|
4. SalariesSince salaries are tied to ranks, it is not surprising that for women both measures fall behind those for men. This is especially true in academia, although Zuckerman (1987), in commenting on the 1981 figures for science and engineering fields, suggests that the salary differences between men and women are largest at the full professor level, less so at the associate professor level, and are not present at the assistant professor level. Unfortunately, this optimism seems perhaps premature when one notes the figures of Table 5, which provides the percentage (of the men's average salaries) that the women's average salary falls short for ranks in the research universities for the 1990-1991 academic year. Women at public institutions seem to fare better than their colleagues at private or church related institutions. Comparable tables for the 1985-1986 and 1988-1989 salaries give essentially the same figures with very little change in these percentages over the years. New assistant professors (full professors) in the mathematical sciences received a salary at 92.7% (98.8%) of that averaged over all fields. These findings are not inconsistent with those of Vetter (1987), who conducted a study of women's salaries as a percentage of men's salaries for doctorates in the mathematical sciences. The study found that this percentage declined from a high of 87.8% in 1973 to a low of 81.3% in 1979, increased again to 87.2% in 1983, but subsequently declined again in 1985 to 83.0%. This was essentially unchanged at 83.4% in 1987 (NSF (1990)). Vetter's study also found that the salary differential widened as the number of years of experience increased. This latter observation also emerged from the matched triad study (see Table 2), which shows for mathematics the percentage of the men's median annual salary by which the women's median annual salary is lower. Notice, in particular, the lower salary for women in the 1975-1978 cohort who by 1979 -- that is, after only one to four years from their doctorate -- are already behind by 2%. Since it is assumed starting salaries were comparable, such a lag is somewhat disturbing. These salary differentials constitute a large cumulative effect over the years, not only in absolute numbers of dollars but also on retirement benefits, which tend to be a function of salary during employment.
The matched triad study included the derivation of prediction equations for salary (see the discussion on ranks). The standardized regression coefficients for the respective predictors for salary are shown in Table 3. As for rank, the most important predictor variable was the time factor, with the sum of coefficients for years since the doctorate and years of full time equivalent experience representing the average year's increase in salary. Notice once again this is less for women than for men. Interestingly, although teaching as a primary activity was a potential predictor, it never in fact entered the equation, while administration as a primary activity did enter as a positive effect on salary for men but as a negative effect for women. Being married at the time of the doctorate had a negative effect on salary for women. Receiving the bachelor's degree from a research university had a greater effect than receiving the degree from a liberal arts college or overseas, but in all cases there was a negative effect on the woman's salary but a positive contribution to the man's salary. When sex was included as a predictor in the equations, the effect for women was always negative and significant and predicted women to have a salary about $1,550 (in 1979) less than the comparable male. These prediction equations were subsequently used to predict the salary a woman would have received had she been paid as a man (using again the matched triads of doctorates from 1958-1978). The residuals (of actual minus the predicted salary) were found for successive five year periods since the doctorate had been awarded. The residuals for the most recent group of women tended to center around zero but they shifted more negatively for each successive five year cohort, substantiating the results of the Vetter (1987) study noted above. A corresponding opposite-sex equation for men had positive residuals for men (see Scott (1977)). Thus, when compared with those of equal ability and attributes, women tend to be underpaid while men tend to be overpaid. While these conclusions are regrettable, they are however correctable. Scott (1977) developed a higher education salary evaluation kit whereby the salaries of women in homogeneous departments (or similar units) which have about fifteen or more white males, can be flagged as being potentially inequitable. Furthermore, the amount by which the salary needs to be adjusted to bring it to the level of a comparable white male is estimated. This study investigated many sets of potential predictor variables for the resultant regression analysis. Surprisingly, the set of predictors consisting of year of birth and year of doctorate (where ``year'' means the last two digits only) generally served very well. Adding additional predictor variables such as number of publications, number of doctoral students, etc., did not greatly improve the prediction. A possible explanation is that, since the regression analysis is performed over the homogeneous unit itself, most individuals in that unit publish papers and produce students at a comparable rate. Therefore, these variables do not tell us very much more than is already known and so do not add any additional predictive power. Whatever predictor variables might be the most relevant for a particular unit, it is important not to use variables such as numbers of articles published, rank, tenure, administration, which are themselves biased against women. While this evaluation kit allows identification of women receiving salaries lower than comparable males, Scott (1979) cautions that, when this method appears to identify a woman as earning an average male salary, care should still be taken as she may still be earning less than what she herself should be earning were she a male when actual accomplishments are taken into consideration. By plotting salary as a function of years since the doctorate, Scott (1979) shows that all women in a certain department were underpaid. This includes a particular woman whose salary was comparable to the men and so would not be flagged from the regression analysis. Yet, on further investigation, by any measure of merit adopted (except salary), this woman was clearly the most outstanding person in the department. Gray and Scott (1980) discuss further the dangers and misuses of the regression analysis. Billard et al. (1990) provided a statistical adjustment method to remove identified gender bias in such salaries.
Copyright ©1991
American Mathematical Society. Reprinted with
permission. |
||||||||||||||||||||||||||||||||||||||||
