Probably the most rehearsed explanation of the gender pay gap is discrimination. After accounting for traditional labor market factors, a large residual gap remains. This residual gap is also called the unexplained gap. Researchers often commit the fallacy of equating unexplained effect to discrimination effect instead of omitted variable bias. In fact, most wage decomposition models are probably contaminated by bias. This article will explain that much of the residual gap is likely due to other causes. In particular, time flexibility. The evidence for the discrimination effect is often ambiguous.


1. Bias in the decomposition of wage gaps
2. Time flexibility: The most important cause of the pay gap?
3. The interesting case of the wage gap in medicine
4. Does gender composition matter?
5. Gig economy also shows the gender gap
6. The impact of sorting effects
7. Marital status and child penalty
8. Gender difference in psychological factors
9. Negotiation also explains the gender gap
10. Unintended consequences of transparency laws
11. Family policies fail?

2023 Gender Pay Gap Report. Retrieved: November 27th 2023.

Payscale is a large online salary survey, weighting toward salaried professionals with college degrees, providing information about their industry, occupation, location and other compensable factors. In 2023, they estimate that women earn 0.83 dollar for every 1 dollar men make but women earn 0.99 dollar after controlling for job title, education, experience, industry, job level and hours worked. In 2015 those estimates were 0.73 and 0.97 dollar. They convert hourly compensation to annual compensation where necessary and re-scale annual compensation to a 40-hour work week where necessary.

Lifetime earnings are also influenced by controls. Over a 40-year career where wage growth is assumed to be 3% annually (that estimate is based on previous research, although larger increases have been observed in recent years), the life time earnings come to 5.2M for men and 5.14M for women after controls. This is a difference of about 70K dollars over 40 years. It should be noted that this estimate does not account for lost benefits, investments, promotions, or other compounding factors on lifetime wealth.

The controlled gender gap reported in Payscale is smaller than what is typically reported from various, smaller surveys, according to many research papers. As there are too many papers, likely thousands, I will focus here on the most important papers I was able to find.

1. Bias in the decomposition of wage gaps.

Huber (2015) explains that the difficulty in disentangling direct and indirect effects is the requirement of controls for mediators and confounders, the latter being typically neglected. For instance, using education (mediator) in the decomposition without controlling for family background (confounder) generally biases the explained and unexplained components in the wage decomposition. The reason is that education is already an intermediate outcome such that conditioning on it without accounting for confounders is likely to introduce bias, as discussed by Rosenbaum (1984) and Robins & Greenland (1992).

Huber (2015) insists on the necessity to apply more realistic identifying assumptions for wage decompositions: “as gender or ethnicity are determined at or prior to birth and therefore precede mediators like education or profession. For this reason, we have suggested the use of an alternative set of identifying assumptions that assumes exogeneity of the group variable and the mediators only conditional on observed confounders. Then, the unexplained and explained components of the decomposition can be nonparametrically identified by using a simple weighting expression that reweights observations by the inverse of the conditional propensity to belong to a particular group given the mediators and confounders.” (p. 188).

Huber & Solovyeva (2020) expressed further concerns in decomposition analyses which often select male wages as the reference wage due to being the non-discriminatory wages: “Sloczynski (2013), however, challenges such an interpretation, as it is conventionally not backed by a theoretical model supporting that male wages correspond to the general equilibrium wages in a world without discrimination. For instance, a weighted version of male and female wages could possibly better reflect the equilibrium wage, if the latter is to be defined as reference wage. In general, different (weights of) reference groups (in terms of their wages) change the magnitudes of the explained and unexplained components and thus, the interpretation of the counterfactual analysis, due to interaction effects between gender and observed characteristics in determining the wage.” (p. 11).

Huber & Solovyeva (2020) also reveals another, much less known bias: the histories (i.e., years) of the mediator over time that are typically omitted in decomposition analyses. Using the longitudinal NLSY79 data, their main specification includes current occupation as well as years in current occupation. When the current values are retained but the histories of occupation are removed, the explained components are reduced and the unexplained components are increased.

Another bias arises when hourly wages are missing and have to be substituted with monthly and annual earnings because these variables are contaminated with labor market interruptions.

Shatnawi et al. (2014) argued that the standard wage decomposition suffers from aggregation bias as it fails to capture gender segregation that occur within firms due to differences in job assignment and promotion. Their improved model treats job hierarchy as a continuum in wage rates which circumvents issues from non-overlapping occupational or job title distributions. The model is then extended by adding an interaction between the discrimination effect and segregration effect that are freed of any dependence on the characteristics of the female sample and on wage discrimination. This augmented model determines better the direction of the bias that results from not knowing the wage structure.

Navarová (2022, Table 4.2) employ a Bayesian Model Averaging (BMA) to estimate the effect of variables on their meta-analytic gender gap. Larger gender gaps were found for studies which do not use Oaxaca-Blinder (0.045) or do not control for selection bias (0.048) or omit experience (0.099) and female share (0.065) or use North American samples (0.082) while smaller gaps were found for studies which use only full-time workers (-0.040) or only college graduates (-0.071) or use Southern European samples (-0.045). The author found no evidence of publication bias.

Wilcox & Forhad (2023) use quantile Oaxaca decompositions to evaluate the heterogeneity of the wage gaps in STEM discipline. While Oaxaca decomposition at the mean reveals no gender gap after conditioning on academic productivity, the quantile decompositions reveal that women earn more (less) than men in top (low) quantiles. This adds to the challenge of interpreting these patterns as discrimination effects.

2. Time flexibility: The most important cause of the pay gap?

Goldin (2014, 2015) published probably the most important papers regarding the causes of the pay gap, and argued that jobs for which bargaining and competing matter the most are also positions that have the greatest nonlinearity (i.e., convexity) of pay structure with regard to working hours. If workers were perfect substitutes for one another, there would be no premium on earnings with respect to working hours, and earnings would move linearly. The corollary is that the pay gap is lower when earnings are more linear with respect to time worked. Goldin (2014) showed that, in the field of technology and science, job characteristics such as time pressure, more contact with others, and less freedom to make decisions, all have strong negative relationship with the female/male earnings ratio, i.e., less flexibility leads to greater earnings gap.

More illustrations are provided using two nonlinear occupations (MBA and law) and one linear occupation (pharmacist). Among MBAs graduating from the University of Chicago Booth School from 1990 to 2006, there is a 30 log point gap in annual earnings 5 years after the MBA and a 60 log point gap at 10-16 years after the MBA. Training prior to MBA accounts for 24%, career interruptions and job experience for an additional 30%, and differences in weekly hours account for an additionnal 30%. Among law graduates from the University of Michigan Law School from 1982 to 1991, the gap is nil at the start of employment, and is small at year 5, after controlling for hours, weeks and time off. The gap at year 15 amounts to 22 log points when working time is controlled and is further reduced to 13 log points once work absences and job tenure are added. Among full-year pharmacists, the adjusted gap is 5-7%. The gap declined over time due to several factors: a decrease in self-ownership, the rise of large corporation and hospital employment, a decreased cost to flexible employment. The profession is now one of the most egalitarian one but it is not due to legislation or anti-discrimination policy or licensing requirements or regulations specific to the pharmacy profession.

Time flexibility accounts for the narrowing of the pay gap over time but also for the much greater contribution of the within-occupation component (compared to between-occupation) to the gender gap, and also why there is a motherhood wage penalty. Self-employment has declined in a large number of professions the past several decades including dentists, lawyers, optometrists, pharmacists, physicians, and veterinarians. The decline has produced a reduction in the premium to long and unpredictable hours. Moreover, the rapidly growing sectors such as those in health and information technologies move in the direction of greater linearity.

Bütikofer et al. (2018, Figure 9) examine the earnings of top earners with an MBA, a law degree, a STEM degree, or a medical degree in Norway. Results from event study analysis show that women in professions with more nonlinear wage structures, such as MBA and law, suffer from a larger child earnings penalty (20+% after 10 years) in contrast to women in professions with a more linear wage structure, such as STEM and medicine.

Barth et al. (2021, Tables 3-4, 4A) examine the LEHD, a very large longitudinal data. Their AKM wage model, based on Abowd et al. (1999), includes individual, establishment fixed effects and time varying covariates. Among college workers, the gender gap across establishments widens by 11 log points from age 25 to 45, accounting for 30% of the widening of the gap. The importance of the establishment component also increases over time. The analysis by occupational category reveals a differential widening of the gap, with 32 log points in professional and management, 18 log points in service and sales, 4 log points in production and transportation. Management and professional jobs have far more need for establishing and maintaining interpersonal relationships, have more structured jobs and more discretion in making decisions. Production jobs have far fewer client and worker contacts and far fewer working relationships with others. Another important finding is that, within establishments, the widening of the gender pay by age is occurring entirely among college-educated workers. This is because college workers are typically found in career-oriented jobs characterized by competition over promotions and non-linear payment structure. When marital status is accounted for, the widening gap across age within-establishment occurs between married, rather than between non-married, men and women. Between establishments, however, the widening gap across age occurred only among married workers but not non-married workers.

Bertrand (2018, Table 2) analyzed the American Community Survey and showed that the median (but not the mean) elasticity of earnings to working hours increases in the higher-paying occupations but also increases from 1980 to 2000 but with no further changes up to 2015. The larger elasticity in higher-paying occupations contributed to the larger pay gap in these occupations.

Destefanis et al. (2023) employed Goldin’s (2014) measure of residual (i.e., unexplained) pay gap as well as Oaxaca-Blinder decomposition, based on Italian data. The residual pay gap increased as a function of the elasticity of earnings with respect to work hours, for both graduates and non-graduates. However, both of these variables are correlated with the occupational characteristics that impose earnings penalties on workplace flexibility only among graduate workers.

3. The interesting case of the wage gap in medicine.

Ramanan & Mathisen (2023) examine the gender gap in medicine in Norway, a profession supposedly more gender fair. Using regression with age, age^2, public-private sector, level of education, and fixed effects of county, they reported a gender gap of 16,2% in total wage and a gender gap of 5,7% in fixed wage. This result is driven by gender differences in overtime work which affects total wage (which includes overtime) but not fixed wage: “Men and women work nearly the same hours of overtime when planned three months in advance, but men are almost twice as likely to work overtime hours when offered the day before (Bolotnyy & Emanuel, 2018). This suggests that women are less spontaneous when accepting overtime, as they might have to fulfill family responsibilities, and it is reasonable to assume that many instances where overtime is necessary for a medical environment occur in emergencies, which are spontaneous and unpredictable.” (p. 52).

Weaver et al. (2015) document that the pay gap is partly due to differences in priorities. After controlling for age, pediatric specialty, practice model, geography, type of clinical work, and productivity measures, they found that substantial pay prioritization is associated with higher earnings. This is important since substantial pay was ranked higher by men, whereas collegiality and control over personal time were ranked higher by women. Differences in priorities may be governed by cultural norms. It is commonly argued that women may be perceived more negatively than men when they negotiate pay.

Whaley et al. (2021) compare US physician earnings, while controlling for the number of hours worked per week and annual amount billed to Medicare (proxy for clinical volume), physician specialty, practice type, year of survey completion, number of years since the completion of the last clinical training. The gender gap increases greatly during the first years of practice and varies greatly across specialty.

Ganguli et al. (2020) analyze the gender pay gap among Primary Care Physicians (PCPs) using the EHR 2011-2017 data. After adjustment for the physician’s age, academic degree, specialty, and number of sessions worked per week, female PCPs generated 10.9% less annual visit revenue than their male counterparts. The pay gap was due primarily to female PCPs providing 10.8% fewer visits, although female PCPs spent 16% more time with patients per visit (and overall). Female PCPs missed more opportunities to bill higher-intensity visit codes on the basis of time spent with patients.

4. Does gender composition matter?

The assertion that a higher share of females reduces the earnings gap received mixed evidence. Tomaskovic-Devey (1993) analyze the 1989 NCEHS data, using regressions by gender group, controlling for job, firm characteristics and race dummies. For the male sample, men’s log hourly wage is reduced by 0.003 (or 0.3%) for a 1% increase in female composition. For the female sample, women’s log hourly wage decreases by 0.001 but the coefficient is non-significant. Budig (2002) analyzed the NLSY79, using a fixed-effects model with job characteristics, industrial sector, human capital and labor supply characteristics, family and demographic characteristics. It was found that all interactions between gender and gender composition were small and non-significant. Penner et al. (2012) analyze a very large US grocery retailer over 9 years. The model, which includes fixed effects of race, establishment, job title and person, shows a coefficient of zero for both female manager and female manager interaction with gender variable. That is, the share of female manager does not impact the pay gap. Stojmenovska (2019) analyzed a British panel data, using a workplace-fixed model, adjusting for year, employee and workplace characteristics. It was found that 1% increase in the share of women managers reduces the gender gap by 0.325 (SE=0.111) pound sterling (£) per week and reduces men’s earnings by 0.245 (SE=0.139) pound sterling (£) per week but this latter coefficient is not significant despite the large sample size. The major cause of the reduced gender gap therefore is attributed to a reduction in men’s earnings rather than an increase in women’s earnings. The coefficient of 0.325 would translate to £3.25 per month for a 10% increase in the share of women managers (the average weekly wage in the sample is £335). Finley et al. (2022, Table 7) use a regression that controls for organization, position and year fixed effects and report that the gender gap in nonprofit organizations reduces by 3.2% (non-significant) if the percentage of female board members is above the median and by 6.5% if the CEO is female.

Cutillo & Centra (2017, Table 5) apply Oaxaca-Blinder model on Italian data (ISFOL). They add two components to the model: the family composition and selectivity effects. Due to the nonrandom allocation of jobs, related to gender differences in preferences, with men seeking pay grade and women seeking job security and employment benefits, ignoring self-selection leads to biased estimates. The pay gap is similar between female-dominated (15.6%) and other jobs (16%). The family composition contributes to 6.8% and 3% of the gap in female-dominated jobs and other jobs and therefore explains 45% (i.e., 6.8/15.6) and 19% (i.e., 3/16) respectively. After selectivity adjustment, the unexplained component rises to 36.5% in female-dominated jobs and falls to 8.8% in other jobs. These authors see this result as supportive of the discrimination theory because “employers take advantage of individuals’ inclinations and penalize on wages of women when hiring them in the jobs they desire to achieve a satisfactory work-life balance.” (p. 26). Again, this story lies on the fallacy that “unexplained” equals “discrimination”.

Library market is a good illustration because it is a field dominated by women (71-74%) and displaying small pay gap. Sweeper & Smith (2010) adjusted their model for marital status, children, degree, occupational and labor market factors, and reported a coefficient of -0.094 for women in the (log) income regression, which translates to EXP(0.094)=0.910, i.e., 9% gap. But the coefficient is very imprecise, owing to the very large standard errors. Galbraith et al. (2019) found that women initially make 93.40% of men’s earnings. After adjusting for only experience, type of library, position, women make 97.80% of men’s earnings, and when adding marital status, children, and education, women make 98.57% of men’s earnings. In light of these findings, one must ask whether gender composition relates to gender pay gap. Sweeper & Smith (2010) informed us that women made substantial gains when they held only a small portion of leadership positions but also that library science has experienced substantial changes due to technological developments such as computing, automation, and the growth of Internet. The reduction in pay gap may instead be due to structural changes.

5. Gig economy also shows the gender gap.

There are two ongoing arguments suggesting that gig economy (i.e., independent workers) helps reducing earnings gap. First, because employers are suspected to discriminate against women. Second, gig economy generates flexibility which favors women. Cook et al. (2020) analyze the Uber drivers, who have full discretion regarding when and where they work. They document a 7% earnings gap. Based on Gelbach’s (2016) wage gap decomposition, they found that the gender gap is fully explained by returns to experience, pay premium for faster driving, and differences in driving locations. Importantly, they do not find women to be disadvantaged by three factors expected to contribute to a gender wage gap: returns to work intensity, preferences or constraints affecting specific hours worked, or customer discrimination.

Litman et al. (2020) investigate an anonymous online microtask platform such as Amazon Mechanical Turk (MTurk) which became a major sector of the gig economy. MTurk connects employers (requesters) to employees (workers) who perform jobs called Human Intelligence Tasks (HITs). Requesters disclose tasks on a dashboard with a short description of the HIT, the compensation being offered, and the time the HIT is expected to take. The gender of workers who complete these HITs is not known to the requesters. The mean pay across gender is $5.70 per hour, with women earning 60 cents (i.e., 10.5%) less than men. The pay gap dropped to 46 cents after adjusting for advertised hourly pay and to 32 cents when also adding covariates such as race, marital status, number of children, age, HIT acceptance speed and number of HIT. When analyzed individually, marital status, children, and experience do not account for the gender gap at all. Overall, such findings indicate that the large gender gap is partially explained by women selecting tasks with lower advertised pay and having lower task completion speed, but with no room left for discrimination.

6. The impact of sorting effects.

Sorting effect refers to the differential selection into firms on the basis of gender. Card et al. (2016, Tables 3 & B6) employ wage regression based on Oaxaca-Blinder model, which allows decomposition of person and firm effects, using Portuguese data. This model, controlling for year, education, and age*education dummies, shows that firm-specific pay premiums explain 20% of wage variation among both men and women, while positive assortative matching (i.e., the positive correlation between person and firm effects) explains another 10%, with sorting effects (i.e., the difference in male wage premiums weighted by the shares of men versus women at each firm) explaining 15-20% and relative bargaining effects (i.e., the male-female difference in wage premiums weighted by the fraction of men at each firm) explaining 10-15% of the gender gap. Furthermore, sorting effects rise with age and are more important among less educated workers, while bargaining effects are larger for highly-educated workers. Importantly, women are disproportionately likely to work at low surplus firms paying small premiums to both genders.

Sin et al. (2020, Table 3) examine the role of firm effects in the gender gap in New Zealand. The wage model include many controls such as demographics, qualifications, household composition, number of children, tenure, job-match characteristics, firm characteristics, ethnicity, and fixed effects of regional council, occupation, industry. A large gender gap is found, of about 20-28%. Occupation explains 9%, across industries 16.4-18.5% and across firms 5-9% of this baseline gap. Then, a decomposition analysis of the total effect is conducted: firm sorting explains 23.4% of the overall gap whereas the within-firm gender differences in wage premia explains 22.2% (43.2%) of the overall (within-firm) wage gap.

Glaubitz et al. (2022, Figures 7-9) use an Oaxaca-Blinder decomposition to analyze the lifetime earnings in the SOEP data in Germany. The model controls for marital status, experience, part time, education, cohort and sector. In the early 20s, the gender gap in hourly wages shows a small difference but widens quickly over the life cycle. By the end of the work life, 90% of the overall gap is due to differences in work experience. The gender gap in annual earnings is even larger but is once again driven by the lesser experience accumulated. Because cross-sectional analysis does not reveal how advantages or disadvantages add up or balance out over the lifecycle, they employ a microsimulation model which simulates missing observations. For women with children the earnings gap is stable over the work like while for women without children the gap is nonexistent at the beginnning but grows with increasing age. Possible causes could be less promotion or self selection into sectors with lower earnings growth potential (e.g., health care).

Gulyas et al. (2021, Table 3) found in Austria that, when the pre-treatment (i.e., years before the transparency law was rolled out) gender gap is decomposed into within-firm and sorting effects, the firm pay policies explain 28-39% of the gap while sorting effects explain 61-72% of the gap. This is consistent with Card et al. (2016) and Morchio & Moser (2019) who found very little effect of firm pay policies on the gender gap in Portugal and Brazil.

Coudin et al. (2018, Figure 8) investigate bargaining and sorting effects in France using a Oaxaca-Blinder decomposition. They found a sorting effect accounting for 11 % of the hourly wage gap while bargaining effect is close to zero. The sorting effect dramatically increases after the first childbirth. The absence/reduction of bargaining effect is due to France having higher minimum wages.

Boll et al. (2017) analyze European countries and found that in all countries (except for two) the male wage premiums are particularly pronounced in those sectors where female workers are over-represented. This sorting explains a huge portion of the gender gap. Much of the sorting seems to take place within rather than between industries.

7. Marital status and child penalty.

Cukrowska-Torzewska & Matysiak (2020a) estimate a meta-analytic wage penalty of 3.6% for each additional children among multiple child mothers and a wage penalty of 3.8% for mothers with exactly one child.

Bertrand et al. (2010, Tables 3-4) examine MBA graduates between 1990 to 2006 from the University of Chicago. The initial gap of 28.7 log points is reduced to 3.8 points after accounting for weekly hours worked, finance classes taken, MBA GPA, pre-MBA characteristics, MBA experience, job function, employer type, cohort*year interaction. The earnings gap (without controls) increases with the number of years, from 8.9 points at year 0 to 56.5 points at year 10-16 after MBA completion, the addition of the controls reduced the gap to nearly zero regardless of the years. Three salient factors accounting for the growing earnings gap over time are: differences in training prior to MBA graduation, differences in career interruptions, and growing differences in weekly hours worked, the latter two being associated with motherhood.

Zhang & Hannum (2015) examine the evolution of gender wage gap in China, using the CHNS data from 1989 to 2009. They employ generalized estimating equation (GEE) models of earnings that account for multiple observations within the same individual and employ Heckman models to include a control to correct for selection into employment. Married women’s wage increases at lower pace than married men’s over time, however single women’s wage increases at similar pace than single men’s over time. This indicates that marital status plays a huge role in the wage gap.

Bronson & Thoursie (2017) use employer-employee matched data to analyze within- and across-firm effects on men’s and women’s wages over time in Sweden between 1985 and 2013. Their wage AKM decomposition model based on Abowd et al. (1999), with year and firm fixed effects, shows that differences in promotion-related growth account for around 80% of the wage growth gap. These wage gains are marginally higher for men, which means that promotion-related growth is driven mainly by gender differences in the probability of receiving a promotion. Moreover, 29% of this promotion probability gap occurs during or immediately after childbirth while another 21% is due to part-time work.

Goldin et al. (2022) analyze the motherhood penalty over time using the NLSY79. Their OLS regression includes fixed effects of age, number of children and their interaction with gender variable, and further controls for unemployment rate and work experience. They found that mothers’ working hours increase relative to those of non-mothers and fathers when the children get older and become more independent. Although mothers narrow the earnings gap with respect to non-mothers, mothers do not reduce the gap with respect to fathers conditioning on hours and weeks.

Garbinti et al. (2023) analyzed the lifetime earnings in France between the 1967 and 1987 birth cohorts, based on the french EDP 2019 data. The Oaxaca-Blinder decomposition model shows that the unexplained factors declined over time, from 60% to 20-35%, due to the declining gender gap in characteristics. Much of the gap is due to differences in working time (30% and 60% for the 1967 and 1987 cohort). Another key aspect in closing the gender earnings gap is the increase in educational attainment among women.

Kleven et al. (2019) analyze the administrative data in Denmark between 1980-2013. They use the Oaxaca-Blinder decomposition while adding within-person time variation around childbirth (i.e., an event study approach). The decomposition shows that the fraction of gender gap attributed to children increased from 40 to 80% between 1980 and 2013, and is robust to controls of education choices made prior to childbirth. This robustness results from the event study variation they use: the education effect is obtained from cross-sectional variation and therefore does not absorb the child effect obtained from within-person time variation. The remaining gap is now all about children because of the falling gender gap in education. To investigate how the child penalty is transmitted through generation, the grandparents’ labor supply level is included in the model while also controlling for the grandparents’ characteristics. They found that child penalty on women is strongly correlated with the division of labor in their own childhood homes, but not with the division of labor in their spouses’ childhood homes.

Gørtz et al. (2023) observe grandparents earnings gap 5 years before and after the arrival of their first grandchild in Denmark. They use an event study approach, controlling for individual and age fixed effects (as dummies), which relies on the smoothness assumption that the changes in individual, family, and work preferences evolve gradually over an individual’s life course as opposed to the sudden event that is the grandchild birth. Five years after the arrival of the grandchild, the earnings gap between grandmothers and grandfathers amounts to 3.8%, grandmothers are 4.2% less likely to work full time than grandfathers, but there is no differential effect on wage rates. The earnings gap increases from 3.8% after 5 years to 10.1% after 10 years. Overall, this finding leads the authors to conclude that child penalty may strike twice.

Similarly, Meng et al. (2023, Figures 2 & 6-7) examine the chinese CFPS data, and found that upon the arrival of a child, the first-time grandmothers strongly reduce their labor supply to provide care for the child. The event study analysis shows that the mothers experience a large drop in labor force, hours worked but the gap quickly vanished after 6-7 years, but not so much for monthly earnings (although the men and women’s estimates overlap). Two findings are worth noting. First, the recovery of women’s labor force participation, hours worked and earnings after childbirth however is due to receiving grandparenting. Second, the recovery of women’s outcomes occurs only among women with one child, not for women with multiple children. Finally, the authors applied Kleven et al.’s procedure for decomposing the gender gap into children and background and education component. They estimate that 29% of the gap is due to parenthood, which is much lower than the estimates of 64% and 83% reported in earlier studies for the US and Denmark, respectively. This could be due to the lower fertility rate in China since 80% of this sample is composed of one-child mothers.

Differential motherhood effect is indeed observed across countries. Cukrowska-Torzewska & Lovasz (2020b, Table 2) examine the parenthood effect across 26 european countries with varying institutional contexts. They divide the gender gap into 3 components: the family gap or parenthood effect (computed using the difference between parent and non-parent) among women and men, the wage gap among non-parents, and then decompose each of these components using the Oaxaca-Blinder method. The wage models adjust for education, age, marital status, regional disparities, time fixed effects. The fatherhood wage premium exists in all countries with some variation. The motherhood wage penalty varies greatly across countries as it contributes marginally to the gender gap in Nordic countries but to a much greater degree in Eastern Europe whereas a motherhood wage premium instead is observed in Southern Europe, except Portugal, and its contribution to the gender gap is the largest. The magnitude of the motherhood penalty might be moderated by childcare access.

8. Gender difference in psychological factors.

The impact of personality on earnings prospects is rarely discussed. Bertrand (2018, pp. 211-214) published a literature review. An important factor is the gender difference in risk aversion, along with the higher income volatility associated with high income jobs, causing women to stay away from these jobs. Even after accounting for differences in risk aversion, a residual gender gap in willingness to compete remains. Perhaps the most interesting cited study is from Reuben et al. (2015) who found that MBA female graduates, despite having expressed interest in obtaining a job in the competitive business sector, are less willing to compete than male students. While industry choice explains 40% of the widening MBA gap (from 11% to 39%), preferences for competition strongly correlates with industry choice. Bertrand concludes that the studies, taken together, likely account for 10% of the gender gap at best. Blau & Kahn (2017, Table 7) provides a summary of these studies, although they often use different personality measures, e.g., Big Five, locus of control, competition, self-confidence, these traits often account for 4-14% of the gender gap. Roethlisberger et al. (2023, Table 6) provides the most updated literature review. Here again, the effect varies a lot across studies, with Big Five accounting for 1.4% to 24% of the gender gap. The heterogeneity of effect being this huge implies that more research is needed to identify its relevance. A less discussed issue in these reviews is that personality measures are less reliable than background measures such as education, hours, and marital status.

Lower confidence in math and science ability among women explain a good portion of the STEM pay gap. Sterling et al. (2020) use path mediation analysis, with controls for industry, major, GPA, degree, year of graduation, having a second degree, and previously having internship with the current employer. They found that self-efficacy, rather than belief in workplace culture or socialized pay preferences, partially mediates the gender pay gap at workforce entry among STEM graduates.

Using survey data on job search behavior, Cortes et al. (2022, Table 9) found that women accept jobs earlier than comparable men. Both field and experimental settings showed that a non-trivial portion of the gender gap in reservation wage (i.e., smallest wage at which a worker is willing to accept a job) at the early career stage is attributed to men’s greater degree of risk tolerance and overconfidence relative to women.

A curious case is presented by Azmat & Ferrer (2016, Tables 8-10) who document the gender gap in performance and earnings among young associate lawyers in the US. In their research, career aspiration appears to be very important: “When asked to rate, on a scale from 1 to 10, their aspirations to become an equity partner in their firm, 60 percent of male lawyers answered with 8 or more, compared to only 32 percent of female lawyers” (p. 21). Their models include individual, firm, education controls, region fixed effects. None of the measures of discrimination used, e.g., the extent to which lawyers interact with the firm’s clients, self-reported measures of discrimination, or in the assignment of cases, could explain the gender gap in performance (measured with hours billed and client revenue). Female lawyers with young children bill 200 fewer hours per year, while male lawyers with young children do not decrease their hours billed, however childrearing does not help explain the gender gap in client revenue. Career aspirations almost fully explain the gender gap in hours billed and client revenue. Their final analysis shows that individual and firm characteristics explain 50% of the earnings gap, while the inclusion of performance measures explains a substantial share of the remainder. Clearly here, discrimination is a poor explanation of the earnings gap.

9. Negotiation also explains the gender gap.

Finley et al. (2022, Tables 4-5) study the role of negotation in the pay gap among nonprofit organizations, based on the SOI data. They use external employment options as a measure of competition intensity (from either nonprofit or for-profit firms) since competition for labor should increase compensation negotiability. Using a specification which controls for organization, position and year fixed effects, they found that the pay gap is larger by 5.2% among firms with high nonprofit competition relative to low nonprofit competition firms and 6.4% among firms with high for-profit competition relative to low for-profit competition firms.

Kiessling et al. (2024) examine the role of boldness in initial wage claims among German students and found that a 20 percentile rank increase of boldness reduces the gap by half, using a specification controlling for major, industry and year fixed effects.

Roussille (2023) analyzes data from, a leading online recruitment platform for fulltime, high-wage engineering jobs, which records previously unexplored components of the salary negotiation process. First, every candidate has to provide the salary they are looking for in their next job. This ask salary is visible to firms recruiting on the platform, along with the candidate’s resume information. Second, companies signal their interest to candidates with a bid salary, indicating how much they would be willing to pay the candidate before interviewing them. Last, the platform records a final salary if the candidate is hired. The initial ask gap is 6.6%, which is reduced to 2.9% after controlling for resume characteristics. Using a model that captures the greatest share of heterogeneity, the ask gaps range from 8.5% to -2.1%, with the largest gap arising among candidates who are not currently employed, have more experience, and fewer credentials. Roussille then documents the relationship between the ask salary and firms’ bid and final offer gaps. The initial bid gap is 3.3%, which is reduced to 2.2% after adjusting for resume characteristics but excluding the ask salary. When ask salaries are adjusted, and even when resume characteristics are not, the bid gap disappears. Similarly, for a given job, resume characteristics account for 3 percentage points of the 4.8% unadjusted bid gap, while further controlling for the ask salary brings the bid gap to zero, indicating that the bid gap doesn’t arise from the composition of jobs for which women interview.

10. Unintended consequences of transparency laws.

Transparency law is designed to reduce the gender pay gap that is typically seen as the product of discrimination. Bennedsen et al. (2023) reviewed a large body of research, concluding that the effect on the gap is small, typically by about 1-2%. In particular, one study found that the effect of transparency laws is more pronounced in unionized workplaces, likely due to the public shaming of firms with high gender gap.

Boheim & Gust (2021) examine the 2011 transparency law in Austria, which requires firms with more than 1,000 employees to publish a pay report (that is company secret), but which also affected smaller firms the following years. The law does not punish a firm if it fails to publish the reports of if there are systematic differences in pay. Results from regression discontinuity shows that the law did not reduce the gap generally, except for newly hired women who received a wage increase in firms above the 1,000 employee cutoff. However, the share of female employees declined in firms above the 1,000 employee cutoff relative to firms below this cutoff. This suggests that firms reacted to the increase in salary costs by reducing the number of female employees. Gulyas et al. (2021) analyze the same data but employ an event-study model, with 2013 as the post-treatment year. Treated and control firms are specified as having ≥150 and less <150, respectively. There is no distinct evolution of the wage gap in treated versus control firms. The most important finding is that the act has reduced job separation rate in treated firms relative to control firms by over 1.1%, which may suggest that job satisfaction has increased.

There are numerous examples of such unintended consequences. Mas (2016) analyze a large sample of city managers in California, who are in charge of day-to-day operations in the city and play an important role in generating tax revenues and providing public services efficiently. Wage cuts were higher in cities where compensation was initially higher, particularly cities where the city managers were paid more than $200k annually. There was no relative decline in the 50th, 75th, and 90th percentiles of the city wage distributions, implying that reductions at the top of the wage distribution reflect pay compression (i.e., decrease in variance). Pay disclosure increased job quit rate by 75% and decreased by 20% the rate of new hires of experienced managers. Mas warned against the tendency of such populist policies to downgrade the quality of the hired managers on the basis of an aversion to large salaries. Obloj & Zenger (2022) examine the pay gap among US academics, using a DID specification controlling for academic tenure, field, institution and year fixed effects. After transparency shock, pay dispersion decreased by nearly 20% and the sensitivity of pay to observed performance as well as the financial rewards associated with promotion across academic ranks both declined substantially.

So far most studies ignored what should have been the most glaring unintended consequence: pay compression, as a result of transparency, increases hidden remuneration. Wong et al. (2023) provided such evidence in their analysis of Chinese firms. Their multilevel SEM is composed of the firm-level predictor (pay transparency), firm-level mediator (variable pay compression), two mediators measured at the employee level (benefit requests) and firm-level outcomes (benefit grants). After controlling for number of employees, firm revenue, tenure, age, education, gender, they found that transparency leads to pay compression which then triggers requests for hidden benefits such as extra vacation days, additional training, supplementary health benefits, collectively termed non-monetary idiosyncratic deals (i-deals). These authors argue that employers decide to reduce pay differences because it is often time consuming and psychologically draining for them to address employee complaints and salary adjustment requests.

11. Family policies fail?

Given the huge impact of childbirth on the earnings gap, economists often proposed family policies to reduce the gap. Using an event-study analysis, Kleven et al. (2020) showed that the enormous expansions of parental leave and child care subsidies over more than 60 years in Austria had no long-term impact on the gender convergence. This finding reinforces the idea, according to several researchers, that the convergence in earnings gap reflects the evolution of gendered preferences and norms. The authors conclude that “This stands in contrast to the strong cross-country correlation between child care subsidization and gender gaps, while within-country quasi-experimental studies of specific reforms have led to mixed results.” (p. 32).


  1. Azmat, G., & Ferrer, R. (2017). Gender gaps in performance: Evidence from young lawyers. Journal of Political Economy, 125(5), 1306-1355.
  2. Barth, E., Kerr, S. P., & Olivetti, C. (2021). The dynamics of gender earnings differentials: Evidence from establishment data. European Economic Review, 134, 103713.
  3. Bennedsen, M., Larsen, B., & Wei, J. (2023). Gender wage transparency and the gender pay gap: A survey. Journal of Economic Surveys.
  4. Bertrand, M. (2018). Coase lecture–the glass ceiling. Economica, 85(338), 205-231.
  5. Bertrand, M., Goldin, C., & Katz, L. F. (2010). Dynamics of the gender gap for young professionals in the financial and corporate sectors. American economic journal: applied economics, 2(3), 228-255.
  6. Blau, F. D., & Kahn, L. M. (2017). The gender wage gap: Extent, trends, and explanations. Journal of economic literature, 55(3), 789-865.
  7. Böheim, R., & Gust, S. (2021). The Austrian pay transparency law and the gender wage gap.
  8. Boll, C., Rossen, A., & Wolf, A. (2017). The EU gender earnings gap: Job segregation and working time as driving factors. Jahrbücher für Nationalökonomie und Statistik, 237(5), 407-452.
  9. Bronson, M. A., & Thoursie, P. S. (2017). The lifecycle wage growth of men and women: Explaining gender differences in wage trajectories. Georgetown University, Manuscript.
  10. Budig, M. J. (2002). Male advantage and the gender composition of jobs: Who rides the glass escalator?. Social problems, 49(2), 258-277.
  11. Bütikofer, A., Jensen, S., & Salvanes, K. G. (2018). The role of parenthood on the gender gap among top earners. European Economic Review, 109, 103-123.
  12. Card, D., Cardoso, A. R., & Kline, P. (2016). Bargaining, sorting, and the gender wage gap: Quantifying the impact of firms on the relative pay of women. The Quarterly journal of economics, 131(2), 633-686.
  13. Cook, C., Diamond, R., Hall, J. V., List, J. A., & Oyer, P. (2021). The gender earnings gap in the gig economy: Evidence from over a million rideshare drivers. The Review of Economic Studies, 88(5), 2210-2238.
  14. Cortés, P., Pan, J., Pilossoph, L., Reuben, E., & Zafar, B. (2023). Gender differences in job search and the earnings gap: Evidence from the field and lab. The Quarterly Journal of Economics, 138(4), 2069-2126.
  15. Coudin, E., Maillard, S., & Tô, M. (2018). Family, firms and the gender wage gap in France (No. W18/01). IFS Working Papers.
  16. Cukrowska-Torzewska, E., & Lovasz, A. (2020b). The role of parenthood in shaping the gender wage gap–A comparative analysis of 26 European countries. Social Science Research, 85, 102355.
  17. Cukrowska-Torzewska, E., & Matysiak, A. (2020a). The motherhood wage penalty: A meta-analysis. Social science research, 88, 102416.
  18. Cutillo, A., & Centra, M. (2017). Gender-based occupational choices and family responsibilities: The gender wage gap in Italy. Feminist Economics, 23(4), 1-31.
  19. Destefanis, S., Mazzotta, F., & Parisi, L. (2023). Goldin’s Last Chapter on the Gender Pay Gap: An Exploratory Analysis Using Italian Data. Work, Employment and Society, 09500170221143724.
  20. Finley, A. R., Hall, C. M., & Marino, A. R. (2022). Negotiation and executive gender pay gaps in nonprofit organizations. Review of Accounting Studies, 27(4), 1357-1388.
  21. Galbraith, Q., Callister, A. H., & Kelley, H. (2019). Have academic libraries overcome the gender wage gap? An analysis of gender pay inequality. College & research libraries, 80(4), 470-484.
  22. Ganguli, I., Sheridan, B., Gray, J., Chernew, M., Rosenthal, M. B., & Neprash, H. (2020). Physician work hours and the gender pay gap—evidence from primary care. New England Journal of Medicine, 383(14), 1349-1357.
  23. Garbinti, B., García-Peñalosa, C., Pecheu, V., & Savignac, F. (2023). Trends and Inequality in Lifetime Earnings in France.
  24. Glaubitz, R., Harnack-Eber, A., & Wetter, M. (2022). The gender gap in lifetime earnings: The role of parenthood.
  25. Goldin, C. (2014). A grand gender convergence: Its last chapter. American economic review, 104(4), 1091-1119.
  26. Goldin, C. (2015). How to achieve gender equality. Milken Institute Review Q, 3, 24-33.
  27. Goldin, C., Kerr, S. P., & Olivetti, C. (2022). When the Kids Grow Up: Women’s Employment and Earnings across the Family Cycle (No. w30323). National Bureau of Economic Research.
  28. Gørtz, M., Sander, S., & Sevilla, A. (2023). Does the Child Penalty Strike Twice, and If So Why?.
  29. Gulyas, A., Seitz, S., & Sinha, S. (2023). Does pay transparency affect the gender wage gap? Evidence from Austria. American Economic Journal: Economic Policy, 15(2), 236-255.
  30. Huber, M. (2015). Causal pitfalls in the decomposition of wage gaps. Journal of Business & Economic Statistics, 33(2), 179-191.
  31. Huber, M., & Solovyeva, A. (2020). On the sensitivity of wage gap decompositions. Journal of Labor Research, 41, 1-33.
  32. Kiessling, L., Pinger, P., Seegers, P., & Bergerhoff, J. (2024). Gender differences in wage expectations and negotiation. Labour Economics, 102505.
  33. Kleven, H., Landais, C., & Søgaard, J. E. (2019). Children and gender inequality: Evidence from Denmark. American Economic Journal: Applied Economics, 11(4), 181-209.
  34. Kleven, H., Landais, C., Posch, J., Steinhauer, A., & Zweimüller, J. (2020). Do family policies reduce gender inequality? Evidence from 60 years of policy experimentation (No. w28082). National Bureau of Economic Research.
  35. Litman, L., Robinson, J., Rosen, Z., Rosenzweig, C., Waxman, J., & Bates, L. M. (2020). The persistence of pay inequality: The gender pay gap in an anonymous online labor market. PloS one, 15(2), e0229383.
  36. Mas, A. (2017). Does transparency lead to pay compression?. Journal of Political Economy, 125(5), 1683-1721.
  37. Meng, L., Zhang, Y., & Zou, B. (2023). The motherhood penalty in China: Magnitudes, trends, and the role of grandparenting. Journal of Comparative Economics, 51(1), 105-132.
  38. Morchio, I., & Moser, C. (2019, June). The gender gap: Micro sources and macro consequences. In 2019 Meeting Papers (Vol. 143). Society for Economic Dynamics.
  39. Navarová, T. (2022). International gender wage gap: A Meta-analysis.
  40. Obloj, T., & Zenger, T. (2022). The influence of pay transparency on (gender) inequity, inequality and the performance basis of pay. Nature Human Behaviour, 6(5), 646-655.
  41. Penner, A. M., Toro-Tulla, H. J., & Huffman, M. L. (2012). Do women managers ameliorate gender differences in wages? Evidence from a large grocery retailer. Sociological Perspectives, 55(2), 365-381.
  42. Ramanan, A., & Mathisen, V. (2023). The Gender Wage Gap in Medicine: An empirical study on the gender wage gap among physicians in Norway (Master’s thesis).
  43. Reuben, E., Sapienza, P., & Zingales, L. (2015). Taste for competition and the gender gap among young business professionals. NBER Working Paper, (w21695).
  44. Roethlisberger, C., Gassmann, F., Groot, W., & Martorano, B. (2023). The contribution of personality traits and social norms to the gender pay gap: A systematic literature review. Journal of Economic Surveys, 37(2), 377-408.
  45. Roussille, N. (2021). The central role of the ask gap in gender pay inequality. University of California, Berkeley.
  46. Shatnawi, D., Oaxaca, R., & Ransom, M. (2014). Movin’on up: Hierarchical occupational segmentation and gender wage gaps. The Journal of Economic Inequality, 12, 315-338.
  47. Sin, I., Stillman, S., & Fabling, R. (2022). What Drives the Gender Wage Gap? Examining the Roles of Sorting, Productivity Differences, Bargaining, and Discrimination. Review of Economics and Statistics, 104(4), 636-651.
  48. Sterling, A. D., Thompson, M. E., Wang, S., Kusimo, A., Gilmartin, S., & Sheppard, S. (2020). The confidence gap predicts the gender pay gap among STEM graduates. Proceedings of the National Academy of Sciences, 117(48), 30303-30308.
  49. Sweeper, D. L., & Smith, S. A. (2010). Assessing the impact of gender and race on earnings in the library science labor market. College & Research Libraries, 71(2), 171-183.
  50. Stojmenovska, D. (2019). Management gender composition and the gender pay gap: evidence from British panel data. Gender, Work & Organization, 26(5), 738-764.
  51. Tomaskovic-Devey, D. (1993). The gender and race composition of jobs and the male/female, white/black pay gaps. Social Forces, 72(1), 45-76.
  52. Weaver, A. C., Wetterneck, T. B., Whelan, C. T., & Hinami, K. (2015). A matter of priorities? Exploring the persistent gender pay gap in hospital medicine. Journal of hospital medicine, 10(8), 486-490.
  53. Whaley, C. M., Koo, T., Arora, V. M., Ganguli, I., Gross, N., & Jena, A. B. (2021). Female Physicians Earn An Estimated $2 Million Less Than Male Physicians Over A Simulated 40-Year Career: Study examines estimated career gap in pay between female physicians and male physicians. Health Affairs, 40(12), 1856-1864.
  54. Wilcox, V., & Forhad, M. A. R. (2023). Reverses in Gender Salary Gaps Among STEM Faculty: Evidence from Mean and Quantile Decompositions. The BE Journal of Economic Analysis & Policy, 23(4), 943-979.
  55. Wong, M. N., Cheng, B. H., Lam, L. W. Y., & Bamberger, P. A. (2023). Pay Transparency as a Moving Target: A Multistep Model of Pay Compression, I-Deals, and Collectivist Shared Values. Academy of Management Journal, 66(2), 489-520.
  56. Zhang, Y., & Hannum, E. (2015). Diverging fortunes: The evolution of gender wage gaps for singles, couples, and parents in China, 1989–2009. Chinese Journal of Sociology, 1(1), 15-55.