as Dr. Treiman. See, e.g., Tr. 113-14, 132-138, and 144-48.
Dr. Michelson initially submitted a report to the plaintiffs on February 20, 1989; on November 1, 1989, he revised that report by means of appendices "F" and "G" to the original report. PX 1057 (report of Dr. Michelson to plaintiffs, dated November 1989 and entitled: "A Study of Salary Rate Differences Between Male and Female Employees of Nassau County, N.Y."). Dr. Michelson himself appears to believe that the revisions of his new report greatly enhance the quality of the initial study: "The result of these refinements is to make this presentation . . . the most exacting model of a salary system, using the most error-free data set, ever presented in litigation." PX 1057 at G-3. In any event, the revisions to Dr. Michelson's first report certainly had the effect of reducing the salary differentials that he attributed to sex. Tr. 132. To the extent that these reports are inconsistent, the court will consider his revised report as superseding his initial report.
Dr. Michelson undertook to evaluate the salaries of "historically" male-dominated jobs and "historically" female-dominated jobs (although he conceded to the court that he had no adequate bases on which to characterize job titles as "historically" male- or female-dominated). Tr. 13. Dr. Michelson proceeded, as have all persons involved in this case, on the arbitrary demarcation established by the plaintiffs as to sex domination: That is, Dr. Michelson accepted as given that a job title comprised of 70% women is "female-dominated" but that a job title with 69% women is "mixed." Tr. 13-14 and 114.
Dr. Michelson considered the mean salaries for Nassau County employees in the years 1983 and 1986. Tr. 9. He found that the raw difference in the average salaries of male-dominated jobs and of female-dominated jobs was $ 7,873 in 1983 and $ 9,166 in 1986. Tr. 15. Dr. Michelson then undertook to subject these "raw differences" to multiple regression analysis. Principally, Dr. Michelson sought to determine how much of these basic disparities were attributable to differences in the descriptions for the Nassau County job titles -- the bases on which the Cresap team had ultimately evaluated jobs and assigned salary grades. In his final analysis -- with Nassau County police officers considered separately (because they have a separate pay system, Tr. 29-30) but with job descriptions accounted for -- Dr. Michelson found that the difference between the average salary of male-dominated jobs and the average salary of female-dominated jobs was $ 2,473 in 1983 and $ 3,298 in 1986. Tr. 16; PX 1057 at G-20. Cast in slightly different terms, Dr. Michelson found -- again, with police considered separately and with job descriptions accounted for -- that every percentage point increase of females in a Nassau County career line corresponded to a decrease in salary of $ 31 in 1983 and of $ 42 in 1986. PX 1057 at G-21. Dr. Michelson testified that these sex-associated differences in salaries are not random. Tr. 104-05. Rather, Dr. Michelson found that the salary disparity in Nassau County is clearly related to the sex of Nassau County employees. Tr. 104-05, PX 1057 at 2.
The methodology of Dr. Michelson's study was to analyze Nassau County job descriptions and to try to create an independent variable for every aspect of these specifications that have a bearing on salaries. He noted that, although the Cresap team formally evaluated each job on the basis of four factors, these factors could be subsumed, in his opinion, under two groups: (1) factors that concern characteristics of the employee sought for the job (such as education and experience); and (2) characteristics of the job itself (such as complexity of duties and supervisory responsibility). Tr. 31. Dr. Michelson also recognized, aside from the job descriptions, that the seniority of employees has an effect on salary, and he attempted to account for that fact in his regression analyses. PX 1057 at 133. But his principal task was to evaluate the effect that the characteristics of the Nassau County job descriptions have on the critical dependent variable against which all aspects of this lawsuit are measured -- salary. It was therefore the objective of Dr. Michelson to account for as many of these characteristics as was both possible and appropriate. PX 1057 at G-21 to G-22.
In the formulation of his data base, Dr. Michelson began with the decision to analyze "career lines" rather than separate job titles. Tr. 34-35. Thus, rather than consider separately the jobs of Probation Officer Trainee, Probation Officer I, Probation Officer II, Probation Officer Supervisor I, and the rest of the probation officer jobs separately, he evaluated the entire probation officer career line as one unit. Dr. Michelson also testified that, in Nassau County job titles, most women are in female-dominated career lines and most men are in male-dominated career lines. Tr. 53.
As noted, the most significant aspect of Dr. Michelson's study was his derivation of independent variables from the Nassau County job descriptions. With the aid of the "writer's manual" produced by Cresap to train the Nassau County team to write job descriptions, PX 620, Dr. Michelson translated the language of these title specifications into several dozen independent variables. Tr. 60. For example, many job specifications indicate that a particular job title requires a certain degree of knowledge about certain subject matter. If the specification indicated that the title required only some or no knowledge about certain subject matter, Dr. Michelson recorded the variable "KNOW01" for that job. If the job specification indicated that "knowledge" simpliciter was required, he coded it as "KNOW2". The variable "KNKOW3" and "KNOW4" represented job specification language of "considerable knowledge" and of "thorough knowledge" respectively. PX 1057 at G-5; Tr. 60; see also PX 1057 at 100-01 and at G-4 to G-5 (comprehensive lists of independent variables). Dr. Michelson conceded that he did not attempt to code every "minute variation" in the language of the job specification and also that such coding of language required subjective judgments, Tr. 73-75; nonetheless, he testified that, in his opinion, he had accounted for all variables that had explanatory power with respect to the salary differentials he had earlier identified. Tr. 77.
Again, after coding the language of the job specifications -- both the characteristics of the job itself and the characteristics of the person required for the job -- into independent variables and after accounting as well for seniority, Dr. Michelson concluded that:
Some of the salary differential is explained by characteristics held differently by males and females, and gained (arguably and assumedly) independent of the actions of Nassau County. Some of the salary differential is explained by differences in the specifications of the jobs males and females are in. However, . . . neither individual nor job characteristics explain all of the sex-related salary differences. Salary appears to be related to gender itself.
PX 1057 at 1-2.
Although the scope of Dr. Michelson's undertaking in this project was ambitious, certain methodological shortcomings prevent this court from assigning more than minimal weight to his conclusions. As a threshold matter, the court must discount somewhat Dr. Michelson's own assertions that he accounted for all statistically significant variables in his regression analysis. He himself testified that he had to revise his initial report because he concluded that important variables had not been considered. Certainly he would not have submitted his initial report as "final" in February of 1989 did he not believe then that it was comprehensive. He also characterized the process of identifying and of coding significant aspects of the job specifications as "unending." Tr. 20.
Perhaps one of the most troubling aspects of Dr. Michelson's study is that, even though he himself did not personally code any of the job specifications, and even though he acknowledges that the coding process required subjective judgments, he did not record any of these judgments for subsequent review or evaluation. Tr. 216. In fact, despite his insistence to the contrary, Dr. Michelson made clear on cross-examination that he cannot now determine whether or not his staff followed his instructions in the coding process because he does not know what instructions he gave. Thus, for example, on the question of how to code the phrase "prepares progress reports", the following exchange took place:
Q. Do you know whether in every case where your staff saw the language "prepares progress reports," they coded it as communication?
A. No, I only know they either did or didn't.