Economists have been studying gender differences in earnings since the 1960s. Figure 1, calculated using data from the U.S. Census Bureau, makes the widely cited comparison between the median earnings of male and female full-time, year-round workers from 1960 to 2015. If men and women were doing the same work, one could expect the ratio of female-to-male earnings would approach 100 percent. It does not. As of 2015, median female earnings were 80 percent of median male earnings for full-time, year-round workers. This is a widely reported and somewhat misleading statistic, because it does not account for factors that are known to determine wages such as education and work experience. In this essay, I will explain an economist’s perspective on the gender gap, what we know and what remains to be determined about why women earn less than men. As an economist who studies the gender pay gap my job is to first use data to document and explain the gender pay gap. I will show that the gender gap explained by factors that influence earnings differs considerably from the 80 percent number given in Figure 1. Second, I will review what economics research tells us about the gender gap. Throughout my focus will be on the persistent gap for highly-educated women—those with a bachelor’s degree or more, because it is among the highly-educated that the gender gap is the largest.
Economic theory argues that equally productive workers will be paid the same for the same work, regardless of characteristics like gender, race, and sexual preference. However, choices men and women make prior to the labor market influence what they ultimately earn. In addition, the evolution of careers and work experience influence earnings. Non-market factors, such as the choices to marry, have children, and who cares for the children influence earnings as well.
When examining the gender gap, the economist’s goal is to make apples-to-apples comparisons. This necessitates controlling for observable factors associated with earnings. I will use the 2013 National Survey of College Graduates (NSCG) to examine gender differences in salaries to demonstrate how the gender gap changes once we adjust for these factors. I limit my sample to those with bachelors, masters, and professional degrees who are prime-aged workers (ages 25-54), work full-time, and full-year. I drop those with doctorates, since many work in academic institutions, and their employment outcomes differ from those of the general population. I also drop full-time workers earning less than the minimum wage ($16,700 per year). My measure of earnings is annualized salary.
The first two bars in Figure 2 compare the overall female-to-male earnings ratio for all workers in the U.S. to those with a college degree in 2013. Whereas the gender earnings ratio is 78 percent for all workers in the U.S., for the college-educated it is 75 percent. Once I add controls for race and immigration status, the gender salary ratio narrows to 76 percent—still larger than for all U.S. workers. I now demonstrate how the female-to-male salary ratio changes using the National Survey of College Graduates, and explanations from economic theory.
Explanations for the Gender Gap
Education. Given that I am interested in the gender gap for the highly educated, my model adds controls for highest degree (bachelors, masters, or professional) and years of work experience. The female-to-male salary ratio increases from 75 percent for the college-educated to 78 percent when accounting for highest degree. In work I have done with co-authors Steven Ceci, Shulamit Kahn, and Wendy Williams, we found that women were significantly less likely to pursue degrees in math-intensive fields. Though women now receive the majority of bachelor’s degrees, 57 percent as of 2014, they do not pursue the same majors. In 2014, women made up over 50 percent of bachelor’s degree recipients in Psychology, Social Sciences (excluding economics), life sciences, and the humanities. Women obtain more than 40 percent of math and computer science bachelor’s degrees, and only 20 percent of engineering degrees.
In addition, occupations that use mathematical skills, such as computer science and engineering, pay significantly higher salaries than those that do not use mathematics. For example, in 2013, median earnings in engineering occupations were $82,000, but only 22 percent of those working in engineering occupations were women. In contrast, median earnings of non-science and engineering occupations were $61,000, and 51 percent of those employed were women.
… occupations that use mathematical skills, such as computer science and engineering, pay significantly higher salaries than those that do not use mathematics. By the time girls matriculate in college, their math courses and skills may not be sufficient to pursue majors that require significant mathematics.
Our research found that girls are more likely to stop taking math courses in middle school and high school. This is not the result of girls’ lack of ability. Over time, girls have closed the average math test score gap, but boys still are more likely to take advanced placement high school courses in math, and are more likely to have the top scores.1 Part of this gap may be the result of risk-aversion. Until recently, the SAT penalized test-takers for guessing. Katherine Baldiga recently designed an experiment using practice questions from the SAT, and then varied the penalty for a wrong answer. When there was no penalty for guessing, women and men were equally likely to complete the test. When the penalty increased, women were significantly more likely to skip questions even after controlling for prior knowledge. Thus, women’s risk aversion may explain part of the observed test score gap in high stakes tests. Competition may also influence girls’ mathematics performance. Muriel Niederle and Lise Vesterlund argue that gender differences in mathematics test scores may result from gender differences in responses to high-stakes tests. By the time girls matriculate in college, their math courses and skills may not be sufficient to pursue majors that require significant mathematics.
To see how field of study influences the gender gap, I introduced controls for 31 major fields of highest degree that include the humanities, science, engineering and social science. The female-to-male salary ratio rises by nine percentage points to 87 percent in Figure 2. Thus, field of study explains over 40 percent of the gender gap among the college-educated.
Greedy Occupations and Work-Life Balance. Sociologist Lewis Coser developed the concept of “greedy institutions”: those institutions and organizations that demand a person’s full attention, and reduce the time spent on competing roles. Economists have found that hours of work demanded in some professions give rise to “greedy occupations” that disadvantage women. Claudia Goldin and her coauthors have examined these greedy occupations.    Among MBAs, her studies found there are small earnings gaps at the start of working, which widen to the point that men make almost twice as much as women 15 years after completing the degree. Goldin analyzed the work requirements in these occupations, finding that these kinds of jobs do not provide flexibility in hours of work, and are associated with larger gender pay gaps. In contrast, Goldin and Lawrence Katz found a small gender gap among pharmacists where women earn 92 percent of men. She attributes this to pharmacists being substitutable for one another. In fact, Patricia Cortés and Jessica Pan have found that women are less likely to select occupations with high work hours. Taken together, these results suggest that part of the gender salary gap is a very high return to working longer hours in these “greedy occupations.”
In my analysis of the NSCG, adding hours of work to salary regressions explains only one percentage point of the gender salary gap, increasing the female-to-male salary ratio to 88 percent.
Long work hours in some occupations come into direct conflict with parenting. In fact, parenthood can be considered another of Cosner’s “greedy institutions” that requires full attention and leaves little time for other activities. Women are often the primary care-givers of children, and the demands of taking care of infants and small children are time-intensive, especially if the child becomes sick (which happens more frequently if a child is in a daycare while the mother works). In addition, women’s earnings in some occupations are not large enough to compensate for the high costs of childcare.
Several studies, including my work with Madeline Zavondy and Marianne Sundström, have shown that men earn a marriage premium and women, in some cases, earn a marriage penalty.  In both studies, our results indicate marriage makes men more productive because women specialize in household production and taking care of children. Claudia Goldin, Marianne Bertrand, and Lawrence Katz found that children were associated with a significant growth in the gender gap among MBAs. Once female MBAs have children, they work fewer hours, shift to less-demanding (and lower-paying) jobs, and some leave the labor force entirely.
… women who leave the labor force after the birth of their first child experience twice the earnings penalty as women with a high school education.
If women do leave the labor force to stay home with their children, they experience a large earnings penalty. Julie Hotchkiss, Melinda Pitts, and Marybeth Walker estimated the effect of leaving the labor market after the birth of a first child on women’s earnings. Assuming that leaving the labor force is exogenous, women who leave earn 51 percent less than women who remain. They also find that college-educated women who leave the labor force after the birth of their first child experience twice the earnings penalty as women with a high school education.
In my analysis of the NSCG, I cannot control for spells that women leave the labor market because of childbirth. However, when I add marital status and children to salary regressions, the gender gap widens and the female-to-male salary ratio is now 85 percent. Since marriage and children have different effects on salaries for men and women, it is instructive to make comparisons based on these characteristics. The female-to-male salary ratio among single childless women and men women is 88 percent. The female-to-male salary ratio among married men and married women without children is 84 percent; and the female-to-male salary ratio among married men and married women with children is 87 percent. Part of the increase in the salary ratio for women with children could be attributed to positive selection: women capable of working and raising children simultaneously may be of higher overall ability, resulting in higher earnings potential.
In some cases, women leave the labor force, or work in a job that differs significantly from their field of study. Together with Joshua Rosenbloom, I have investigated women’s participation in computer science majors and careers. The lack of women in computer science and information technology has gained much notice in the popular press. Computer science is also an outlier: in 1983 at the start of the PC revolution, women received over 35 percent of bachelor’s degrees in computer science. In 2014, that number had dropped below 20 percent, making computer science the only science, technology, engineering, and mathematics (STEM) field where women’s representation has decreased since the 1980s. We found women were more likely to leave computer science and information technology jobs than men, and their departure was associated with having young children. In addition, we found that women computer science/information technology majors were almost twice as likely to be working in a job unrelated to their degree 10 years after they graduated. With Shulamit Kahn I have also investigated whether women were more likely to leave engineering occupations. We found women were more likely than men to leave engineering occupations, and this occurred when they left the labor force entirely after having children. However, women were less likely to leave engineering than other STEM occupations.
Our final model in Figure 2 includes controls for employment sector and whether the worker is using their degree in their employment. Women are significantly more likely to work in education and the non-profit sectors, making up 68 percent of employees. These sectors also pay less, with median salaries in education being $42,000 and in non-profit being $57,000. In contrast, only 35 percent of employees in the for-profit sector are women, and median salaries are $80,000. Once I add controls for employment sector and using one’s degree, the female-to-male salary ratio climbs to 87 percent. Overall, after accounting for education, experience, field of study, hours of work, marital status and children, the female-to-male salary ratio climbs from 75 percent to 87 percent. These factors explain nearly half of the 25-percentage point gender gap.
Figure 3 shows one of the persistent puzzles in studies of the gender gap: the female-to-male salary ratio falls with years of work experience. The blue bar shows the unadjusted female-to-male salary ratio and the red bar shows the same ratio after adjusting for education, experience, major, hours of work, marriage and children, and employment sector. Both the unadjusted and adjusted female-to-male salary ratios fall with years of work experience. Some of this decrease may be women who left the labor force and then re-entered, but it remains a puzzle for those seeking to understand why women earn less than men. Behavioral economics is where we now turn to address this puzzle.
Behavioral Economics and the Gender Gap: Competition, Gender Norms, and Negotiation. As you can see from the above discussion, much of the gender gap can be explained by education, experience, major choice, family, and employment sector. However, a non-trivial gender gap remains. Behavioral economics provides additional insights; it is the study of psychology as it relates to economic decisions. Much of this research is based on experiments that are difficult to include in the gender gap estimates that I provided above. I review the evidence here.
Muriel Niederle and Lise Vesterlund ran a series of experiments where women and men were given a task of adding up numbers and then paid based on the number of correct answers. They found that women were less likely to compete, and men were overconfident about the return to competition. Ernesto Reuben, Paolo Sapienza, and Luigi Zingales tested MBA students for their preferences for competition and then evaluated their earnings. They found gender differences in preferences for competition explained 10 percent of the gender gap, and that individuals who preferred competition worked in higher-paying industries nine years after their MBA. Linda Kamas and Anne Preston found competitive students were more likely to major in engineering and natural science than non-competitive students. They surveyed these students after they entered the labor force and found the gender earnings gap disappears for women who are both competitive and confident. Thus, taste for competition has the potential to explain a significant proportion of the gender gap.
… much of the gender gap can be explained by education, experience, major choice, family, and employment sector. However, a non-trivial gender gap remains. Behavioral economics provides additional insights …
Marianne Bertrand, Emir Kamenica, and Jessica Pan examined whether gender norms influence women’s earnings. They found that share of household income earned by the wife drops sharply to the right of one-half, where women earned more than their husbands. They argue this empirical regularity is consistent with gendered norms that make wives averse to earning more than their husbands. They also found that within couples where a wife’s potential earnings were likely to exceed her husbands’, the wife was less likely to be in the labor force, and if she worked, she was less likely to earn up to her potential. In households where wives did earn more than their husbands, wives did more household chores, and these couples were also more likely to divorce. Thus, Bertrand and her co-authors argue that gender identity within the household influences the gender earnings gap.
Finally, Linda Babcock and Sara Laschever have written a series of books on the gender gap in negotiation. This work started with Babcock’s observation that male graduate students were more likely to ask for career-enhancing opportunities than female graduate students. The estimates in Figure 3 show that there is a small initial starting salary gap that grows over time. One potential explanation could be the implications of gender differences in negotiation. Suppose both a male and female college graduate are offered a job with a starting salary of $50,000. The woman accepts the position for $50,000 and the man negotiates a higher salary of $52,500. Assume both employees are equally productive and receive a 3 percent raise each year. Over a 20-year career, the woman will receive $67,176 less in total earnings. If the man is successful at negotiating for a slightly higher wage each year, this earnings gap will grow. Babcock’s research also showed that women were penalized if they aggressively negotiated for salary or other employment outcomes in a manner resembling men’s negotiation styles. In other words, negotiation is also gendered, and women who negotiate like men do not receive their desired outcomes.
Instead, Babcock and Laschever’s second book offered negotiation strategies tailored to women consistent with gendered norms. They recommend women understand their best alternative without negotiation, use data to support their case for increased salaries, and practice negotiating in large and small contexts in order to become more comfortable with the process. Andreas Liebbrandt and John List conducted a natural field experiment to examine how gender differences in negotiation influenced starting salaries. They found men were more likely to negotiate than women when negotiating the employment contract was ambiguous, and the resulting gender wage gap was larger in these jobs. In contrast, the gender gap in negotiation goes away when salaries are considered negotiable.
… Babcock and Laschever’s second book offered negotiation strategies tailored to women consistent with gendered norms. They recommend women understand their best alternative without negotiation, use data to support their case for increased salaries, and practice negotiating in large and small contexts in order to become more comfortable with the process.
After controlling for factors that influence salaries and earnings, the gender gap narrows considerably but does not disappear. The data I have presented on the female-to-male salary ratio does not allow me to incorporate competitiveness or negotiation in the model. However, the larger gender gap for married women without children is consistent with the gender identity norms found by Marianne Bertrand and her colleagues.
Implicit Bias and Discrimination. Many attribute the unexplained gender gap to implicit bias or discrimination. Implicit bias occurs when attitudes or stereotypes influence our action towards certain groups in an unconscious manner. For example, my colleague, Monica Biernat, performed an experiment testing the stereotype of height for women and men. In the experiment, subjects were shown pictures of women and men of equal height standing next to a table. The subjects were asked to estimate the height of the man or woman in the picture. Despite having a table in the picture that provided a frame of reference, the experimental subjects estimated that the woman was shorter than the man—a result consistent with a stereotype. Ernesto Ruben, Paola Sapienza and Luigi Zingales ran an experiment that indicated implicit bias against women working on a mathematics task influenced the hiring decisions even after providing self-reported ability to those making the hires.
In contrast to implicit bias, discrimination is an explicit decision to treat men and women in the labor market differently. Discrimination can take two forms: explicit prejudice and statistical discrimination. Explicit prejudice by employers, employees, and customers would prevent women or men from being hired and if hired, pay them different rates. For example, customer discrimination may affect the choice of a doctor: women may prefer to receive treatment from female doctors instead of male doctors. If this is the case, Obstetrics and Gynecology practices will have more women than men. Per the Association of American Medical Colleges, in 2015 women made up 85 percent of OB/GYN residents.
Job-seekers in given occupations can use these data to understand median and mean hourly earnings as well as annual mean wages paid in their metropolitan area or state. Armed with this information, women can use data to negotiate their starting offer.
Statistical discrimination differs from prejudice, but it is related to implicit bias. Statistical discrimination occurs when the average characteristics of a group are attributed to individual members of that group. Our research showed that women have lower average SAT math scores than men.1 Suppose a woman with higher than average math ability applies for a job that requires math, but the employer does not directly observe her skills. The employer may hire a man for the job because on average women have worse math skills than men. This would be an example of statistical discrimination.
Although there is substantial evidence of implicit bias and discrimination from experimental studies, it is very difficult to identify these effects in data that compare average differences in salaries by gender. Discrimination is much easier to prove in specific cases, but high-profile cases have resulted in mixed results for women claiming discrimination or sexual harassment. 
Policies and Practices To Narrow the Gender Wage Gap
Given the significant impact that marriage and family have on the employment decisions of mothers, providing paid family leave for workers could narrow the gender wage gap. California, Rhode Island, Washington, New Jersey, and the District of Columbia all provide paid family leave to care for a sick family member or new child. Charles Baum and Christopher Ruhm have examined the impact of California’s paid family leave on mothers’ employment and hours of work. They found paid family leave increased the probability that mothers took leave and were employed within nine-to-twelve months after the birth of a child. To the extent that paid family leave allows mothers to remain employed, it is likely that their wages will increase with continuous labor force attachment. Thus, greater access to paid family leave provides one mechanism for narrowing the gender salary gap.
The Paycheck Fairness Act has been introduced in Congress since 2005. This act “prohibits retaliation for inquiring about, discussing, or disclosing the wages of the employee or another employee in response to a complaint or charge, or in furtherance of a sex discrimination investigation, proceeding, hearing, or action, or an investigation conducted by the employer.” While better information on the gender salary gap may prompt some women to seek redress for differences in pay, it is unclear whether this information would actually close the gender gap given the difficulties in proving salary discrimination.
That said, women can help themselves by doing their homework on salaries and negotiation. The Bureau of Labor Statistics publishes salary information by occupation for the nation, state, and metropolitan areas. Job-seekers in given occupations can use these data to understand median and mean hourly earnings as well as annual mean wages paid in their metropolitan area or state. Armed with this information, women can use data to negotiate their starting offer.
Observable characteristics explain about half of the gender salary gap, but approximately 13 percentage points remain unexplained. Behavioral economics suggests that gendered norms, including tastes for competition and negotiation practices have potential to explain the remaining salary gap.
I also recommend that women learn more about how to negotiate. I have read both of Linda Babcock and Sara Laschever’s books on women and negotiation. When writing this essay, I went searching for my copy of Babcock and Laschever’s Ask for It, to no avail. I have lent my copy out to so many students and colleagues that I have lost track of it! Reflecting on my own experiences while reading these books, I realized I had done a reasonably good job negotiating with my employer, but not as good of a job negotiating with my husband when it came to doing work around the house. I have followed Babcock and Laschever’s advice to practice negotiating in contexts large and small, finding that much in life is negotiable. It is important for women to ask for what they need and want.
• • •
The gender salary gap is significantly more complex than a simple comparison of median earnings by gender. This essay has demonstrated the gender salary gap among the highly-educated can be explained by many factors, some starting as early as middle school mathematics courses. Different choices for college majors, attachment to the labor force, marriage, and children can influence how much women earn, and whether they remain in the labor force. Observable characteristics explain about half of the gender salary gap, but approximately 13 percentage points remain unexplained. Behavioral economics suggests that gendered norms, including tastes for competition and negotiation practices have potential to explain the remaining salary gap. Although policies such as paid parental leave show promise in closing the gender salary gap, given the uncertainty of public policy since the 2016 election, women will be best-served by understanding what they can expect to be paid in their chosen occupation where they live and negotiating for the best starting salary possible.