Gender Differences in Salary
…attitudes have come a long way since F.Y. Edgeworth worried about whether women should receive equal pay for equal work…
—Nancy M.Gordon, et al., American Economic Review, 19741
Perhaps the most basic way to contrast the differing career outcomes of men and women in science and engineering is by comparing their salaries. Salary reflects both the type of employment obtained and success in meeting the goals associated with the position held. As such, salary is a form of recognition for professional contributions and a measure of worth in the scientific community. Merton (1973 reprinted from 1942) argues that there is a strong presumption in science that recognition, including monetary rewards, should be determined on the basis of universalistic criteria related to scientific achievement. To the extent that female scientists and engineers receive fewer financial rewards than men for comparable achievements, their work is undervalued and they are underpaid.
Studies of gender differences in salary for scientists and engineers can be divided into two groups. The first group examines salaries within a
single academic institution (see, for example, Becker and Toutkoushian 1995; Ferber 1974; Fox 1981; Gordon, Morton, and Braden 1974; Hoffman 1976; Katz 1973). Single institution studies have the advantage of more detailed data on each individual and are based on a more complete understanding of the nuances of the local context of employment, but they are limited by the unique characteristics of that institution. A second type of study uses a large sample to study differences across fields, and often across sectors of employment. For example, Ferber and Kordick (1978) examined Ph.D.s in all fields with degrees from 1958–63 and 1967–72. Ahern and Scott (1981), the precursor to our study, examined salaries in five broad fields for Ph.D.s from the 1940s through the early 1970s. Many of these studies of salary are restricted to academics, such as Barbezat (1988), Farber (1977), Gregorio, Lewis, and Wanner (1982), Johnson and Stafford (1979), and Tolbert (1986), or a single field such as Hansen, Weisbrod, and Strauss (1978) or Morgan (1998).
While studies of salary differences for men and women in science and engineering differ widely in their samples, focus, and methodology, each study has found that the average female scientist or engineer earns less than her male counterpart. There have been several proposed explanations for this gap in earnings:
Women earn less because they are less qualified than men. While our analysis in earlier chapters found few gender differences in educational backgrounds, it is still possible that qualifications attained at the completion of formal education may be lower. Due to longer periods out of the labor force, women accumulate fewer years of experience and during periods of absence from S&E their skills may depreciate. Consequently, when women reenter the S&E labor force they will earn a lower salary than at the time of exit and will have foregone the salary increases due to accumulated experience. In anticipation of time out of the labor market, women may choose to invest less in on the job training or employers may invest less in female employees. Lower investment in training early in the career will produce lower future female earnings (Duncan and Hoffman 1979). Or, even with similar education and experience, women may be less productive than men in the scientific workplace. See Cole and Zuckerman (1984), Long (1992), and Xie and Shauman (1998) for a review of the literature on gender differences in productivity.
Cumulative advantage, as defined by Merton (1973 reprinted from 1942), suggests that men are the beneficiaries of gender inequities early in the career and that these early advantages are magnified over time. Even if salary is based entirely on productivity, early disadvantages in employment for women may lead to a pay gap that will grow over the course of their careers.
There may be crowding of women into certain subfields either because of choice, social norms and mentoring, or entry barriers to other subfields. Because salaries are the result of interactions between supply and demand, increases in supply will put downward pressure on wages in these more female friendly subfields. See Bergman (1974) for a general treatment of this phenomenon.
The theory of comparable worth (Bellas 1994) posits that fields that employ a higher proportion of women pay lower salaries because women’s work is devalued by society (Treiman and Hartmann 1991). According to this theory, the suppressing effects of gender composition occurs after controlling for economic factors that affect salaries.
Finally, and perhaps most controversially, female scientists and engineers may receive less pay than men for equal work as a result of subtle or blatant discrimination by employers. This discrimination may take the form of lower wages for women doing the same work as men at all levels of experience. For example, Bellas (1994) and Ahern and Scott (1981) found that the effects of experience on salary were larger for men than women, indicating that men are compensated more than women for any given level of experience. Discrimination may also be reflected in society’s tendency to devalue women’s work, paying lower salaries in fields where large numbers of women work (see point 4 above). Or, discrimination may come in the form of barriers to entry into certain prestigious subfields or jobs resulting in crowding of women into less prestigious, lower paying alternatives.
In this chapter we use data from four years of the SDR to examine the extent and causes of gender differences in salaries. We begin by describing the gross gender differences in salaries without controls for characteristics of either individuals or their employers. We find that men have had a nearly constant 20 percent advantage in salary during the 23 years from 1973 to 1995. To understand why men receive higher salaries and why there has not been an improvement, we add controls for variables that have been suggested by prior research. This is done initially by simply comparing the median salaries of men and women in, for example, the same fields or with the same year of Ph.D. To control simultaneously for a large number of factors, we estimate a series of multiple regressions. The differing characteristics of men and women, such as in experience and field of study, can explain much of the gross gender difference in salary. However, even with numerous controls, gender differences in salary remain. Reasons for these differences are discussed in the summary.
Salary data from 1973, 1979, and 1989 were converted to 1995 dollars using adjustment factors for inflation from the U.S. Census Bureau (1999). Multiple regression was used to estimate salaries for men and women after controlling for a large number of variables simultaneously. The effects of the control variables were allowed to differ by gender. In these regressions, the dependent variable is the natural log of salary in 1995 dollars. Since raises are generally based on a percentage increase, a loglinear model provides a better fit. See Hodson (1985) and Becker and Toutkoushian (1995) for further details. A loglinear model predicts the log of income for a given set of characteristics. Since an unbiased estimate of the predicted income (as opposed to the log of income) cannot be computed by simply taking the exponential of the predicted log income, we use Duan’s (1983) nonparametric smearing estimator to compute predicted incomes. For additional details, see Chapter 2.
GROSS GENDER DIFFERENCES IN SALARY
Figure 7–1 plots the median salaries of men and women in the full time, year-round U.S. labor force and for our sample of full time scientists and engineers for the years of the SDR used in our report.2 Doctoral scientists and engineers, whether male or female, are well-paid professionals who earn substantially more than the average worker in the U.S. economy. The median salaries of male scientists and engineers have remained about 100 percent higher than those of full-time men in the general population, while the median income of female scientists and engineers have declined from being 200 percent greater than those of women in the general population in 1973 to around 150 percent greater in later years. This decline for doctoral women corresponds to a rise in income for women overall in the U.S. labor force while the real income of women in S&E declined slightly (Figure 7–2).
Since 1973 the median income of male scientists and engineers has been approximately 20 percent higher than the median salary of female scientists and engineers, as shown by Figure 7–1. While large, the earnings gap between male and female doctoral scientists and engineers is much smaller than the gap in the entire U.S. labor force, which would be
expected given that male and female scientists and engineers are more homogenous in their characteristics than are men and women in the general population. Further, the gender gap in earnings is smaller than that for other female professionals (e.g., physicians, executives) or for scientific occupations that require less than a doctorate, such as technicians and programmers (U.S. Department of Labor-Women’s Bureau 1994). However, there has been no sustained improvement in the salary disadvantage for doctoral women in S&E during the 22 years since 1973, while there has been a steady improvement in salaries for women relative to men among full-time, year-round workers in the U.S. population. According to the National Commission on Pay Equity (1996), the shrinking gap is due to the gains women have made in real wages relative to men as a result of increasing years of work experience, increasing equality of education, improved market skills, and the decreased number of high-paying jobs for men. Men’s real wages (in constant dollars adjusted for inflation) drifted downward, while women’s real wages increased.
While gross gender differences in salaries for scientists and engineers have not narrowed since 1973, salary is the outcome of a stratification process that involves many steps, each of which is associated with differences in pay. Earlier chapters showed that due to the increasing entry of women in recent years, female scientists and engineers are on average younger than their male counterparts. Accordingly, we would expect the younger women to earn less. Further, there are gender differences in field of study, sector of employment, and primary work activity. Each of these dimensions of the career is associated with differences in salary and we find generally that women are more likely to be in positions associated with lower salaries. In the remainder of this chapter, we decompose the overall gender differences in salaries, attempting to determine the degree to which men and women with similar characteristics are paid differently.
PROFESSIONAL AGE AND DOCTORAL COHORT
While there has been no improvement since 1973 in the pay discrepancy between the average male and female scientist or engineer, we know from Chapters 3 and 4 that the average professional age of women is less than that of men. Since salary is strongly affected by years of experience (Ahern and Scott 1981), even if women were compensated in the same way as men, we would expect the average salary for the younger population of women to be lower than that of men. If, however, men had a slight salary advantage at the start of the career, this small difference in starting salary would multiply over time since raises are often calculated on a percentage basis. Further, if women have more interruptions after the
Ph.D., this loss of experience would lead to increasing gender differences over time. Ferber and Kordick (1978) found such an increase in a study of Ph.D.s from 1958–63 and 1967–72, and found convergence in income after women reentered the labor force.
Panel A of Figure 7–3 plots the median salaries of men and women in 1973 by the number of years since the Ph.D. The median salary in any given year is a 5-year average centered on that year. At the start of the career, men are making 12 percent more than women, compared to the 22 percent gross difference we found when the different age structures for men and women were ignored. The gender difference in salary increases steadily to 20 percent in year 15. For the next 10 years, there is an overall increase, although there are substantial fluctuations due to the small number of women with Ph.D.s from the years prior to 1958. Panel B plots similar data for 1995. The first thing to note is that the salaries for both men and women are lower at all stages of the career compared to those in 1973. Since data in both figures are in 1995 dollars, this documents a decline in real income for scientists and engineers between 1973 and 1995. Second, in 1995 the gender gap begins at 20 percent in year 1. For later years, the differences in salaries are generally smaller than in 1973, but for all career years men earn at least 10 percent more than women of the same career age.
The conclusions that we can draw from Figure 7–3 are limited since we are not plotting the salaries of the same group of people as they age over the career. Instead, each year of the career corresponds to a different Ph.D. cohort. For example, in 1973 those in year 5 received degrees in the years around 1969 (recall that we are plotting five-year averages), while those in year 10 received degrees in the years around 1965. Cohorts of Ph.D.s from different years are used to approximate what might happen to a cohort from a single year as it progresses through the career. When interpreting results based on these synthetic cohorts, it is impossible to differentiate empirically between alternative explanations of the results. For example, in Panel A it appears that women encounter a “glass ceiling” around year 20 while men’s salaries continue to increase. An alternative explanation is that the cohorts of women that received their Ph.D.s more than 20 years earlier faced obstacles earlier in their careers that limited their incomes later in the career. If more recent cohorts do not face these obstacles, their salaries would continue to increase as they age. Using this argument and data for engineering, Morgan (1998) concludes that the “earning penalties to women are more a matter of when they started their careers than of how long they have worked.”
Given the limitations of synthetic cohorts and the results of Morgan (1998), it is important to examine what happens to the same cohort of Ph.D.s over time. This is done in Figure 7–4, which plots gender differ-
ences for four cohorts defined by the Ph.D. year at four years of the SDR. Each bar shows the percent higher median salaries for men at a given number of years since the Ph.D.; the number at the top of each bar is the approximate career age for that cohort in a given survey year. The set of four bars above shows that the salary advantage for men with degrees from 1959–1968 increased from 17 percent at career year 11 to 19 percent by year 17, with a drop in year 27, ending with a difference of 21 percent in year 33. A steady increase in the salary advantage for men is also seen in the 1969–1978 cohort. By comparing those with similar career ages in different cohorts (e.g., age 11 for the 1956–1968 cohort, age 7 for the 1969– 1978 and 1979–1988 cohorts, and age 5 for the most recent cohort), we find some evidence of a modest decrease in the salary differences for men and women in more recent years.
While these results demonstrate that some of the overall gender difference in salaries can be explained by gender differences in professional age, substantial differences remain. These results are based on years since the Ph.D. Ideally, we would compare salaries of individuals with the same years of full-time professional experience, taking into account interruptions in the career and part time employment. Unfortunately, com-
plete data on years of postdoctoral work experience are not available. Since women are more likely to have interruptions, perhaps due to family obligations, the results given above may over-estimate the age standardized gender differences in salary. For example, career age for women is more likely to over-estimate professional experience than for men. Using data from 1983, Lewis found that career interruptions had equal effects on the salaries of male and female scientists and engineers, but that women were more likely to have interruptions.
The gender differences in salary may also be accounted for by gender differences in other dimensions that affect salary, such as field and type of employment. These dimensions of the career and their effects on salary are now considered.
The link between a field’s sex makeup and its salary level led us to ask whether more female fields pay less partly because their practitioners are mostly women.
—Marcia L.Bellas and Barbara F.Reskin, Academe, 19943
Fields differ substantially in the median salaries received by Ph.D.s employed in those fields, as shown in Figure 7–5. Engineers have the highest median income, followed by physical scientists, with mathematicians, life scientists, and social/behavioral scientists following. Field differences have been increasing since 1973, confirming the results of Bellas (1997). For example, in 1973 the median salary in engineering was 8 percent greater than in the social and behavioral sciences; by 1995 the difference was over 20 percent.
While Johnson and Stafford (1979) found no discernable pattern of field differences in salary, a series of papers by Bellas and collaborators (Bellas 1993, 1994; Bellas and Reskin 1994) demonstrated that fields employing higher proportions of women pay lower salaries. Her work is based on the concept of comparable worth that argues that since women’s work is devalued by society (Treiman and Hartmann 1991), occupations that are predominantly female receive lower compensation. A simple labor market supply and demand framework can also explain this phenomenon. With the influx of women into science, certain fields saw more absolute growth of employees than others, possibly due to free choice of entering women or to entry barriers imposed to prevent female entry into other fields. In particular, psychology, life sciences and the social sciences were the destinations for many female entrants. With the large increases
in supply of employees, and without similar increases in demand, wages were depressed in these fields relative to fields without these large supply increases. Studies of comparable worth have, however, included controls for labor market conditions. For example, Bellas (1994) used the 1984 National Survey of Faculty sponsored by the Carnegie Foundation (1984) and found that the negative effects of gender composition persisted after control for individual characteristics and labor market conditions.
For 1989 and 1995, Figure 7–6 shows the negative relationship between the percent of Ph.D.s who are female in the full-time labor force of a field and the median salary for that field. There was a weaker relationship in 1973 and 1979 (not shown) since there was little variation in the percent women among fields. Clearly, women are more frequently found in those fields with the lowest salaries. For example, women are much less likely to get degrees in the more highly paid field of engineering and much more likely to obtain degrees in the social and behavioral sciences. There are also differences in subfields. For example, women are much less likely to obtain a doctorate in economics, where salaries are higher, than in anthropology, where they are lower.
While comparable worth suggests that both men and women, not just women, earn less in those fields where there are proportionally more women, our data suggest that women receive less than men even within lower paying fields. Figure 7–7 shows that men have higher salaries in all fields in each of the years examined. However, with the exception of the social and behavioral sciences, there has been a within field decline in the
salary advantage for men. In the social and behavioral sciences, men have had a nearly constant 10 percent salary advantage. Thus, women are most likely to have degrees in the broad field that pays the least and in which salary advantages for men have persisted longest. Keep in mind, however, that these figures do not control for professional age.
Regardless of the explanation, women are more frequently found in those fields with the lowest salaries. Overall, field differences accounts for a significant proportion of the gross differences in salary that were documented in the last section.
EMPLOYMENT SECTOR AND PRIMARY WORK ACTIVITY
Figures 7–8 and 7–9 plot median salaries by sector of employment and primary work activity. In each year the salaries are highest in industry, which in large part explains the higher overall salaries of engineers. While in 1973 the median salary in government was close to that in industry, since 1973 government salaries for Ph.D.s have dropped significantly relative to those in industry. Salaries are lowest in academia, where women are most likely to work. Even larger salary differences exist among work activities, as shown in Figures 7–10 and 7–11. The highest salaries are in management, due in large part to managers having more work experience than the average Ph.D. Salaries drop steadily as we move from pro-
duction work to applied research, and finally to the lowest salaries for those who are teaching. Overall, differences in salaries by sector and activity are important for understanding gender differences in salaries since women are more likely to be employed in those sectors that pay less and in work activities associated with lower salaries.
There are also differences among sectors and work activities in the degree to which men receive higher salaries than women. Figure 7–10 shows that the salary advantages for men are greatest in the nonprofit sector, with a steady increase from 23 percent in 1973 to 32 percent in 1995. Differences are smallest in government, with a small increase between 1973 and 1995, including a spike to nearly 25 percent in 1979. In both industry and academia, there has been an overall increase in gender differences, although there is evidence of a decrease between 1989 and 1995. Gender differences also vary by work activity, as shown by Figure 7–11. Differences are largest in management, production, and basic research, with smaller differences in teaching and applied research. While there is no clear trend over time, it is important to keep in mind that these figures do not control for gender differences in professional age.
The results so far have controlled for only a single factor at a time (e.g., professional age, sector). But, many key dimensions of the career are interrelated. For example, employment in industry is more likely in engineering and less likely in the social and behavioral sciences. And, within some sectors applied research is more likely, while in other sectors basic research is more common. Interpretation is further complicated since there are significant gender differences in years of professional experience with increasing entry of women occurring at different rates across fields and sectors. Accordingly, to more fully understand gender differences in salary it is necessary to control for these simultaneously. In this section we use regression to examine gender differences in salary after controlling for multiple dimensions of the career. Our strategy is to estimate separate regressions for men and women, which allows the effects of each variable to differ by sex. For each pair of regressions, one for men and a second for women, the predicted salaries for men and women are computed for the combined male and female average levels of the control variables in the equation.4 These predictions are used to compute the percentage differ-
ence in the salaries of men and women. Additional variables are added to the regressions and the advantage in salary for men is computed after controls for the additional variables. See Chapter 2 for further details.
Figure 7–12 shows changes in the salary advantage for men as additional variables are added cumulatively to the regression. The two panels present the same information organized to highlight different aspects of the results. The first set of bars in Panel A plots the percent higher salaries for men when only the gender of the individual is used to predict salary. As shown earlier, there is no consistent pattern over time, with men earning between 22 percent and 26 percent more than women. For the second set of bars, career age is added to the regression. Gender differences drop only 2 points in 1973, with drops of between 7 and 10 percentage points in later years. By 1995 the percentage advantage for men has decreased to 13 points after controlling for differences in career age. Keep in mind that we had to use career age rather than years of full-time experience due to missing data for the experience variable. Since women have more time lost to interruptions, we expect that gender differences would have been even smaller if controls for experience were used. The third set of bars adds field of doctoral study, reducing the adjusted gender difference by only 1 point in 1973, with decreases of over 5 points in 1989 and 1995. The male salary advantage continues to drop as controls for sector and primary work activity are added.
With all controls added, the advantage for men was cut in half to 14 percent in 1973 and 1979. In 1989, the advantage was reduced an additional two-thirds to slightly below 10 percent and by 1995 the advantage for men was further reduced to slightly above 5 percent, a drop of three-quarters. After adding controls for differences in background and work experience, a steady decrease over time in the salary advantage for men is found.
Bayer and Astin (1975) argued that the explained variation (i.e., R2 or coefficient of determination) in salary regressions for women should be smaller than for men. Their argument was that the salaries of women are more strongly affected by discrimination and consequently would not be explained by other variables such as field or years of experience. It is also likely that the careers of women are less predictable than those of men due to a greater number of career interruptions. Figure 7–13 shows that this was clearly the case in 1973 and 1979, but that the difference has declined and is nearly eliminated by 1995. There has also been a steady decrease in the amount of variation that can be explained by the structural variables included in our models. This decrease in what can be explained may reflect the changes in the scientific and engineering labor market that have occurred since 1973.
SALARIES IN INDUSTRY AND GOVERNMENT
The effects of age, field, and work activity may differ by sector of employment. For example, the salary advantage for men in engineering may be larger in one sector than another. To allow for this possibility, a series of regressions was run for each sector separately for industry, government, and academia; there were too few cases for separate analyses of those working in the nonprofit sector.
Figure 7–14 shows the percentage difference in salaries for men and women in industry after controlling for age, field, and work activity. With all controls, shown by the set of bars labeled “+PWA”, the higher salaries for men are reduced from an 18 percent to a 7 percent advantage in 1973; in 1979 the male advantage was over 15 percent even with controls. By 1995 there was a substantial reduction to an adjusted difference of less than 5 percent. These results are consistent with Vetter’s (1992) finding that there has been convergence in the salaries of doctoral chemists in industry.
Figure 7–15 presents similar data for those employed in government. Overall, the salary advantage for men is smaller than that in industry, and by 1995 after controlling for age, field, and sector, women are estimated to have marginally higher salaries than men.
SALARIES IN ACADEMIA
…[academic] salaries are not of the nature of wages and that there would be a species of moral obliquity in overtly so dealing with the matter.
—T.Veblen, Higher Learning in America, 19185
Despite Veblen’s warning of moral delinquency, the majority of studies of the salaries of scientists and engineers are focused on the academic sector, often being further restricted to those with faculty positions. A key advantage to studying the academic sector is that more is known about characteristics of the employing institutions, the work activity, and, to some extent, productivity. These studies include: Bayer and Astin (1968; 1975), Becker and Toutkoushian (1995), Ransom and Megdal (1993), Barbezat (1988), and Toutkoushian (1998). Gray (1993) provides a detailed review of statistical analyses of faculty salaries used in court cases. Overall, these and many other studies have concluded that there has been substantial progress in academia in reducing gender differences in salaries. Barbezat (1988) concluded that “salary discrimination” in the academic market is less than in other sectors of the economy. In this section, we begin by examining gender differences in salaries among all full-time, doctoral academic employees. We then restrict our analysis to the influential group of tenure-track faculty at research universities.
Figure 7–16 plots the percentage salary advantage for academic men after controlling for key dimensions of the academic career. Analyses are based on doctoral scientists and engineers employed full time in academia, regardless of work activity or type of institution. The two panels present the same information organized first by the variables added to the regression and second by the year of the survey. The first column in Panel A shows the overall gender differences in salaries without any controls. The higher salaries for men increase from 18 percent in 1973 to a high of 24 percent in 1989 before dropping back to 20 percent in 1995. The results labeled “+Age” show that the increasing overall differences in academic salaries during this period were due to the younger professional age of women in academia. Controlling for professional age substantially decreases the salary advantage for men, particularly in 1979 and later. By 1995, the advantage for men is reduced to 10 percent. If data on years of experience had been available, these decreases would probably have been even larger. Looking at Panel B, we see that the effects of professional age only became large after 1973 (shown by the large drop from the solid black bar to the adjacent bar). This corresponds to the rapid influx of
women into academia during this period. Since women are more likely to have interruptions due to family obligations, our measure of experience as years since the Ph.D. is likely to overestimate the professional experience of women. If we had a measure of years of work experience, the reduction in gender differences in salary would likely be even greater. Ferber and Kordick (1978:227), in a study of Ph.D.s from 1958–1963 and 1967–1971, concluded that “the relatively lower earnings of highly educated women can be explained largely by their career interruptions…” She found that once women reentered the labor force on a permanent basis, gender differences in salary were reduced. Unfortunately, more recent data are not available.
In academia, as in other sectors, there are significant salary differences across fields. Feldberg (1984:315) found that in academia, as in science as a whole, faculty in fields where there are proportionately more women receive lower salaries even after controlling for human capital and scientific productivity. Bellas (1994) confirmed this result in several studies that were discussed earlier. Note, however, that women tend to be found least often in those fields in which there is the greatest demand from industry, and accordingly salaries would be expected to be higher. While we confirm the direction of field differences from past research, the magnitudes are small after controlling for differences in years of experience. In 1973 and 1979, controls for broad field resulted in only trivial reductions in gender differences, with somewhat larger reductions of 3 points in 1989 and 2 points in 1995. Since our measure of field is based on the doctoral degree, it is possible that the effects of field of employment would be larger. However, since there is relatively little switching across broad fields, this difference is unlikely to be large.
Different Carnegie types of institutions have substantially different rates of pay. For example, in 1995 academics in the elite Research I universities were making 5 percent more than those in Research II universities, 15 percent more than in Doctoral universities, 20 percent more than in Master’s, and 33 percent more than in Baccalaureate institutions. As shown in Chapter 6, women are more likely to be employed in those institutions with lower median salaries. Figure 7–16 shows that adding controls for Carnegie type to the regression containing professional age and field does not substantially reduce the overall gender differences in salary. However, if we examine the gender difference within each type of institution, we find some important differences. Figure 7–17 plots the percentage higher salaries for men by Carnegie type of employer based on the regressions described above. The plot is computed for an academic 15 years from the Ph.D. who is average on other characteristics. The results show that gender differences in salaries have declined since 1973 in all types of institutions, but that the largest changes since 1973 are found in
those institutions with doctoral degree programs. This finding is explored further in the next section where we focus on academics located in Research I universities.
Adding controls for the type of work activity further reduces gender differences in salary, but by a relatively small amount. If we further refine work activity to include distinctions among faculty ranks, shown by the last set of bars in Panel A in Figure 7–16, the overall salary differences between men and women are reduced to less than 8 percent in all years and just 5 percent in 1995.
Tenure Track Faculty at Research Universities
Our findings above have shown that a great deal of the overall gender differences in salaries can be accounted for by the differing professional ages of men and women in academia, with smaller reductions introduced by controls for field, type of institution, and work activity. This section provides a more detailed analysis of faculty with tenure-track positions in research universities (i.e., Research I or Research II universities according to the Carnegie classification). We limit our analyses to this group of scientists and engineers for two reasons. First, work environments differ widely among types of academic institutions. Consequently, the effects of variables such as rank and productivity may operate differently at different types of institutions. By restricting our analyses to a more homo-
genous group of academics, the meaning of our findings should be clearer. Second, tenure-track positions at research universities are often considered to be the most prestigious academic appointments and these faculty train the largest number of Ph.D.s and produce the majority of research in the United States. Accordingly, it is appropriate to give more detailed consideration to this group of academics.
From 1979 to 1995,6 the overall salary advantage for tenure-track men in research universities dropped slowly from 30 percent in 1979 to 26 percent in 1989 and finally 25 percent in 1995. Note that salary differences in these positions were greater than in the population as a whole. A possible explanation for the slow progress in overall salaries for female faculty in research universities is the greater professional experience of male faculty. Not only is the average male faculty member older, but some research has suggested that the salary advantage for men increases with age. Using data from 1970, Johnson and Stafford (1979) found that the salary disadvantage for women starts small but rises dramatically over time. They conclude (Johnson and Stafford 1979:241): “As time passes, the earnings differential between the sexes grows, and this can be attributed to cumulative effects of discrimination or to the market’s reaction to voluntary choices for reduced hours of work and on the job training by women.” More recently, faculty salary data from the American Association of University Professors show a salary gap between women and men at each rank and across all academic fields, with the widest gap among full professors who tend to be the oldest Ph.D.s (Magner 1996b).
Our data, shown in Figure 7–18, show a more complicated picture. In 1979 (shown by the solid line), there was an increase in the salary advantage for men during years 1 through 5, a nearly constant 6 percent difference from year 5 till year 15, followed by increasing differences until a decline beginning in year 18 (which is based on a small number of female faculty). In 1995 there are larger differences in most years, with a gap of 14 percent in year 1, dropping to 11 percent in year 5. The remaining years track closely with the results from 1979. An alternative way to examine salary differences over the career is to examine gender differences by academic rank. Figure 7–19 shows the percentage higher salaries for men by academic rank for the years 1979, 1989, and 1995. As with years of experience, after controlling for rank women are increasingly less well paid than men later in the career with no evidence of improvement by 1995. Keep in mind that we have not yet controlled for other variables.
Barbezat (1988) reviews the debate on whether rank should be in-
cluded in regressions predicting salary in academic positions. The argument (Hoffman 1976) is that since women may be discriminated against by slower advancement in rank, estimates of discrimination in salary that include rank may be downwardly biased. Ahern and Scott (1981) found that both academic rank and salary are explained by the same set of individual-level variables and did not use rank to predict salary because “rank itself is influenced by gender.” Nonetheless, we believe that there are important reasons to examine salary differences within rank. It is important to know if men and women in the same rank receive comparable salaries. If, in fact, women are promoted more slowly, the allocation of raises on a percentage basis would make their salaries higher than men within a given rank (since women have been in rank longer), thus providing a lower bound for gender differences independent of the process of rank advancement. Accordingly, in the regression results that follow, rank is included as a predictor of salary.
Figure 7–20 summarizes the most important results of our regression analyses of faculty salaries in research universities. The first set of bars for each survey year shows the predicted percent difference in salaries for men and women after controlling for rank and professional age. The results are similar to those presented earlier, showing that controlling for professional age is largely equivalent to controlling for academic rank. The second set of bars for each year adds controls for characteristics of the scientists, including field, prestige of the Ph.D., elapsed time from baccalaureate to Ph.D., whether the employing institution is public or private, the prestige rating of the individual’s department, and whether it is a Research I university. Significantly, this substantially increases the predicted gender differences in salaries, with predicted differences in salary of 12 percent in 1979, 8 percent in 1989, and 10 percent in 1995. Essentially, these results indicate that men have substantially higher salaries than women with very similar educational backgrounds, institution locations, and experience. We have not, however, included controls for productivity.
As argued by Merton (1973 reprinted from 1942), rewards in science should be based on contributions to the body of scientific knowledge. In academia, unlike many locations in industry and government, these contributions are freely published. Johnson and Stafford (1979) argue that lower salaries may be due to lower productivity. Barbezat (1988), however, questioned whether differences in productivity might also be due to discrimination in publications and found that adding publication variables decreased gender differences in salaries. While there is a huge literature on how to measure scientific productivity (see Long 1992, Gray 1993, and the references cited therein for details), our analysis is limited to simple counts of publications obtained from the Institute for Scientific Information (see Chapter 2 for details). For 1979 we used publications
from 1981–1986; for 1989, we used publications from 1987–1992; and for 1995, from 1990–1995. We do not have information on productivity for prior periods (e.g., total productivity over the career) or measures of contributions to service, administration, or teaching. Thus, our measure of productivity is crude. Nonetheless, the last column for each year in Figure 7–20 shows that controlling for publications has a major impact. Controlling for publication substantially reduces the gender differences in salary, with smaller remaining differences at higher ranks. Still, even with these controls, significant differences in salaries remain: in 1995 female assistant professors were earning 9 percent less than men, female associate professors 6 percent, and female full professors 4 percent. Keep in mind that these differences might be further reduced if we had a better measure of professional experience.
This chapter presents a lot of detailed information that reflects the complexities of the scientific and engineering labor market. Salary is closely related to the outcomes analyzed in earlier chapters: professional age, sector of employment, and work activity. The more recent entry of women into science and engineering that was documented in Chapters 3 and 4 leads to women in the S&E labor force being younger, which in turn affects their salary. The greater tendency of women to be in the academic sector and in off-track positions, as shown in Chapters 5 and 6, leads to employment in jobs with lower salaries. Yet, adding controls for these dimensions of the career does not eliminate gender differences in salaries.
Figure 7–21 summarizes our findings by comparing gender differences in salaries across increasingly similar groups of men and women. Among all scientists and engineers, regardless of field, sector, professional age or other characteristics, there is a 28 percent salary advantage for men in 1979 that drops to 23 percent in 1995. If we consider only academics, where salaries are more homogeneous than across all sectors, the salary difference drops slightly. Restricting our comparison to faculty in tenure track positions results in an additional small decline. To standardize for age and achievement, we further restrict the comparison to full professors, resulting in a differences of around 10 percent. Since salaries differ across field, the next comparison considers only those in the social and behavioral sciences, where women are more likely to work. In 1979 this leads to few further reductions, but reduces the salary advantage to less than 3 percent in 1995. The last column shows that looking at more similar groups of men and women will not necessarily reduce the salary advantage for men. Among full professors in Research I universi-
ties in the social and behavioral sciences, men were earning nearly 15 percent more in 1979 reduced to 10 percent in 1995.
While controlling for background differences eliminates much of the gender difference in salary, it does not eliminate it altogether. Why? One possibility is that key variables are missing from the analyses or that others are measured poorly. Indeed, we can suggest many ways in which our analyses could be improved with more and better data. Our results show that at least for those in research-oriented locations, such as Research I universities, controls for productivity are essential and that not having such controls may grossly overestimate gender differences. But we can think of no compelling reason why other variables that we might like to have had, or to have measured better, would account for the remaining differences in salary. Moreover, while comparing men and women who are more similar reduces much of the overall salary difference, this does not change the fact that overall, women have significantly less well paying jobs in sciences. Discrimination need not come only in the form of differential salaries for equal work, but also through differential access to higher paying jobs. As Conway and Roberts (1983) put it: “Another type of possible discrimination is placement discrimination, which refers to the ‘shunting’ or ‘steering’ of females or minorities into lower job levels than their qualifications warrant.” Further, with each progressive stage of the stratification process, it becomes more difficult to distinguish outcomes that are the result of individual differences between women and men-
from outcomes that are the result of men’s cumulative advantage over women in science.
In summary, the male/female earnings gap in science is not fully explained by the individual or contextual factors that have so far been measured. Even analyses that methodically control for measurable differences between women and men and attempt to measure discrimination directly leave a large residual of the wage gap unexplained. While some of the remaining differences may be due to measurement error and it is possible that some control variables are missing, there appear to be differences that remain. Clearly, more research is necessary—not only on faculty salaries but also on gender differences in other employment sectors—in order to assess the effects of discrimination against women at each step in the stratification process.