Full opinion text
MEMORANDUM OPINION AND ORDER KRAM, District Judge. Plaintiffs in this case claim that the State University of New York at New Paltz (“New Paltz”) discriminated against women in violation of 42 U.S.C. § 2000e et seq. (Title VII). The Court has certified a plaintiff class, which, pursuant to a stipulation, is composed of all full-time academic rank faculty members who were employed in the Division of Liberal Arts and Sciences at New Paltz at any time from academic years 1973 through 1984. In the same stipulation, the plaintiffs limited their claims to discrimination in salary, rank, and initial placement in rank. Plaintiff Roberta Ottaviani is the class representative. Plaintiff-intervenors Dorothy Jessup, Joan Marie de la Cova, and Harriet Klapper are members of the plaintiff class because they were full-time academic rank members of New Paltz’ Division of Liberal Arts and Sciences during the class period. Plaintiff-intervenor Caro-lee Schneeman is not a member of plaintiff class because she was a member of the Fine Arts Division. Each individual plaintiff brings separate claims against New Paltz which are discussed infra. This case was tried to the Court pursuant to Fed.R.Civ.P. 52. The following are the Court’s findings of fact and conclusions of law. I. TITLE VII: GENERAL LEGAL STANDARDS A. Burden of Proof in a Class Action The plaintiff class claims that New Paltz treated women differently from men on the basis of sex. In order to prove such a claim of disparate treatment, plaintiff must prove discriminatory intent. See Melani v. New York City Bd. of Higher Education, 561 F.Supp. 769, 773 (S.D.N.Y.1983). Plaintiffs must demonstrate that “unlawful discrimination has been a regular procedure or policy followed by an em-ployer_” Int'l Bro. of Teamsters v. United States, 431 U.S. 324, 360, 97 S.Ct. 1843, 1867, 52 L.Ed.2d 396 (1977). The occurrence of sporadic or isolated discriminatory acts is not sufficient; gender discrimination must be New Paltz’ standard operating procedure. Id. at 336, 97 S.Ct. at 1855; Coser v. Moore, 739 F.2d 746, 749 (2d Cir.1984). A finding of classwide discrimination creates a rebuttable presumption in favor of individual claims of discrimination. Franks v. Bowman Transp. Co. Inc., 424 U.S. 747, 772, 96 S.Ct. 1251, 1268, 47 L.Ed.2d 444 (1976); Teamsters, 431 U.S. at 359, 97 S.Ct. at 1866. If the class claims are unsuccessful, individual claimants may still pursue their actions, but they must satisfy the burden of proof described in the next section. B. Burden of Proof in Individual Claims The Supreme Court has established a three step process for adjudicating individual Title VII claims. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802-804, 93 S.Ct. 1817, 1824-25, 36 L.Ed.2d 668 (1973). First, the plaintiff has the burden of proving a prima facie case of discrimination by a preponderance of the evidence. Id. at 802, 93 S.Ct. at 1824; Zahorik v. Cornell University, 729 F.2d 85, 92 (2d Cir.1984). The burden of production then shifts to the defendant “to articulate some legitimate, non-discriminatory reason for the employee’s rejection.” McDonnell Douglas, 411 U.S. at 802, 93 S.Ct. at 1824. If the defendant satisfies this burden, the plaintiff must show that the reason is merely a pretext for discrimination. Zahorik, 729 F.2d at 92. Although the burden of production shifts to the defendant in the second step, the burden of persuasion remains with the plaintiff throughout. Id. McDonnell Douglas sets forth four elements a plaintiff must prove to make a prima facie showing of race discrimination: 1) plaintiff belongs to a racial minority; 2) plaintiff applied and was qualified for a job for which the employer was seeking applicants; 3) despite plaintiff's qualifications he or she was rejected; and 4) the job remained open after the plaintiff was rejected and the employer continued to seek applicants with plaintiffs qualifications. McDonnell Douglas, 411 U.S. at 802, 93 S.Ct. at 1824. This test is flexible and should be tailored to a specific case. Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 253, n. 6, 101 S.Ct. 1089, 1094, n. 6, 67 L.Ed.2d 207 (1981). In fulfilling its burden of showing a legitimate reason for its action, the defendant need not prove absence of discrimination. Lieberman v. Gant, 630 F.2d 60, 65 (2d Cir.1980). Rather, the defendant need only show that it acted on a neutral basis. Id. The employer’s hiring practices need not be “rational, wise, or well-considered — only ... non-discriminatory.” Powell v. Syracuse University, 580 F.2d 1150, 1157 (2d Cir.), cert. denied, 439 U.S. 984, 99 S.Ct. 576, 58 L.Ed.2d 656 (1978). The evidence the employer presents should be objective. Sweeney v. SUNY Research Foundation, 711 F.2d 1179, 1185 (2d Cir.1983). Objective factors include evidence that a plaintiff’s qualifications do. not match those required for a job. Id. A defendant’s subjective evaluation of objective criteria also is acceptable, but should be specific and not speculative. Id. at 1185-86. A plaintiff can demonstrate that a neutral explanation is pretext “either directly by persuading the court that a discriminatory reason more likely motivated the employer or indirectly by showing that the employer’s proffered explanation is unworthy of credence.” Burdine, 450 U.S. at 256, 101 S.Ct. at 1095. Evidence of pretext includes discriminatory statements or admissions, the mix of the workforce, an atmosphere of discrimination, the employer’s general practices, comparative evidence, and statistics. Penk v. Oregon State Board of Higher Education, 816 F.2d 458, 462-63 (9th Cir.), cert. denied, — U.S. -, 108 S.Ct. 158, 98 L.Ed.2d 113 (1987). C. The Relationship Between the Class Claims and the Individual Claims Evidence of intentional discrimination against women in individual cases is relevant to the class claims of disparate treatment. It brings statistics to life, demonstrates how college policies affected the treatment of individual women, and is proof of the state of mind of college officials. Craik v. Minnesota State University Board, 731 F.2d 465, 471 (8th Cir.1984). Similarly, evidence of classwide discrimination is relevant to individual claims because it can demonstrate that a neutral explanation for a decision is pretextual. In short, all evidence of intentional discrimination is relevant to all claims and must be considered as a whole. II. STATISTICAL EVIDENCE OF CLASSWIDE DISCRIMINATION Plaintiffs assert that statistical evidence alone proves their claims of class-wide discrimination. Gross statistical disparities can be sufficient to prove a prima facie case of discrimination. Hazelwood School District v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977). Statistics are not irrefutable, however, and their usefulness depends on the facts and circumstances of a particular case. Id. Statistics used in a Title VII action against a university must account for the fact that a college hires specialists on a decentralized basis. Coser, 739 F.2d at 750. Plaintiffs’ statistical expert, Dr. Mary Gray, utilized a technique known as multiple regression analysis to determine whether differences in male and female salaries were based on different qualifications or other unexplained factors. See Melani, 561 F.Supp. at 774. A multiple regression analysis determines the influence that various factors (independent variables) have on an observed phenomena (the dependent variable). For each independent variable, the statistician calculates a coefficient, which is the effect the variable has on a dependent variable. The coefficient is derived from the data about all the faculty members, and measures, for example, the effect a year of service (the independent variable) has on salary (the dependent variable). A multiple regression analysis can determine the effect gender has on salary either by calculating a coefficient which measures the ef-feet of gender (either male or female), or by comparing a woman’s actual salary with her predicted salary. The salary is predicted by multiplying the coefficient for each independent variable by the value each variable assumes for the faculty member, and adding the results. Even if the results of the regression demonstrate that gender influences salary, this is not necessarily relevant. Rather, the difference between male and female salaries, measured by either the size of the gender coefficient or the difference between the actual and predicted salary, must be statistically significant. That is, the likelihood that the difference in salary is due to chance must be less than .05 percent. Melani, 561 F.Supp. 774. Stated another way, the odds must be less than one in twenty that the apparent difference is actually no difference for it to be statistically significant. This probability is also expressed in terms of “standard deviations”. A standard deviation is a measure of the distance on a graph between the probability of an observed result and zero. The higher the number of standard deviations, the further the result is from the expected result, and the less likely it is. This probability can be measured in two ways. The first, known as a two-tail test, measures the probability of the result falling on either side of a bell curve. The second, known as a one-tail test, measures the probability of the result falling on either the left or the right of the center. When measuring the results based on only one tail of the curve, the standard of statistical significance is 1.65 standard deviations. When using both tails, it is 1.96. A. Plaintiffs’ Statistical Studies 1. Plaintiffs’ main salary study Plaintiffs’ exhibit 882 presents the results of the regression analysis that Dr. Gray testified best represents the difference in salary between men and women at New Paltz. Dr. Gray measured salary differences for each year of the class period, 1973 to 1984. This study indicates that women earned from $1,036 to $2,277 less than their predicted salaries for the relevant years. The standard deviations associated with this ranged from — 2.72 to — 6.88. Underlying this study are a number of judgments. a. Use of rank variables Plaintiffs’ main study used the following independent variables: number of years of full-time teaching experience prior to hire at New Paltz; number of years teaching experience in academic rank at New Paltz; possession of a doctorate degree; number of years since obtaining the doctorate degree; field or discipline; number of publications; other experience prior to hire at New Paltz; and years of full-time high school teaching experience. Dr. Gray stated that three variables — pri- or academic rank, current rank, and years in current rank — may influence salary. Dr. Gray, however, did not include these variables in her main study for two reasons. First, current rank and years in rank are under the control of New Paltz, the institution accused of discriminating against women, and thus might be “tainted” by discrimination. Second, rank is not a qualification, but a result of qualifications. The variables she used measure qualifications. Including rank variables would “double-count” the effect of the qualifications and decrease the extent of the difference between men’s and women’s salaries. Dr. Gray submitted evidence of academic rank discrimination at New Paltz. First, Dr. Gray compared the number of men and women in the four academic ranks in academic years 1973 through 1984. (Pl.Exhs. 891 and 940) Her study indicates that women were disproportionately concentrated in the two lower ranks in each year of the study. This differential remained consistent when Dr. Gray accounted for possession of a Ph.D. Dr. Judith Stoikov, defendants’ statistical expert, criticized this evidence as merely descriptive, not demonstrating that the results were statistically significant. Dr. Stoikov also criticized the study because factors other than a Ph.D.— such as years at New Paltz, prior experience, and publications — determine rank. Second, Dr. Gray compared the rank at hire for men and women from 1973 to 1984. (Pl.Exhs. 889 and 952) The results indicate that New Paltz hired 34 women, 33 into the two lower ranks. New Paltz hired 80 men, 60 in the lower ranks. New Paltz thus hired a disproportionate number of women into the lower ranks. Dr. Gray divided the new hires into those with and without a Ph.D. Nobody without a Ph.D. was hired into the two higher ranks. A woman without a Ph.D. was more likely than a man without a Ph.D. to be hired into the lowest rank. Among those with a Ph.D., a woman was more likely than a man to be hired into the two lower ranks. On rebuttal, Dr. Gray presented a study of the statistical significance of these results. (PLExh. 952A) The probability that gender had no effect on initial hire for all new hires and for those with a Ph.D. was less than .05 percent. Dr. Stoikov testified that there were not enough observations to complete a statistical study of the rank at hire results reported in exhibit 952, with the exception of faculty hired without a Ph.D. For those faculty, she found that the differences in rank at hire were not statistically significant. Third, Dr. Gray compared the average years to promotion for men and women promoted from 1973-1984. (Pl.Exh. 895) She testified that the only promotion for which there was enough data to make a meaningful comparison was from assistant professor to associate professor. The average number of years to promotion for a woman was 8.55 years, and for a man, 7.00 years. Examining only those faculty members with a Ph.D., Dr. Gray found that the difference in years to promotion was 1.909 years. Dr. Stoikov criticized this study because it did not analyze differences in the rate of promotion. Finally, Dr. Gray compared the results of a regression analysis which excluded rank variables with a regression which included them. (Pl.Exhs. 104 and 1028) The difference between the average female underpayments in each study is statistically significant. That is, the amount of the underpayment absorbed by the rank variables, determined by the underpayment in a study which did not use the rank variables, is statistically significant. Dr. Gray asserts that since rank should be based on the same factors as salary, any statistically significant difference in the residuals between the rank and non-rank studies is due to discrimination in rank. Dr. Stoikov also criticized this conclusion on the ground that Dr. Gray's study did not show that rank is discriminatory, but only that men and women were unevenly distributed among rank. b. Department chair variable Dr. Gray testified that although New Paltz adds a stipend to the salary of the chairperson of a department, she did not include a variable to reflect this in her main salary study. (PLExh. 882) Her reason was that selection of department chairs was under the control of New Paltz, and tainted by discrimination. In support of this claim, Dr. Gray presented two exhibits which list the department chairs at New Paltz by gender. (PLExhs. 900 and 980) The total number of chairpersons was 53.5 men and 6.5 women. The total number of years served was 194.5 to 9.5. Dr. Gray testified that the probability that gender had no effect on this difference was less than 1 out of 1000 (PLExh. 1022). Dr. Gray also studied the distribution of department chair positions among faculty with more than 7 years of college teaching experience, since at New Paltz, most department chairs were tenured, and the waiting period for tenure was 7 years. The result was the same distribution of chairperson years between men and women as when she measured all faculty. The probability that gender had no effect was 5 out of 1000. Dr. Stoikov criticized Dr. Gray’s failure to account for department chair stipends in her main study. Dr. Stoikov testified that Dr. Gray’s proof that selection to departmental chair was discriminatory was flawed because the only factor it accounts for is whether the faculty member had seven years of experience, and selection to departmental chair depends on many other factors. Defendants also criticized Dr. Gray for failing to use a variable for prior service in an administrative position because she did not demonstrate that women were denied administrative positions on discriminatory grounds. Defendants assert that excluding these variables from the salary regression overestimates the differential between men and women, as both factors have a positive influence on salary (Def.Exh. S6-A, Table H) and more men than women were chairpersons and administrators. c. “Males only” regression Dr. Gray’s main study is a “males only” regression. She calculated the coefficient for each independent variable using only the data for male faculty members. Then she multiplied each coefficient by the corresponding value for a female faculty member. She added together all of these figures; the sum is the female faculty member’s predicted salary. That is, it is the salary she would be earning, based on all her relevant qualifications, if she were a man. Dr. Gray then subtracted the woman’s actual salary from her predicted salary. This was the woman’s “residual”, or the measure of the difference between her actual salary and the salary she would earn if she were a man. Dr. Gray repeated this process for all the class members, added all the residuals, divided by the total number of women, and derived the average female residual. This residual is the basis for calculating the degree of statistical significance. A males only regression is different from a total population regression. In such a regression, the characteristics of the entire population, male and female, are used to calculate the coefficients for each variable. A gender variable, either male or female, is added to the list of independent variables. The size of the gender coefficient is the measure of the effect of being male or female. For example, if the gender variable equals female, and the coefficient is — $100, the effect of being female on salary is — $100. The number of standard deviations, that is, the likelihood that gender does not influence salary, is based on the gender coefficient. Dr. Gray preferred the males only regression because the purpose of her study was to determine if New Paltz treated men differently from women. Dr. Gray testified that the males only regression, in predicting a female’s salary based on the value that the various independent variables have for a man, was superior in measuring what a woman would earn if she were paid the same as a man. The total population regression measures a woman’s salary against the salary for an average person. This, Dr. Gray testified, reduces the size of the coefficients by combining the male and female values, and in turn reduces the predicted salary differential between men and women. For example, if the coefficient for a Ph.D. is $1,000 for a male and $500 for a female, and there are an equal number of males and females, the Ph.D. coefficient is $750. Dr. Stoikov, on the other hand, testified that a males only regression is not as accurate as a total population regression because it only accounts for values existent in the male population. d. Excluding the nursing faculty Dr. Gray did not include members of the nursing faculty in her main study, even though nurses were in the liberal arts faculty and are members of the class. Dr. Gray excluded them from her study for two reasons. First, there were no male nursing faculty, and thus there were no males with whom to compare the females. Second, she determined that although the nurses were members of the faculty of liberal arts, they had different duties and received much higher average salaries than other women on the liberal arts faculty. (Pl.Exh. 894) New Paltz started its nursing program in 1981. There are only six nurses who are class members, and there were no more than four nurses in any year. Including the nurses in the regression thus has a minimal effect on the results. (Def.Exh. S(6), Table 20) Nevertheless, defendants argue that the nurses are members of the class, decisions about their salaries were made by the same people who made decisions about the other class members, and hence they should be included in the regressions. e. One-tail measurements Dr. Gray expressed the statistical significance of the results in her main study according to both a one-tail and a two-tail test. The one-tail results demonstrated a higher degree of probability that gender was the cause of the observed salary differentials. Dr. Gray preferred the one-tail test because all the results were negative, that is, all the female residuals fell under the left tail of the curve. 2. Plaintiffs' other current salary studies Although Dr. Gray was confident that her main salary study accurately reflected the different treatment of men and women at New Paltz, she completed other regressions which incorporated variables or employed methodologies which were not reflected in her main study. a. Total population regression In the second half of exhibit 882, Dr. Gray did a total population regression using the same variables as in the main study. She measured the net effect of being male. In each year, men had a higher salary than predicted. The number of standard deviations and the one and two-tail probabilities were similar to those in the main study, and all were statistically significant. b. Regressions including rank variables Dr. Gray completed five regressions which included rank variables. The first three regressions have two parts, a males only regression and a total population regression. The fourth and fifth are a males only and total population regression, respectively. The regressions include variables for current rank (Pl.Exh. 883); current rank and years in rank (Pl.Exh. 884); current rank, years in rank, and prior rank (Pl.Exh. 885); prior rank only (Pl.Exh. 989); and current rank with instructor treated as a separate rank, years in rank, and prior rank (Pl.Exh. 988). In the first three regressions, the current rank variables included only three ranks: assistant professor, associate professor, and professor. Dr. Gray testified that she combined the ranks of instructor with assistant professor for two reasons. First, there were very few male instructors, and in some years there were none. In light of this, Dr. Gray felt it would be inaccurate to predict salaries of the female instructors based on the characteristics of a small number of males. Second, as expressed earlier, supra, at II. A. 1. a., Dr. Gray prepared a study which she believed demonstrated that, based on their qualifications, a disproportionate number of women were hired as instructors. By combining the ranks, she hoped to control for some of this difference. Dr. Stoikov criticized this. She testified that the result of combining rank was that a woman instructor was predicted to have the salary of a male assistant professor. This overestimates the woman’s salary, and hence the differential. Examining the males only regressions in exhibits 883 through 885, and comparing them to plaintiffs’ main study (Pl.Exh. 882), yields the following observations. First, as each rank variable is added, the female residual and the degree of statistical significance consistently decrease. However, the residual remains statistically significant in a one-tail test for every year of the class period except 1984/85 in the test that includes all rank variables. The results are statistically significant under a two-tail test in all measures but three: the regression including current rank and years in rank in 1978/79 and 1984/85 (Pl.Exh. 884); and the regressions including all three variables in 1984/85 (Pl.Exh. 884). The total population regressions which correspond to the males only regressions in exhibits 883 through 885 show a similar consistent decrease in the net effect of being male and a decrease in the number of standard deviations. The total number of differences which are not statistically significant also decreases. The measurements which are not statistically significant under either a one or two-tail test occur in exhibit 883, 1984/85; exhibit 884, 1984/85; and exhibit 885, 1979/80 and 1984/85. Under two-tails, the differences are not statistically significant in 1978/79, 1979/80, and 1980/81 in exhibits 884 and 885. It is also worthwhile to compare the total population regression which includes all three rank variables (Pl.Exh. 885) with the regression which includes all three rank variables and treats the rank of instructor separately (Pl.Exh. 988). The net effect of being male decreases in every year except two in exhibit 988. The net effect is not statistically significant under either tail in four of the years, 1975/76, 1978/79, 1979/80, and 1984/85. It is not statistically significant under two tails only in 1983/84. The final rank regression Dr. Gray presented was a males only regression in which she added only prior rank to the variables she used in her main study. (PI. Exh. 989) Comparing the results of this regression with the main study indicates that prior rank does not account for much of the difference between male and female salaries. In 1982/83, prior rank increased the difference. c. Regressions including rank and department chair variables Dr. Gray prepared three males only regressions which consider the impact serving as a department chairperson has on salary. (PhExhs. 888, 889, and 890) These studies are all variations of the males only regression in exhibit 885, which includes variables for all three ranks, so comparisons with that study are appropriate. In exhibit 888, Dr. Gray simply added a variable for department chair. While the female residual remained negative, it decreased slightly in every year except 1983/84. The residual was not statistically significant in 1984/85 and 1978/79 under either tail, and under two tails alone in 1979/80. In exhibit 889, Dr. Gray did not add a department chair variable. Rather, she subtracted the department chair stipends from the salaries of department chairpersons. The average female residuals were generally slightly higher than in exhibit 888, but generally not as high as in exhibit 885. The only year in which the residuals were not statistically significant was in 1984/85. Finally, exhibit 890 combines the effects of exhibits 888 and 889 by adding a department chair variable and subtracting the department chair stipend. The results of this study were virtually identical to the results reported in exhibit 888. d. Regressions including nursing faculty Plaintiffs’ exhibit 898 is a total population regression which adds a nursing variable to the variables used in plaintiffs’ main study. It should thus be compared to the total population regression of plaintiffs’ main study. The male coefficients are positive in each year in exhibit 898, and are virtually identical to the male coefficients in the total population regression reported in the second part of exhibit 882. In plaintiffs’ exhibit 897, Dr. Gray adds a nursing variable to the results of the total population regression of exhibit 885, which includes all the rank variables. The results of the two regressions are virtually identical. 3. Salary at hire Dr. Gray also reported the results of a males only regression analysis, using the same variables as in plaintiffs’ main study, for faculty hired from 1973 to 1984. (PI. Exhs. 939) Females received a salary at hire that was an average of $535 less than men, a standard deviation of — 2.25. B. Defendants’ Salary Studies 1. “Validation” of the rank variable Dr. Stoikov attempted to “validate” the use of rank variables in her salary regressions by proving that rank at New Paltz was not discriminatory. She studied the initial rank of those hired between 1973 and 1984 and promotions between 1973 and 1984. a. Initial rank In Table 4 of defendant’s exhibit S(6), Dr. Gray determined the effect six variables had on rank at hire: possession of a Ph.D.; publications; years of prior college teaching; years of related research experience; years of other related experience; and pri- or rank. The results indicate that only 4.8 percent of those hired as instructors held a Ph.D. at hire, while 68.4 percent of those hired as assistant professors, and all hired as associate professors and professors, held a Ph.D. Similarly, rank at hire increased among those with one or more publications. The most influential experience variable, however, was prior college teaching: as years of prior college teaching experience increased, so did rank at hire. Finally, only one faculty member — a male —was hired at a rank lower than his prior rank. Dr. Stoikov broke down these factors by gender. She determined that a higher percentage of males than females had a Ph.D. and at least one publication at hire. Males also had more years of experience in all three categories. Finally, the prior rank of women was proportionately lower than that of men: nearly three-quarters of the women held the rank of instructor or below, while less than half of the men did. Conversely, more than half the men held the prior rank of assistant professor or higher compared to one-fourth of the women. In Table 6 of exhibit S(6), Dr. Stoikov presented the results of a study of the statistical significance of the influence of various factors on initial rank. Since the initial rank variable could only assume four values — instructor, assistant professor, associate professor, and professor — Dr. Stoi-kov employed a “maximum likelihood estimate” procedure known as logit or categorical modeling, which is designed for dealing with a dependent variable with a limited number of values. Dr. Stoikov also added the following variables to the test: years of hire; academic field; and gender equals female. The statistical analysis indicates a very high probability that a woman would be hired into the predicted rank. (Table 6, col. 5) Dr. Gray divided the results into five categories. The probability that a woman would be hired correctly into: 1) one of the four ranks was 63 percent; 2) one of three ranks (combining professor and associate professor) was 87 percent; 3) the rank of instructor as opposed to the other three ranks was 44 percent; 4) an entry level rank as opposed to a non-entry level rank was 100 percent; and 5) the rank of instructor as opposed to assistant professor was from 44 to 80 percent. Dr. Stoikov concluded from her studies that gender did not influence assignment to initial rank from 1973 to 1984. She testified that entry into the two lower ranks was a function of whether one had a Ph.D. or was about to get one, and that entry into the higher ranks was a function of years of college teaching experience and prior rank. b. Promotions In Table 10 of defendants’ exhibit S(6), Dr. Stoikov studied the probability that a female would be promoted compared to a man. She studied three different promotions: instructor to assistant professor, assistant to associate professor, and associate to professor. The factors she used in her regression analyses were years since Ph.D., years in rank, department group, and gender equals female. Dr. Stoikov concluded that gender did not have an influence on promotion. The most significant influences on promotion from instructor to assistant professor were years since Ph.D. and years in rank. Years in rank was the only factor which had a significant influence on promotion from assistant to associate professor, and none of the measured factors influenced promotion from associate professor to professor. Dr. Gray criticized Dr. Stoikov’s attempt to “validate” rank on two grounds. First, Dr. Stoikov did not account for people placed in rank before 1973. Since half of the class members were hired before 1973, Dr. Stoikov’s studies do not account for whether these women were hired into the correct rank. Thus, even if the waiting time for promotion was equal, they still might not be in the proper rank. Second, plaintiffs assert that the population Dr. Stoikov relied on was not a large enough population to produce reliable results in a categorical modeling or logit regression. 2. Salary decisions made between 1973 and 1984 a. Initial salary Dr. Stoikov did a regression analysis of the initial salaries of all New Paltz faculty hired between 1973 and 1984. (Def.Exh. S(6), Table 11) She controlled for the same factors as in her study of initial rank, and did not include any rank variables. The results, reported at Part I of Table 11, indicate that the difference between male and female starting salary is not statistically significant. b. Rates of salary increases In Part II of Table 11, Dr. Stoikov measured the annual rates of salary increase. Dr. Stoikov found that the net effect of being female was positive. Females’ salaries increased faster than similarly qualified males. c.Share of discretionary increases Dr. Stoikov also studied whether women received their “fair share” of discretionary pay from 1973 to 1984. She found that with the exception of 1977, when the percentage of women was actually higher, the percentage of female faculty receiving awards was equal to the percentage of female faculty. (Def.Exh. S(6) Table 15) Dr. Stoikov also found that the amount of the discretionary pay women received was proportionately equal. In some years women received more than their fair amount, and in others they did not. In no year was the difference statistically significant. 3. Current salary studies: 1973-1984 Dr. Stoikov did a multiple regression analysis of salaries paid to all full-time faculty in academic rank at New Paltz from 1973 to 1984. (Def.Exh. S(6), Tables 17 and 18) Dr. Gray included the nursing faculty in her study. The variables she controlled for were: 1) year of hire (if in 1973 or later) or years at New Paltz (if hired before 1973); 2) years of prior college teaching; 3) years of other related prior experience; 4) prior rank; 5) years since Ph.D.; 6) current rank; 7) years in current rank; 8) department group; 9) prior administrative position; 10) department chair; and 11) gender equals female. Dr. Stoikov also deducted the chair stipend from the appropriate salaries. Dr. Stoikov’s regressions are total population regressions. In Table 17, Dr. Stoikov reports the female coefficient — the effect that being female has on expected salary — in terms of dollars. In eight years the effect of being female was negative, in four years it was positive. The range in the negative years was — $15 to — $325, and the range in the positive years was $33 to $281. In none of the years was the effect of being female statistically significant under a two-tail test. In Table 18, Dr. Stoikov reported the effect of being female in natural logarithms, that is, the percentage of total salary. In this study, the effect of being female was negative in 11 of 12 years. The range of the effect was from — 2.6 to — 0.1 percent. The effect of being female was not statistically significant under a two-tail test. Dr. Gray’s most persuasive criticism of Dr. Stoikov’s salary was her use of a year of hire variable. Dr. Gray presented evidence that the variable was extremely volatile. The effect on salary of being hired in 1973, for example, was drastically different for each year in the period, and hence unreliable. Dr. Gray also presented evidence that the year of hire variable did not measure what it purported to measure— that bad budgetary years had a negative influence on salary. She showed that being hired in 1979, a year defendants contend was a poor budgetary year, had a statistically significant effect on salary in only one year, and in that year it was positive. 4. Salaries in 1971-1972 Dr. Stoikov also prepared a regression analysis of salaries paid to all full-time liberal arts faculty from 1971 to 1972. (Def.Exh. S(6), Table 23) She used the same regression technique as in Tables 17 and 18 but used different variables because she did not have as much information. She used years since Ph.D., year of hire (if hired after 1963), years at New Paltz (if hired before 1964), current rank, years in rank, department group, and gender. In 1971, the female coefficient was — $472. This was associated with — 1.7 standard deviations, which is not statistically significant under a two-tail test. In 1972, the effect of being female decreased to — $263, and the number of standard deviations decreased to — 0.99. C. Conclusions from the Statistical Evidence 1. Including rank variables in the regressions Rank is an improper variable if there is evidence that rank is discriminatory, and appropriate when there is evidence that it is not. See Valentino v. United States Parcel Service, 674 F.2d 56, 72-73, n. 30 (D.C.Cir.1982) (absent clear evidence that rank is not discriminatory, there is no assurance that rank is an appropriate variable); Melani, 561 F.Supp. at 782-83; Presseisen v. Swarthmore College, 442 F.Supp. 593, 614, 619 (E.D.Pa.1977), aff'd, 582 F.2d 1275 (3d Cir.1978) (inclusion of rank variable appropriate when evidence showed no discrimination with respect to promotion). The plaintiffs have failed to prove that rank at New Paltz was discriminatory. Dr. Gray’s studies of rank (Pl.Exhs. 891 and 940), rank at hire (Pl.Exhs. 899 and 952), and waiting time for promotion (PI. Exh. 895), were mere compilations of data which neither accounted for factors (with the exception of possession of a Ph.D.) which Dr. Stoikov’s studies persuasively showed lead to assignment to rank or promotion, nor demonstrated that observed differences were statistically significant. The exhibits which Dr. Gray introduced to show statistical significance (Pl.Exh. 891A which was not admitted into evidence and Pl.Exh. 952A which was) were unpersuasive. The only factor they measured was possession of a Ph.D. Dr. Gray’s comparison of the regression which included rank variables with one that did not (Pl.Exhs. 1021 and 1028) proved only that rank had a statistically significant influence on salary, and that men and women were unequally distributed across the ranks. It did not show that the ranks were unfairly distributed or discriminatory. Finally, as is described more fully below, there is no evidence of discrimination in rank in individual cases either. The defendants’ statistical evidence, on the other hand, demonstrated that there was no discrimination in either placement into initial rank or promotion between 1973 and 1984. Plaintiffs’ criticism of Dr. Stoi-kov’s statistical study on the ground that it did not include a large enough population to be reliable is not well founded. Plaintiffs failed to prove to the Court that Dr. Stoikov improperly relied on the logit and categorical modeling regression techniques. See Craik, 731 F.2d at 476 n. 14. The plaintiffs’ second criticism of Dr. Stoikov’s rank validation—that she did not account for faculty hired before 1973—is a more significant criticism but is nevertheless unpersuasive. Dr. Stoikov’s studies do explain the rank at hire for at least half the class members, and explain the promotions for all class members during the relevant period. 2. Other variables A regression analysis should account for holding a departmental chair in the absence of evidence that selection to departmental chairs at New Paltz was discriminatory, as it increases a faculty member’s salary. Plaintiffs failed to prove that selection to a departmental chair at New Paltz was discriminatory. There is no evidence that New Paltz deprived women of departmental chairs on discriminatory grounds. Furthermore, Dr. Gray’s studies are not persuasive. Exhibits 900 and 980 merely show that more men than women held departmental chairs. They do not show that this distribution was discriminatory. Dr. Gray’s attempt to prove that the distribution was discriminatory (Pl.Exh. 1022) is also unpersuasive because it does not account for any of the factors that influence selection to departmental chairs. The study in which Dr. Gray controls for faculty with at least seven years of prior college teaching is similarly unpersuasive because it accounts for only one factor. The defendants also showed that holding a prior administrative position had a positive influence on salary at New Paltz. It is thus also appropriate to use this variable absent some showing of discrimination. Plaintiffs did prove that all prior administrators were men, but they did not show that assignment to administrative rank was discriminatory. Finally, the Court finds that the year of hire variable is, as plaintiffs demonstrated, too volatile to be reliable. It is also not particularly helpful. As plaintiffs have shown, it does not indicate what it is supposed to—that salaries decreased in bad .budgetary years. 3. Males only or total population regression The males only regression predicts how women would be treated if they were treated the same as men, but does not account for factors which males do not possess. The total population regression accounts for factors not observed among males, but tends to underestimate any underpayment to women. The Court will view a males only regression from the perspective that it overestimates male/female differences to some degree, and will assume that a total population regression underestimates them. 4. Means of expressing probabilities: one-tail or two Dr. Gray’s position is that since all male coefficients and female residuals fell under either the positive or negative side of the tail of the bell curve, the probabilities that the results were due to chance should be expressed according to a one-tail test of probability. See Brunet v. City of Columbus, 642 F.Supp. 1214, 1230 (S.D.Ohio 1986), appeal dismissed, 826 F.2d 1062 (6th Cir.1987). Defendant’s criticism of a one-tail test is also compelling: since under a one-tail test 1.64 standard deviations equal the statistically significant probability level of .05 percent, while 1.96 standard deviations are required under the two-tailed test, the one-tail test favors the plaintiffs because it requires them to show a smaller difference in treatment between men and women. See Craik, 731 F.2d at 475-76, n. 13. This conflict provides the Court an opportunity to review the standards for determining the legal significance of statistical disparities. The Supreme Court has recognized that 2 or 3 standard deviations is generally sufficient to render data suspect to a social scientist. Castaneda v. Partida, 430 U.S. 482, 496 n. 17, 97 S.Ct. 1272, 1281 n. 17, 51 L.Ed.2d 498 (1977). Nevertheless, there is no threshold of statistical significance above which discriminatory intent is shown as a matter of law, and below which it is not sufficient. See EEOC v. American National Bank, 652 F.2d 1176, 1192 (4th Cir.1981), cert. denied, 459 U.S. 923, 103 S.Ct. 235, 74 L.Ed.2d 186 (1982); Gay v. Waiters’ and Dairy Lunchmen’s Union, 694 F.2d 531, 551 (9th Cir.1982); Coser, 739 F.2d at 754 n. 3 (5 percent chance of probability has no talismanic significance). A court’s approach to statistical disparities should be flexible, inferring from them what it can about the existence of discriminatory intent. When reviewing statistics which show relatively small disparities, such as one to three standard deviations, a court should be cautious about drawing an inference of discrimination. Coates v. Johnson and Johnson, 756 F.2d 524, 547 n. 22 (7th Cir.1985); EEOC v. American National Bank, 652 F.2d at 1192. When the standard deviations are small, the court’s conclusions about the anecdotal evidence are particularly important to the outcome of the case. See Segar v. Smith, 738 F.2d 1249, 1278 (D.C.Cir.1984), cert. denied, 471 U.S. 1115, 105 S.Ct. 2357, 86 L.Ed.2d 258 (1985). The small difference between a one-tail and two-tail test of probability is not relevant. The Court will not treat 1.96 standard deviation as the dividing point between valid and invalid claims. Rather, the Court will examine the statistical significance of the results under both one and two tails and from that infer what it can about the existence of discrimination against women at New Paltz. See Chang v. University of Rhode Island, 606 F.Supp. 1161, 1205 (D.R.I.1985) (comparing one-tail and two-tail test results). 5. Accuracy of data During the course of the trial, the parties disputed the admissibility of defendants’ statistical reports because defendants’ expert changed the data for various faculty members after she had provided it to plaintiffs’ expert. The defendants assert that the plaintiffs’ main studies, which do not use the changed data, are inaccurate. Only eleven pieces of data out of the thousands involved in this case are at issue. Both experts submitted studies at trial which used the data of the other expert. (PI. Exhs. 985, 986, and 987; Def.Exhs. S(6)(B), Tables 26 and 27) The results indicated that the data differences do not make a significant difference in either study. The slight differences in each database do not affect the weight the Court accords to either experts’ studies. 6. Conclusions from plaintiffs’ statistics The Court makes the following conclusions about plaintiffs’ statistical studies. First, plaintiffs’ regression analy-ses which do not control for the variables of rank, current rank, and prior rank are not of sufficient weight to demonstrate discrimination. This includes plaintiffs’ main study, exhibit 882. Plaintiffs’ studies which do not include all three rank variables are not sufficient either. (Pl.Exhs. 883, 884, and 989) This also includes plaintiffs’ study of salary at hire, which does not control for prior rank. (Pl.Exh. 939) Second, plaintiffs’ rank regressions which also control for service as department chair (Pl.Exhs. 888, 889 and 890) are more persuasive than the rank regression that does not control for department chair. (Pl.Exh. 888) Third, even those studies that do control for department chair are slightly less persuasive because they do not control for prior administrative rank. Fourth, plaintiffs’ males only regressions are acceptable, even though they exclude the nursing faculty and instructors, with the caveat that they slightly overestimate the difference in treatment. This leaves plaintiffs’ exhibits 888, 889 and 890 as their most persuasive studies. These control for education, experience, current rank, years in rank, prior rank, and department chair. In all three studies, the average female residuals are negative. The following table summarizes the range in which the standard deviations in each exhibit fall. Total number of standard deviation measures which fall between: 0-1 1-2 2-3 3-4 4-5 5-6 6+ Ex. 888 1 2 3 2 3 0 0 Ex. 889 0 2 3 2 3 1 0 Ex. 890 1 3 2 3 2 0 0 The residuals are thus generally associated with standard deviations between one and five. Two residuals fall below one standard deviation, one residual is greater than five standard deviations, and none are higher than six. These deviations alone, particularly in light of the caveats described above, are not sufficiently high to support a prima facie claim of salary discrimination. On the other hand, the statistics do not preclude an ultimate finding that plaintiffs have made a prima facie showing. The Court will thus consider these statistics in conjunction with other anecdotal evidence of discrimination. 7. Conclusions from defendants’ statistics The Court makes the following conclusions about defendants’ main study of current salary, at Tables 17 and 18 of exhibit (S)(6). First, the volatility and questionable reliability of the year of hire variables renders the results somewhat questionable. Second, as a total population regression, it tends to underestimate the male/female disparities. Defendants’ current salary studies demonstrate a much lesser degree of discrimination against women than do the plaintiffs’. In light of the Court’s finding that the defendants’ total population regressions underestimate the degree of salary differential between men and women at New Paltz, however, the residuals are not low enough that they alone rebut a finding of class-wide discrimination. They do, however, require the Court to scrutinize plaintiffs’ statistical studies and other evidence very carefully. Defendants’ studies of salary decisions made between 1973 and 1984 (Exhibit S(6), Tables 11 and 15), are of limited usefulness in assessing salary discrimination because they do not account for the residual of any pre-1973 discrimination. See Bazemore v. Friday, 478 U.S. 385, 106 S.Ct. 3000, 3006-07, 92 L.Ed.2d 315 (1986). These studies are relevant, however, in assessing the state of mind of the New Paltz Administration from 1973-1984, which is the class liability period. They demonstrate that the Administration did not discriminate against women in initial salary, rate of increase, or distribution of discretionary money. III. EVIDENCE OF COLLEGE-WIDE DISCRIMINATION AGAINST WOMEN A. Affirmative Action Program The existence of an affirmative action program “is the antithesis of a pattern and practice of discrimination based on sex.” Coser, 739 F.2d at 751. Evidence of good faith adherence to an affirmative action program is relevant to the issue of discriminatory intent. Craik, 731 F.2d at 472. The absence of a plan does not prove class-wide discrimination, but indicates that heightened scrutiny of university employment practices is appropriate. Chang, 606 F.Supp. at 1183-84. Plaintiffs attempted to prove that New Paltz did not have a viable affirmative action program. 1. The New Paltz Affirmative Action Plan Lorraine Bagley’s first responsibility as New Paltz’ first affirmative action officer in 1974 was to develop an affirmative action plan. Bagley completed the plan in 1975. (Pl.Exh. 151) The plan required the affirmative action officer to review all faculty employment decisions, including hiring, contract renewal, tenure, promotion, and salary. The plan also implemented a new search procedure for all faculty searches. It required search committees to place advertisements in journals with wide circulation among minority and female teachers and students, and to submit all pertinent information about the search to the affirmative action officer. The affirmative action officer had to approve the procedures before a new faculty member could be hired. The plan also stated that New Paltz had completed a study of salary inequity among the faculty which determined that some salaries were inequitably low, and that this was not due to discrimination, but to the poor condition of the budget in the year the faculty member had been hired. The 1975 plan did not include goals and timetables for the hiring of female faculty. Nor did it include an underutilization analysis, which compares the percentage of women on the faculty with the percentage of women holding a doctorate nationally, and is the basis for developing goals and timetables. Bagley testified that she did not prepare an underutilization analysis because the only data available was a tabulation of the percentage of women with doctorates. She needed data which divided the women by discipline, such as history, math, and chemistry, and even further into sub-disciplines such as Asian history and Medieval history. In 1979, the University of Colorado published data which satisfied Bagley. She did an underutilization analysis, and found that with one exception, none of the departments at New Paltz underutilized women. She thus did not prepare goals or timetables because she did not believe they were necessary. Bagley did not save this analysis. Bagley’s handling of the affirmative action plan contravened SUNY policies in two ways. First, SUNY required all its campuses to set goals and timetables. Second, SUNY required each campus to update its plan annually, and Bagley never updated the 1975 plan. (Pl.Exh. 16) She testified that she was too busy monitoring employment decisions to do so. In 1980, New Paltz appointed Margaret Wade as affirmative action officer, and she updated the plan in 1981. (Pl.Exh. 6) This updated plan contained an underutilization analysis, goals and timetable. New Paltz underutilized women in 6 out of 30 departments, and set hiring goals accordingly. (Exh. 16, p. 53) The record does not indicate whether New Paltz achieved these goals. Nor does it disclose New Paltz’ efforts, other than regular affirmative action procedures, to achieve these goals. 2.Part-time affirmative action officer Bagley had been a full-time affirmative action officer, but New Paltz appointed Wade on a part-time basis. Prior to her appointment, Wade was an assistant professor in New Paltz’ political science department. After her appointment, she remained in the department on a half-time basis. This arrangement benefited both Wade and New Paltz: it allowed Wade to continue teaching but stopped her tenure clock while she earned a Ph.D. New Paltz eliminated Bagley’s $17,000 per year salary and replaced it with an annual stipend of $1,000 it paid to Wade. Wade worked as affirmative action officer approximately 25 hours per week. She spent less time monitoring searches than did Bagley, but her other responsibilities were the same. 3.Monitoring employment decisions The record indicates that the affirmative action office was closely involved in faculty employment decisions. Whenever an opening arose, the department notified the affirmative action officer. She informed the department chair of the requirements for the search, and insured that the search was done according to the affirmative action guidelines. In 1983, this system was revised slightly. (Def.Exh. H(7)(b)) The affirmative action officer placed adverise-ments and received resumes herself. The affirmative action officer also monitored promotions, tenure decisions, contract renewals, and discretionary salary increases. Bagley testified that the Administration gave her lists of all faculty it considered each year. She reviewed decisions which she felt might have been unfair by comparing the personnel files of unsuccessful applicants with successful applicants. She then made any objections to the academic vice-president. If he did not concur, they took the dispute to the president, who made the ultimate decision. 4.Conclusions New Paltz’ affirmative action program was relatively weak. It did not include hiring goals and timetables until 1981. Although some of this delay might be excused by the absence of refined data, Bagley could have done a less sophisticated study and developed rough goals and timetables. New Paltz also reduced the affirmative action officer’s job to part-time in 1980 without a corresponding reduction in workload. On the other hand, the program had some strengths. The affirmative action officer discussed all employment decisions with the academic vice-president or president and had to approve all initial hires. In summary, New Paltz’ uneven commitment to affirmative action requires the Court to scrutinize employment practices at New Paltz with great care. B. Salary Equalization 1. Salaries at New Paltz: General Each faculty member at New Paltz received an across-the-board salary increase each year. The union representing all the faculty in the State University of New York (“SUNY”) system negotiated this increase with SUNY. New Paltz was not involved in these negotiations. Each faculty member was also eligible for a discretionary salary increase on an individual basis. SUNY and the union also negotiated the amount of discretionary money available to each campus. New Paltz used a rather complex system for distributing discretionary pay increases. Many different parties made recommendations, including the academic vice-president and the Central Committee on Promotion and Salary. The final distribution of the money, however, was at the discretion of the president. New Paltz divided its discretionary salary increases into three broad categories: inequity, merit, and promotional. In the earlier years of the liability period, New Paltz also used discretionary money to pay chairpersons’ stipends. 2. Methods for distributing inequity increases New Paltz granted an inequity increase to a faculty member whose salary it determined as inequitable. New Paltz defined an inequitable salary as one which was lower than a comparable faculty member. The cause of the inequity could be anything, including discrimination. During trial, and in the relevant documents, New Paltz asserted that inequitable salaries were due to year of hire: a faculty member hired in a bad fiscal year would generally receive a lower salary than one hired in a good fiscal year. There was no basis in the record for this explanation, and as described earlier, the statistics do not bear this out either. During the years in question, New Paltz employed various means for distributing inequity increases. a. Academic years 1973-1975: mandatory minimum salaries In these three years, New Paltz set minimum salaries for faculty of the same rank who had approximately the same number of years in that rank. Those faculty furthest below these salaries were eligible for inequity increases. (Pl.Exhs. 130, 174, and 280) The amount and percentage of awards in the four different categories in the years 1973 to 1975 are as follows: (Pl.Exh. 280) Year Promotion Merit Inequity Dept. Chair 1973/74 $ 6,400/8% $11,975/15% $61,190/77% 1974/75 $ 9,100/18% $27,200/54% $ 8,864/18% $5,285/10% 1975/76 $10,500/15% $43,300/61% $17,400/24% Totals $26,000/12.9% $82,475/40.9% $87,454/43.4% $5,285/2.6% In two of the three years New Paltz awarded more money for merit than inequity, but overall, New Paltz awarded more inequity than merit money. As plaintiff’s expert testified, the minimum salary system has a major weakness as a remedy for salary inequities. Even if New Paltz increased a woman’s salary to the minimum, her salary might still be below a comparably qualified male’s salary. b. 1976: The Kent State Salary Study Bagley and the New Paltz Administration were unsatisfied with the minimum salary system because it did not account for factors such as date of hire, rank at hire, promotion, and publications. (Pl.Exh. 130) Bagley recommended that New Paltz use a newly developed method, described in the Kent State Salary Study. (Pl.Exh. 780) The Kent State method was a multiple regression analysis which controlled for rank, years of service, and median years for all faculty at a given rank. (Pl.Exh. 154) Bag-ley used the Kent State Study to calculate the predicted salary for each faculty member at New Paltz. However, SUNY then informed New Paltz that due to budgetary constraints, no discretionary money would be available that year. New Paltz never actually applied the Kent State Salary Study. c. 1977-1979: The AAUP Kit From 1977 to 1979, New Paltz employed a new system: the Higher Education Salary Evaluation Kit published by the American Association of University Professors, (the “AAUP Kit”) (PLExh. 155) To calculate predicted salaries for women, the AAUP Kit recommended a males only regression which controlled for year of birth, highest degree, years since highest degree, and department. It recommended against using a rank variable. The AAUP Kit recommended that a college investigate the circumstances of all women whose salary fell below the predicted level. Bagley’s application of the AAUP Kit at New Paltz deviated from the AAUP recommendations in two ways. First, she included a rank variable in the study. Second, she investigated only those women whose salaries were at least .5 standard deviations below the norm in 1977 and one standard deviation below in 1979. (Pl.Exhs. 158 and 777) d. 1980, 1981, and 1982 academic years: no method In the 1980 school year, New Paltz decided that it would no longer employ a formal method for determining salary inequity. New Paltz determ