first objection concerns Dr. Welch’s use of the “job requested” variable. The “job requested” variable reflects the job or position which is requested on the application for employment. Dr. Welch justified the use of this variable because of perceived problems with the data. The application used by BCBSM provided four blanks for candidates to fill in their prior employment experience. If a candidate had inore than four prior jobs then this information was not picked up in the data. Further, several applicants did not bother to fill in more than one or two prior jobs, although it could be inferred that the candidate had more experience. More importantly, there existed the problem of classifying the prior experience and education provided by the candidates in a way that allowed for statistical comparison. At trial Dr. Welch presented a study (summarized in Exhibit DF) which demonstrated this problem. The study showed that using reasonable statistical assumptions about the information provided, accurate salary levels could not be readily predicted. Many individuals, both male and female, actually received salaries much higher or lower than would be predicted from their application material. Dr. Welch explained that this occurred because of the ambiguity of the applicants’ descriptions of their backgrounds, the unfamiliarity of the statistician with certain fields, which led to improper valuation of experience or education, and the absence of other relevant information which was not written on the application. Dr. Welch opined that the “job requested” variable overcomes this problem because “the particular job for which an individual applies can summarize some important dimensions of the individual’s level and type of skill at the time of hire.” Dr. Welch’s December 16, 1983 Report, at page 18. According to Dr. Welch, this variable’s prime value is that it “has the capability of distinguishing a special niche in skills between people with specialized talents and the firm ...” Transcript, March 29, 1984, at page 839. Dr. Welch also supplied several studies to support the reliability of the “job requested” variable. The first showed the high correlation between applicants’ prior careers and the job requested. Dr. Welch examined those who requested positions in the areas of computer programming and accounting among all hires. This study showed that 85.1% of the men and 84.9% of the women who applied for the position of programmer had related experience or education. The study also revealed that 88.7% of the men and 82.6% of the women who applied for accounting positions had prior experience or education in that field. The study further demonstrated that 93.0% of men and 100.0% of the women who requested programmer positions had their longest previous prior occupation in that field. While 88.6% of the men and 93.8% of the women who requested accounting had their longest prior occupation in the accounting field. A second study submitted by Dr. Welch explored the correlation between the prior career and the job requested in the area of clerical employees. The following table summarizes that study: Type of Prior Experience of Those Requesting Clerical Jobs All Hires Experience Women Men None 29.1% 31.0% Secretary 16.2% 1.6% Other Clerical 29.1% 13.9% Subtotal 74.4% 46.5% Experience Relevant to Insurance Industry 1.1% 7.1% MAT Hires Experience Women Men None 5.0% 21.7% Secretary . 15.0% 0.0% Other Clerical 25.0% 21.7% Subtotal 45.0% 43.4% Experience Relevant to Insurance Industry 7.5% 13.0% Another study offered by Dr. Welch concluded that “[m]en and women not only arrive at BCBSM with very different kinds of backgrounds, they also seek to utilize their abilities in different kinds of jobs.” Dr. Welch’s December 16, 1983 Report, at page 20. This study revealed that two-thirds (66.9%) of all women hired, in MAT and non-MAT positions, asked to be placed into clerical positions, compared to 19.2 percent of the men. Furthermore, 53.2% of all men hired requested positions as a computer programmer, financial, and analyst categories compared to 8.7% of all women hired. The final studies relevant to the reliability of the “job requested” variable concerned the likelihood that men and women would receive the job and pay wage they requested. The results of this study are reported in Tables 5.1 and 5.2 of Dr. Welch’s December 16, 1983 Report. These studies show that women and men were equally likely to get the job and pay wage requested. The plaintiffs raise several arguments against the use of the “job requested” variable, but only one deserves comment. There does appear to be a taint to the reliability of this variable due to the possibility of “steering” performed by BCBSM personnel. The studies presented by Dr. Welch would be very convincing if they were unquestionably generated by the applicant and not the result of actions by BCBSM personnel. However, the practice and policies of the defendant contain the potential for steering. First, several of the job requested blanks on employee applications (as high as 16%) were filled in by someone other than the employee, most likely BCBSM personnel responsible for hiring. Second and importantly, many of the applications have requests for positions under the very specific and unique names used by the defendant. Apparently the names of these positions would not be known to persons not already or previously employed by defendant unless informed by a BCBSM employee. Terry .Rotare, defendant’s senior personnel representative, testified that approximately 75% of the MAT hires apply to BCBSM due to the defendant’s use of advertising agencies, employment agencies, federal and state agencies, and newspaper and trade journal advertisements. Such candidates would usually fill out their applications when they came to the defendant for an interview. She further testified that the BCBSM would many times give the employment agencies or sources the generic name of the position to be filled and not the specific name used by the company. Ms. Rotare testified under cross-examination as follows: Q. And when you send the listing of job names to the agencies, is there any policy requiring whether or not you use generic names or the company’s specific name for the job? A. A notification of a job opening that is sent to, let’s say, an employment agency, for example, or even some of the more professional type associations, we send a copy of what we call a job posting, which would have the generic title on it. Transcript, April 3, 1984, at page 1152. Ms. Rotare explained further during redirect examination: Q. ... You indicated the title you refer to, whatever source we’re talking about, would be the generic title. That was a term suggested to you by [plaintiff’s counsel] in his question. What is your understanding of the term generic? A. Um, I guess you’d say original, true, clear. sfc # sjt ¡{c s-s s}s Q. What is the difference, in your view? And I refer you to your direct testimony, where you said you gave the specific job title to the source. You testified, in response to (plaintiffs’ counsel’s) question, that you would give the generic title. What is the difference? A. Okay. I’m thinking of, for example, when we’re looking for an industrial engineer. We would put an ad in the newspaper that we’re looking for Industrial Engineers, but our job title says, Production Analyst. Q. All right. And the generic title is what? A. Well, I would say Industrial Engineer. Q. In that instance, that’s the name you would put in there? A. Right. Q. And are there other times when you put no specific title, the way you use it, into the advertisement? A. Um, it varies, right ... We wouldn’t catch the eye of the applicant if we said Production Analyst, for example. Transcript, April 3, 1984, at pages 1157-58. This testimony revealed that the job requested information was generally not provided until the candidate was at an interview and that many of the candidates could not have filled in the specific job title which appears for job requested on their application without the guidance of BCBSM personnel. Thus, there is the possibility that BCBSM personnel could “steer” women into less favorable jobs than male applicants by instructing them on which job to request. The possibility of steering taints the reliability of the “job requested” variable, but this possibility by itself is not compelling enough to discount totally the value of this variable. Dr. Welch presented several studies discussed above which supported its use. Of particular significance is his study which correlates the job requested by applicants with their education and prior career. This study suggests that the job requested information was not influenced by steering by BCBSM personnel. Given the possibility of steering, on the one hand, and the studies showing the reliability of the variable on the other, the court concludes that the “job requested” variable cannot be fully accepted by the court nor totally rejected, but its value should be appropriately discounted. Plaintiffs’ second major objection concerns Dr. Welch’s use of a sex interaction term. This term considers and weighs, within the regression, whether certain data is derived from male or female employees. Dr. Welch used these terms in two situations. First, some of the employees’ applications did not have any information for the job they were requesting. Dr. Welch used a female interaction term where the job requested was left blank. Second, the prior careers of several candidates were not fully described in their applications. The applications only permitted candidates to fill in their last four prior jobs. If the candidate had more than four jobs, then a “gap” in the information occurred because there was no information concerning activities from the time the applicant ended his or her education until the time of the earliest recorded job. Dr. Welch labeled this gap of information a “suspicious leading gap”—“suspicious” because the employee may have been active during this period in his or her career, but that time could not be specifically evaluated. Dr. Welch also used a female interaction term for individuals who had these leading gaps. The use of the female interaction term is somewhat controversial. Although it is not inherently improper, because of its underlying assumptions, it should be used with circumspection. The term attempts to measure possible differences in men and women where there is a lack of information. Discussing the use of the interaction term in the case of “suspicious leading gaps” will best illustrate the relation of the female interaction term to missing information and certain assumptions about men and women. In his report of December 16, 1983, Dr. Welch presented a study which related the length Of MAT employees prior careers— the period from when education ended and employment with BCBSM began—to starting salary. The study showed that when men and women are hired out of school, with little or no prior career, into MAT positions there is virtually no difference in their respective starting salaries and initial assignments. The study further showed, however, that the longer the prior career, the larger the difference in starting salary between the sexes to the disadvantage of women. Thus, when the sheer length of the potential career (the period between the completion' of education and hire by BCBSM) is measured without regard to its quality, women whose prior career has a potential of twenty years receive, on average, substantially less than men with potential prior careers of the same length. Since there is no sex differential in initial assignments for those who have no prior career, it was legitimate for Dr. Welch to search for other variables which might explain the differences in initial assignment for those with lengthy careers. Dr. Welch presented an additional study which attempted to show that the quality of the prior career had substantial effect on the initial salaries of individuals. The study showed that the greater the length of the potential prior career the greater the differences in initial salaries between men. According to Dr. Welch: “Diversity in preBCBSM activities results in increased divergence in starting BCBSM salaries among men with long prior careers relative to men with short (or no) prior careers____ Evidently, how a new employee has spent his prior career affects how much- money he receives from BCBSM.” Dr. Welch’s Report of December 16, 1983, at pages 30 and 32. Dr. Welch presented a further study, Table 3.4 of the December 16, 1983 Report, which shows the effect of certain career activities on initial salary at hire. The study shows that starting salary differs systematically for persons with different prior careers. The study also showed that individuals with suspicious leading gaps who had all four blanks of the application filled in received wage increases per each year of the gap of 2.1% more than those who did not have suspicious time gaps. Those, however, with suspicious leading gaps who did not fill in all the blanks had the initial salary diminished by 1.0% per each year of the gap. The study is summarized in the following table: Results in Salary Time Spent in . . . Increase per year Other BCBSM Job (not for BCBSM) 5.1% Other Job 4.7% Military 5.3% Job While in School 6.3% In School 6.9% Leading Gap: 4 prior jobs placed on application 2.1% Fewer than 4 jobs —1.0% For these studies, Dr. Welch opined that productive prior experience was greatly valued at BCBSM and that the suspicious leading gap was a period when the individual added to his or her human capital—as the 2.1% figure suggests. Dr. Welch further made the reasonable assumption that the missing information in the suspicious leading gap was actually career information which influenced the employee’s placement in BCBSM. Relying on his studies concerning the background of men and women MATs, which conclude that women as a whole come to BCBSM with fewer qualifications, women do not generally, in the human capital sense, have as productive prior careers as do men. Dr. Welch concluded that any residual difference in initial salaries of men and women with long potential prior careers is due to the fact that men spent the period known as the “suspicious leading gap” more productively than women. Dr. Welch uses similar reasoning in implementing the term for blank job requests. Because men brought more skills to BCBSM and because men generally requested higher level professional and managerial positions, according to Dr. Welch, it is likely that a blank job requested had a different meaning for men than for women. Thus, Dr. Welch used a female interaction term to measure the possibility of these occurrences. In this way Dr. Welch attempted to characterize the missing information in terms of time and factors which may have added to the human capital of the individual. Plaintiffs delineate several problems with the use of the interaction term. Although the interaction term is useful in finding the possibility of differences in pri- or careers, it may also inadvertently mask discrimination. The interaction term accounts for a certain percentage of the sex differential. Dr. Welch asserts that the term measures real differences in the productive qualifications of men and women. The following example will demonstrate the potential problem with this model. A man and woman are hired at BCBSM with comparable qualifications and each has a suspicious gap of the same length and, although unbeknownst to the statistician, each spent that period performing the same activity. If the woman is placed in a lesser position due to discrimination, Dr. Welch’s interaction term could hide this bias. Dr. Welch’s regressions also included a variable representing sex alone. It is the coefficient of this variable which is the primary measure of discrimination. However, it is conceivable that the pattern of discrimination was complex and is also reflected or hidden in the coefficient of interaction terms between sex and other variables— e.g., suspicious leading gap. Second, the interaction term appears to account for too much of the sex differential considering the value of the information it is attempting to reflect. According to plaintiffs’ calculations, the interaction term with the “suspicious leading gap” accounts for 1.2% of the sex differential while the interaction term with the blank job request forms accounts for 2.4% of the differential. It does not seem likely, although possible, that this substantial difference in the differential would be caused by the information the interaction term purports to account for. Although productive time in a leading gap is somewhat important, it does not seem reasonable to assume that an employer would place this much weight on this factor in fixing initial salaries. Certainly the employees last four prior jobs should be much more compelling in that determination as is evidenced by the fact that BCBSM requests it. Similarly, although to a lesser extent, it does not appear that the interaction term for blank job requested applications should account for so much. Dr. Welch also submitted alternative studies which avoid the controversy of the interaction terms and are highly significant and revealing. In these studies he omits the interaction term and the aspect of the variable to which it was related. The first study contains a regression which includes all the variables discussed, but excludes individuals who did not fill in the job requested blank. This study, Table 3.1 of Dr. Welch’s Surrebuttal Report, showed a smail, statistically insignificant difference in the starting salaries of men and women. A second study, Table 3.2 of Dr. Welch’s Surrebuttal Report, omits individuals who have “suspicious leading gaps.” The results of "this report also show no large or significant difference to the detriment of women. Putting aside the problems with the “job requested” variable, Dr. Welch corrects summarized that “[ajmong persons with complete data, men and women start at BCBSM at comparable salaries.” Dr. Welch's Surrebuttal Report, at page 23. Plaintiffs contend that Dr. Welch should simply tuve excluded the interaction term in perfoi ming these tests and should not exclude the individuals with missing data. Essentially, plaintiffs’ argument in this regard requests that the court choose their assumptions over Dr. Welch’s but without sufficient evidence to warrant doing so. First, concerning the “suspicious leading gap,” plaintiffs assert that since the sex differential is non-existent for those with no prior career experience and increases as the length in potential prior career experience rises, the reason for: a larger disadvantage is because an employee can’t [sic] have a leading gap without a fairly long career. In other words, it is necessarly true that employees with leading gaps are more likely to be older and more experienced than those without such ‘gaps.’ It is hardly surprising that there is a greater difference among male and female [sic] with a B.A. in Business who have been in the labor market for ten (10) years as opposed to the males and females who have just graduated with a B.A. in Business. The reason is simple. Women with leading gaps, who suffer discrimination, will be more likely to suffer a larger loss simply because they are the most likely candidates for the higher paying jobs. Not only is this simple fact the most likely explanation for the higher disadvantage for women with leading gaps, but it is the best reason for not omitting such employees from the regression. Plaintiffs’ Post-Trial Brief, at page 33. Plaintiffs did not present any evidence or studies which support this assumption. They cannot point to any reason which makes their assumption more possible or acceptable than those offered by Dr. Welch. Instead, the court has before it two conflicting assumptions and insufficient information to test the validity of either. Dr. Welch presents studies which tend to support the common belief that men may have spent the time in a suspicious leading gap more productively than women. This court hesitates to adopt that position wholeheartedly because, as previously discussed, in the context of this case it may mask discrimination. On the other hand, plaintiffs request the court to assume the opposite without supporting evidence. Rather than choosing between these two possible assumptions, it is more prudent and logical, as Dr. Welch suggested, to simply exclude these individuals from the regression. The same analysis applied to the individuals with blank job requests. Plaintiffs argue that the court should treat blank job requests as expressing the applicant’s desire for any position. Again, plaintiffs offer no evidence to support nor compelling reason to accept this assumption. There are a number of reasons why the job request information was left blank. Since several applications have “anything” written for job request, it can be reasonably assumed that applicants who were interested in any position indicated so and did not leave the job request blank. Further, it may be there was no need for the applicant to indicate the job sought because of the circumstances surrounding the interviewing process. As opposed to plaintiffs’ assumption, Dr. Welch presents studies, which although not determinative, do support his assumption that there is a possibility that men and women who left the job requested blank were seeking very different positions, that is, men sought technical or professional positions while women sought clerical positions. Again, rather than attempting to choose between these possible assumptions, the better course is to simply eliminate these people from the regression. In summary, Dr. Welch’s initial assignment study found no sex differential in starting pay wages for men and women. Although most of Dr. Welch’s study can be accepted, there exists the problem of the interpretation of the “job requested” variable. Since there is the possibility of “steering” by BCBSM personnel, the court must discount the effect of that variable so that there is some, albeit small, showing of a sex differential to the disadvantage of women. Plaintiffs’ concerns about the sex interaction term are adequately satisfied by Dr. Welch’s regressions which exclude that information and those individuals affected from the studies and show no statistically significant differential. B. Promotions Plaintiffs also allege that women employees were discriminated against in terms of promotions within the MAT ranks. Plaintiffs concede that there is no statistical showing that women were overall less likely to receive promotions. Both Dr. Bergmann and Dr. Welch agree that, overall, women are actually promoted at higher rates than men. Further, plaintiffs admit that, overall, women MATs receive higher salary increases than men. Plaintiffs claim that discrimination in the promotion of women does not occur, overall, in MAT ranks but only in the higher levels. The levels of MAT positions range generally from exempt 1 through exempt 9. Dr. Bergmann presented several studies which were intended to show that women were victims of discrimination in moving into the higher MAT ranks. Her studies focused upon promotions into exempt 7 and above. Although Dr. Bergmann’s studies were seriously flawed and basically unusable, plaintiffs have maintained their claim that women were discriminated against in this fashion. Rather than relying upon Dr. Bergmann’s studies, plaintiffs attempt to show discrimination by arguing that the regressions of Dr. Welch should be modified in a manner which would result in sex differentials to the disadvantage of women. Dr. Welch presented several studies which indicated that women were not discriminated against in either overall MAT promotions nor in promotions into exempt 7 and above. See Dr. Welch’s December 16, 1983 Report, at pages 54-60. Among these studies was an analysis of the percentage of the exempt 7 and above positions which were occupied by women. This study indicated that the participation of women into these ranks doubled in the period of 1976 (12.6%) to 1982 (26.4%). Another study submitted by Dr. Welch showed that the percentage of women in exempt 7 and above was, overall, equal to or higher than the percentage of qualified available women in Michigan. Finally, Dr. Welch presented a regression which purported to measure the promotion rate of comparably qualified men and women into grades exempt 7 and above and which showed no significant sex differential. Plaintiffs contend that two of the variables Dr. Welch used in this latter regression were improper. The independent variables included tenure at BCBSM, education, grade, “computer relatedness” (if the individual was in an area related to computers), and positions in various divisions of BCBSM. In their post-trial brief, plaintiffs objected to the use of the grade and division variables. Plaintiffs, however, failed to offer substantive evidence or cogent reasons to warrant the exclusion of these variables. The grade variable controls for the grade of individuals and their chances of promotion into exempt 7 and above. This variable reflects the reasonable assumption that individuals, male or female, in exempt 5 and 6 have greater opportunity for being promoted in exempt 7 and above than individuals in lower MAT positions. This assumption appears to reflect the actual practices at BCBSM. Both Dr. Bergmann and Dr. Welch presented studies which indicated that promotions for exempt 7 and above generally come from the next lowest grades and rarely come from lower MAT levels. Plaintiffs do not dispute that mid-range MAT grade employees were promoted much more readily than low level MAT grade employees. In fact, courts which have considered this issue have rejected regressions measuring promotions which have not controlled for grade under similar circumstances. E.g., E.E.O.C. v. Federal Reserve Bank of Richmond, 698 F.2d 633, 658 (4th Cir.1983). See Valentino v. United States Postal Service, 674 F.2d 56 (D.C. Cir.1982). Moreover, Dr. Welch presented a study in his surrebuttal report which measured the difference in promotion rates by grade between men and women—after controlling for productivity-related factors, but without controlling for division or whether the employee holds a computer related job—and found that men and women were promoted at the same rate within each grade over the liability period. In other words, there is no disparity in the movement for men and women from grade to grade. Without the grade variable the regression would assume that all individuals are equally likely to be promoted into exempt 7 or above, every year, regardless of their grade. Since women hold a greater percentage of the lower MAT positions, it is certain that such a regression would indicate a sex differential to the disadvantage of women because of the chances of promotion from the lower exempt positions into exempt 7 are not as great as from the middle level exempt positions. The “division” variable purports to control for the effect of employment in a certain division of BCBSM on chances of promotion into exempt 7 and above. Dr. Welch included this variable because his studies indicated that individuals, male or female, in some divisions had greater opportunity for promotion into high level MAT ranks. In his report of November 7, 1983, Dr. Welch presented a study which showed that persons in the Marketing and Data Services Divisions generally received more promotions than other employees. BCBSM is organized into twenty divisions. Dr. Welch’s study presents the average annual promotion and salary growth rates during the period of 1976 to 1982, for the six largest divisions, separately with the remaining 14 divisions combined. The following table summarizes that report: Division Percent Promoted Membership and Regular Business Claims 17.9% Government Business 17.7% Marketing 21.2% Auto and National Marketing 15.5% Health Care Affairs 13.5% Data Services 21.6% Others 14.4% Significantly, Dr. Welch submitted another study in his report of November 7, 1983, which showed that women were promoted at the same rate as men within divisions. Hence, although employees’ chances for promotion depend on which division they were in, there is no showing of disparity of treatment between men and women within divisions. Plaintiffs’ arguments for excluding these variables are without merit. As for the division variable plaintiffs contend: “Division proves nothing. It is a clearly inappropriate variable. Using Division as an explanatory factor begs the question. If Division is a better career path why aren’t more women in Division A.” Plaintiffs’ Post-Trial Brief, at page 39. Concerning the grade variable, plaintiffs make a similar argument: ... if grade assignments are fair, they can be used to supplement the information we already have relative to the employees’ qualifications by telling us something about how otherwise equally qualified employees distinguished themselves on the job. The problem is that based upon this record we cannot assume nondiscriminatory initial placement and/or advancement from below. Plaintiffs’ Post-Trial Brief, at page 42. These statements merely reflect plaintiffs’ conclusions and have no support in the record nor in reason. It is plaintiffs who beg the question with these arguments. Plaintiffs’ contentions here are not the same as their arguments concerning the sex interaction term. Unlike the underlying assumtions of the sex interaction term, these variables do not assume that women possess less productive qualifications than men. The theory and implementation of these variables are completely neutral as to sex. The studies by Dr. Welch, which plaintiffs do not challenge or question, show that women and men were promoted equally from grade to grade and within divisions. Plaintiffs’ assertions are nothing more than claims that the vestiges of past discrimination are pervasive, that women were not initially placed into the MAT grades which would allow them to be promoted into the upper MAT ranks. The plaintiffs did not and cannot point to any study or analysis which would suggest that the grade or division variables bias Dr. Welch’s regressions. Nor, more importantly, do they show in any way that women were discriminated against in terms of promotion at any level at any time. Plaintiffs cannot properly rely upon this vestige of past discrimination argument without showing the past discrimination. The circumstances of this case are not akin to those faced by the court in Segar v Smith, 738 F.2d 1249 (D.C.Cir.1984). In Segar, the plaintiffs were black Drug Enforcement Agents who claimed discriminatory treatment in initial job assignments and promotions. Similar to the argument pressed here, the Segar plaintiffs claimed that they were unable to move to the upper level grades of the government’s grade system. They argued that blacks received less opportunity to advance above grade GS-12. The Segar plaintiffs’ statistical evidence, however, did not show discrimination in promotions above grade GS-12. Nonetheless, the circuit court upheld the ■district court’s findings of discrimination in promotion because plaintiffs proved “discrimination in the factors that bear most strongly on promotion.” Id. at 1283. The court found it significant that the plaintiffs “made a prima facie showing of discrimination in initial assignments, supervisory evaluations, and discipline. These are precisely the factors that determine a special agent’s prospects for discretionary promotions at higher levels of DEA.” Id. The plaintiffs in this case have not shown discrimination concerning the factors bearing upon promotion. They have not shown that women were discriminatorily kept from mid-range MAT positions or from divisions which exhibited the highest promotion rates. Thus, plaintiffs have failed to establish that Dr. Welch’s findings of no disparate treatment in promotions are unjustified. IV. CONCLUSION Pursuant to the consent decree the court is to determine the range of damages, from a low of $1.5 million to a high of $4.5 million, based upon the “strength” of plaintiffs’ case which, in turn, was to be primarily based upon statistical evidence. As provided in the decree, the court is to award plaintiffs upon a “weak” showing of discrimination the minimum of $1.5 million—if “the plaintiffs fail to meet their required burden of proof of showing class-wide discrimination, whether defined as a prima facie case, or on the basis of the statistical presentation ...”—and upon a “strong” showing an award of $4.5 million. A showing between “weak” or “strong” would ma