Discrimination and qualified supply 53

faculty rank, by academic department, or by individual institution.

Essentially the Scott study employs a complex statistical analysis (multiple-regression analysis) for arriving at predicted salaries for men and women on the basis of the data. The analysis involves taking into account the influence of 30 variables -personal characteristics, education, experience, and institution—that are associated with differences in faculty members' salaries. Among the 30 predictor variables are age, highest academic degree, year degree was obtained, prestige of granting institution, years employed in higher education, years employed in present institution, scheduled hours of teaching per week, number of published articles, and number of published books.

A predicted average salary was calculated after statistically controlling for all 30 predictor variables in a multiple-regression equation. A separate analysis is made for men and for women, since some of the coefficients for particular variables are significantly different for men and women. For example, advanced degrees are calculated to be much more significant salarywise for men than for women; the coefficients show that, for men, salaries vary directly with the quality of the employing institution within each category of institution, and not for women; but the effect of the number of publications on salary was determined to be about the same for both sexes (Carnegie Commission, 1973b, pp. 206-207).36

On the basis of the female regression equation, the predicted average salary for male faculty members for all institutions is found to be below the average of their actual salaries by $2,264 a year; for Research Universities I that predicted-actual differential is $2,729. On the basis of the male regression equation, the predicted average for female faculty in all institutions is

Granting Universities I and II are three categories that awarded at least 50, 40, and 10 Ph.D's, respectively, in 1969-70, and in the case of Research Universities II are on the list of the 100 leading institutions in terms of federal financial support. The three categories total 121 institutions. Comprehensive Universities and Colleges I and II are 453 institutions that offer liberal arts programs, have one or more professional or occupational programs, and may offer master's degrees. For more detail on these categories see Carnegie Commission (1973a).

36 It is interesting to observe that the Astin and Bayer study (1973, pp. 350-351), using the same 1969 data and similar statistical methods, seems to contradict these conclusions, with respect to the significance of the three variables as stated in the text.

Antibias regulation of universities 54

$1,407 above the actual salary; for Research Universities I, the predicted average for women is $2,009 above the actual average. The predicted-actual differences are largest in dollar terms for Research Universities I of any of the six categories of institutions and are smallest for Comprehensive Universities and Colleges. Indeed, for Research Universities I for the biological and physical sciences, the difference between the predicted and actual average salaries is most surprising. In those sciences, a great majority of the male faculty received salaries in 1969 that ranged roughly from $1,000 to $7,500 above the predicted salaries on the basis of the female regression equation, and slightly under 10 percent of the male faculty in those sciences received salaries $10,000. to $13,000 above the predicted amounts (Carnegie Commission, 1973b, p. 116 and Chart 14, p. 117). In considering the magnitude of those differentials, one should bear in mind the average salary figures for university faculty in the 1968-69 salary survey of the American Association of University Professors; they ranged from an average of $10,534 for assistant professors to an average of $17,000 for full professors ("The Threat of Inflationary Erosion," 1969, Appendix Table 5, p. 208).

Questions can be raised about the validity of the predicted salaries themselves, and the soundness of their use for assessing actual responsibility for discrimination. Those aspects will be considered after attention is given to the Astin-Bayer study that uses the same basic data and similar methods of analysis.

Before considering the Astin-Bayer study it should be mentioned that Scott also compared differences in faculty salaries associated with race by a similar method. There was a slight tendency for the predicted average salary of black male faculty to be above the actual average, on the basis of the equation for all men, and a slight tendency for the predicted average salary of black women to be slightly below the average actual salary for female faculty, on the basis of the equation for all women. But the differences are not statistically significant (Carnegie Commission, 1973b, pp. 209 and 224).

The Astin-Bayer study uses multiple-regression analysis to predict salaries of male faculty compared with female faculty and also to make similar predictions with respect to academic rank and tenure. Instead of using the regression equation of the opposite sex, Astin and Bayer apply to the women's data the regression weights of the predictor variables calculated from the men's sample.

Discrimination and qualified supply 55

The difference between actual average salary and predicted average salary for women for the whole sample is $1,040– actual salary averaged that much less than predicted salary (Astin & Bayer, 1973, p. 353). The authors say that it is a minimum figure because the regression weights of the predictor variables for the male sample are used, and the result does not include salary discrimination attributable to discrimination in rank. Astin and Bayer do not provide breakdowns by disciplinary areas or type of institution.

With respect to both the Scott and Astin-Bayer studies it should be pointed out that the use of large numbers of predictor variables as a basis for judging the existence and extent of sex discrimination in salary is subject to serious weaknesses. Only quantifiable aspects of each variable are included in the analysis. Unmeasurable, quality aspects of teaching, research and publications, and administrative service, although usually important in determining an individual's "productivity," tend to be neglected. Scheduled teaching hours are not all equal, as Scott clearly recognizes. Successful teaching of advanced courses and lecturing to large numbers of undergraduates in a rapidly changing field are more demanding in terms of scarce competencies than teaching sections in an elementary course, year after year. The former is more likely to be done by men, the latter by women in some departments.

Qualitative differences that affect productivity can be illustrated by examples. In many universities, a significant part of the instruction in courses in non-Western languages is given by native-speaking persons, mostly women. Some of them may ultimately earn a Ph.D. degree but still be qualified mostly for language teaching and not for advanced courses in literature or history. Their teaching per hour cannot be considered equal to that of a distinguished professor in the literature or history of the country where that language is the native tongue. Also Astin and Bayer make no allowance for part-time as distinguished from full-time teaching although, for reasons indicated in the JohnsonStafford paper, part-time service may have important implications for salary in future years.37

37 Briefly, the reason is that inclusion in the analysis of part-time service years as years of experience can be misleading if the rest of the person's time while on part-time teaching was occupied with nonprofessional activities, which means a decline or interruption in the development of professional skills and reputation that influence "productivity" and earnings.

Antibias regulation of universities 56

Furthermore, it is noteworthy that neither Scott nor Astin-Bayer compare the salaries of single women and male faculty. As already explained, the pattern of advancement in rank and salary for single women resembles that for men and presumably would show little sex differences in salary.

Certainly, one cannot assume that 30 predictor variables can serve to predict the average salary that women would be paid if there were no sex discrimination in salary determination on the part of institutions of higher education. The difference between the predicted average salary and the actual salary of female faculty in the two studies is in the nature of a residual still to be explained. At least part of that residual could be due to weaknesses in the predictor variables or to sex differences in productivity factors (e.g., career motivation, ability to perform demanding teaching assignments, standing in the discipline and the profession) not taken into account in the analysis.

It should be noted that analyses of faculty salaries by sex made in individual major universities, some in field investigations under the Federal Equal Pay Act, generally do not find a “pervasive pattern" of lower salaries for female faculty of the magnitude indicated by the Scott and Astin-Bayer calculations and especially by Scott's calculations for biological and physical sciences in major universities, where laboratory research contributions (often with graduate students) are such an important factor in academic reputation and in faculty salary determination. Faculty salary studies by individual institution (microeconomic case studies)38 are likely to provide much more reliable results than macro-type analyses based on the use of many predictor variables.

The Johnson-Stafford study analyzes factors that may explain much of the differential in faculty salaries between men and women. Their analysis is based mainly on data from the National Register of Scientific and Technical Personnel for 1970.39 Six disciplines are chosen for particular examination. They are anthropology, biology, economics, mathematics, physics, and sociology.

Johnson and Stafford point out that salary data for the first

38 For such a microeconomic case study, see Malkiel and Malkiel (1973).

39 Source of data: Tapes of the National Register of Scientific and Technical Personnel, National Science Foundation.

Discrimination and qualified supply 57

academic appointment after completing the Ph.D. degree show comparatively small differences by sex in most of the disciplines. At that point, the average differential between men's and women's salaries in academia ranged from women's salaries 4 to 8 percent lower in five disciplines to 12 percent lower in biology.

For all six disciplines, the percentage differential between male and female salaries widens during the first 20 years of experience after acquiring the Ph.D. degree. During those 20 years, the sex differential widened for the six disciplines as follows: anthropology, 8 to 23 percent; biology, 11 to 26 percent; economics, 5 to 15 percent; mathematics, 6 to 26 percent; physics, 6 to 22 percent; and sociology, 4 to 14 percent. The salary differential widened most rapidly between the fifth and fifteenth years of postdoctoral and professional experience. Generally, during those 10 years the women were between 35 and 45 years of age, when they were most likely to have heavy household and child-care responsibilities.

Johnson and Stafford also made an analysis of salaries and experience of faculty members who, in 1972, were assistant, associate, and full professors at Michigan State University in the six disciplines. They found that the starting salaries for the female faculty in those disciplines averaged only 3 percent lower than corresponding salaries for male faculty, and that figure does not have much statistical significance. At 15 years of reported academic experience, the differential had grown to 20 percent, and thereafter it declined somewhat with more years of experience above 15.

The National Register data also show a narrowing of the sex differential in half of the six disciplines after a certain period of experience. Classifying experience by five-year intervals up to 30 years, the sex salary-differential is largest in percentage terms at 20 years for economics and physics, at 25 years for sociology, and at 30 years for anthropology, biology, and mathematics. In other words, in the first three disciplines the sex differential for salaries narrowed somewhat (1, 2, or 3 percentage points) after 20 or 25 years of experience; in the other three disciplines the sex differential continued to enlarge a few percentage points up to the thirtieth year of experience, the final year included in the analysis.

The changes in the size of the sex salary-differential over time are explained by Johnson and Stafford largely in terms of dif