Skill Level

Paper finds no evidence that participation in high school sports has positive spillovers on wages, LFP or college attendance, highly correlated but its all self-selection. Core of paper, "...We examine the extent to which participation in high school athletics has beneficial effects on future education, labor market, and health outcomes. Due to the absence of plausible instruments in observational data, we use recently developed methods that relate selection on observables with selection on unobservables to estimate bounds on the causal effect of athletics participation. We analyze these effects in the US separately for men and women using three different nationally representative longitudinal data sets that each link high school athletics participation with later-life outcomes. We do not find consistent evidence of individual benefits reported in many previous studies - once we have accounted for selection, high school athletes are no more likely to attend college, earn higher wages, or participate in the labor force. However, we do find that men (but not women) who participated in high school athletics are more likely to exercise regularly as adults. Nevertheless, athletes are no less likely to be obese...."
Their evidence

Positive outcomes driven by selection,"...Athletic participation is strongly positively correlated with a number of outcomes—including high school graduation, college attendance, college graduation, wages, exercise habits, and absence of obesity—but we find that this correlation is almost completely due to selection. For most of the outcomes that we consider, we find that even if the correlation between athletic participation and unobservable characteristics is a small fraction of the correlation between athletic participation and observable characteristics, then there is no effect of sports.Across several different outcomes and different samples, we find no consistent benefit from high school sports. However, in a few cases that we discuss below, we do find statistically significant effects from sports participation that are arguably causal...."

“…In Table 1 and Table 2 we present basic descriptive statistics of athletes and non-athletes for men and women, respectively. Each table contains a list of outcomes and control variables from each survey, along with the respective sample means for non-athletes, athletes, and the full sample. We denote with an asterisk sample means of athletes that are statistically different from non-athletes at the 5% level. The descriptive statistics provide a high-level understanding of how selection on observables and unobservables might mitigate the effects of sports. We divide the variables into three categories: background characteristics, school characteristics, and outcomes…Athletes tend to have higher cognitive test scores, be disproportionately white, have parents with higher levels of education, be more likely to co-reside with parents, and come from homes with higher incomes. In short, our basic summary statistics reveal that athletes are strongly positively selected on personal and family background traits….On the school side, athletes are less likely to be absent from school, more likely to be found in private schools and schools with smaller student bodies, more likely to be found in rural schools, and more likely to attend schools that are more racially segregated. These results hold for both men and women, and are in line with existing literature and theory. Namely, overwhelmingly white, private, and rural schools provide more opportunities for student athletes, for whatever reason. Possible explanations include differences in school funding, or that it is statistically easier to make the team at a school with a smaller student body….In addition to observing that athletes have different background and school contexts, we also observe that athletes have very different adult outcomes. They attain higher levels of education, measured either by grades completed or degrees attained. Athletes also earn more as adults: about 15 percent higher wages for men, and about 12 percent higher wages for women. Athletes are much more likely to report exercising regularly. Male athletes are neither more nor less likely to be obese as adults, while female athletes are much less likely to be obese. Athletes of both genders report a higher frequency of alcohol abuse as adults….The results in Tables 1 and 2 are striking in that the different surveys exhibit not only the same sign of sports effects, but also many of the same magnitudes, in spite of the fact that athletic participation is measured quite differently across the three surveys….”
Educational outcomes

“….Tables 4 and 5 contain estimates of α, the effect of sports participation on educational outcomes, from all three surveys, respectively for men and women. The first two rows of the table report OLS estimates, while the remaining rows report RCR estimates. Consistent with the large body of prior research on the topic, as well as the evidence reported in Tables 1 and 2, when we assume that students are randomly assigned to be athletes, we find economically large and statistically significant effects of sports participation on high school graduation, college attendance, and college graduation. Furthermore, the inclusion of explanatory variables dramatically reduces the size of the estimated effect, by more than half in some cases. The estimated effects are slightly smaller for women, but the general pattern is very similar.....thus, the correlation between sports participation and college graduation propensity for men in the NELS:88 is robust to a large amount of selection on unobservables, although we would not reject a null hypothesis of a null effect if λ = 1. On the other hand, for women we find little evidence to support a causal effect of sports participation on the educational outcomes that we examine. For all three of these samples, even if the correlation with the unobservables is only half as much as the correlation with the observables, there is no beneficial effect of sports participation….”

“…Tables 8 and 9 contain estimates for our measures of health and risky behaviors: regular exercise, obesity, and alcohol abuse. In the case of regular exercise, both surveys indicate that male athletes are more likely to participate in regular adult exercise. Interestingly, after controlling for observables, the athlete/non-athlete difference does not change much (and actually increases in the Add Health sample). This result is graphically depicted in Figure 1 (b) for the NELS:88….”

Their takeaway, "...Our largely null results inform the policy debate on high school sports by providing evidence against claims that sports foster skills that improve educational or labor market outcomes. However, despite having very little human capital value, sports may still have a place in high school as a social or cultural activity. We generally confirm the assertion of sports commentator Heywood Hale Broun, who said, “Sports reveals character, it doesn’t build it”

Michael Ransom and Tyler Ransom, "Do High School Sports Build or Reveal Character?Institute of Labor Economics, October 2017, https://docs.iza.org/dp11110.pdf

Summery, “…Intelligence test scores and educational duration are positively correlated. This correlation could be interpreted in two ways: Students with greater propensity for intelligence go on to complete more education, or a longer education increases intelligence. We meta-analyzed three categories of quasi-experimental studies of educational effects on intelligence: those estimating education-intelligence associations after controlling for earlier intelligence, those using compulsory schooling policy changes as instrumental variables, and those using regression-discontinuity designs on school-entry age cutoffs. Across 142 effect sizes from 42 data sets involving over 600,000 participants,we found consistent evidence for beneficial effects of education on cognitive abilities of approximately 1 to 5 IQ points for an additional year of education.Moderator analyses indicated that the effects persisted across the life span and were present on all broad categories of cognitive ability studied. Education appears to be the most consistent, robust, and durable method yet to be identified for raising intelligence….”

Stuart Ritchie and Elliot Tucker-Drob, "How Much Does Education Improve Intelligence? A Meta-Analysis,"Psychological Science, 2018, https://labs.la.utexas.edu/tucker-drob/files/2019/08/Ritchie-Tucker-Drob-2018-Psych-Science-How-Much-Does-Education-Improve-Intelligence.pdf

“….Table 2 shows the results from each study design after the inclusion of multiple moderators. We included as many moderators as possible for each design, though we chose to include the “achievement” categorization of the outcome test for the school-age-cutoff design (instead of the alternative categorization involving composite, fluid, and crystallized tests, which we used for the other designs) because it had such a substantial effect in the single-moderator model. Including multiple moderators reduced the τ statistic—to a larger degree for the control-prior-intelligence design than for the others— though significant heterogeneity remained in all cases. The moderators that were individually significant (e.g., age in the control-prior-intelligence design or achievement tests in the school-age-cutoff design) were also significant in the multiple-moderator model, indicating that their effects were incremental of the other moderators that we included….”

“…We tested whether studies that included only male participants showed a differential educational effect. This was not the case for control prior intelligence (effect for the 2 male-only estimates: 2.261 IQ points, SE = 0.897, p =.012; effect for 24 mixed-sex estimates: 1.027 IQ points, SE = 0.110, p = 6.79×10−21; difference SE = 0.905, p =.173), or for policy change (effect for the 8 male-only estimates: 1.683 IQ points, SE = 0.507, p =.001; effect for 22 mixed-sex estimates: 2.215 IQ points, SE = 0.788, p =.005; difference SE = 0.941, p =.572). There were no male-only school-age-cutoff studies….”

Multiple moderators.

“….Splitting the outcome cognitive tests into three broad categories—composite, fluid, and crystallized tests—we tested whether the category moderated effect sizes. For the control-prior-intelligence design, there were stronger educational impacts on composite tests (1.876 IQ points, SE = 0.467, p = 5.94×10−05) than on fluid tests (0.836 points, SE = 0.097, p =.152), and the difference between composite and fluid was significant (1.039 points, SE = 0.496, p =.036).There was only one crystallized outcome test for the control-prior- intelligence design (1.893 points, SE = 0.348, p = 5.34×10−08), so we did not include it in the moderator comparison here. For the policy-change design, there were significant effects for both composite (2.314 IQ points, SE = 0.869, p =.008) and fluid (2.272 points, SE = 0.765, p =.003) but not crystallized (1.012 points, SE = 1.125, p =.368) tests; however, the effects on the three different categories were not significantly different from one another (all difference p values >.35). Finally, for the school-age-cutoff design, there were significant effects of a year of education on composite (6.534 points, SE = 2.433, p =.007), fluid (5.104 points, SE = 0.621, p = 2.05×10−16), and crystallized (5.428 points, SE = 0.170, p = 1.04×10−223) tests; there were, however, no significant differences between effect sizes across outcome types (difference p values >.5). We then split the outcome tests into “achievement” tests versus “other” tests. There was only one achievement test in the control-prior-intelligence design, so we did not run this analysis. For policy change, there was no significant difference in the educational effect on the 7 achievement tests (2.760 IQ points, SE = 0.968, p =.004) versus the 23 other tests (1.784 points, SE = 0.553, p =.001; difference SE = 1.011, p =.334). However, for school-age cutoff, which had the largest proportion of achievement tests (38 of the 86 tests were classed as achievement tests), achievement tests showed a substantially and significantly larger educational effect (6.231 points, SE = 0.339, p = 2.85×10−75) than other tests (3.839 points, SE = 0.412, p = 1.11×10−20; difference SE = 0.371, p = 1.19×10−10)….”

Male-only studies.

Outcome test category.

“…Figure 1 shows the effect sizes in the first two study designs as a function of the participants’ mean age at the outcome test. For the control- prior-intelligence studies, outcome age was a significant moderator: The effect size of education declined by −0.026 IQ points per year of age (SE = 0.012, p =.029). At the youngest age (18 years) the effect size of having had an additional year of education was 2.154 IQ points, whereas at the oldest age (83 years) the effect size was 0.485 IQ points. This effect was smaller but still significant if the largest effect size (> 3 IQ points for an additional year) was excluded (−0.011 IQ points per year of age, SE = 0.005, p =.018). There was no significant moderating effect of age at outcome for the policy-change studies (0.014 IQ points per year, SE = 0.022, p =.543): The effect at the youngest age (18 years; 1.690 IQ points) was not significantly different from the effect at the oldest age (71 years; 2.413 IQ points). There was comparatively little variation in the age at the outcome test for the school-age- cutoff design (SD = 1.6 years; see Table 1); there was no significant age moderation effect (−0.027 points per year, SE = 0.399, p =.947)….”

“…. For the policy-change design, we tested whether the age at which the educational policy change produced an increment in compulsory schooling moderated the intervention effect. This was not the case: The effect size increased by a nonsignificant 0.038 IQ points per year of age at the intervention (SE = 0.228, p =.867). The meta-regression model implied that at the youngest intervention age (7.5 years), the effect size was 1.765 IQ points, and at the oldest (19 years) it was 2.204 IQ points….”

Age at outcome test.

“…For the control-prior-intelligence design, we tested whether the age at which the participants had taken the initial (control) cognitive test, or the gap between this early test and the outcome test, moderated the effect size. The age at the early test, which did not vary substantially (see Table 1) was not significantly related to the effect size (−0.024 IQ points per year, SE = 0.064, p =.706). For the youngest early-test age (10 years), the meta-regression model indicated that the effect size of 1 additional year of education was 1.243 IQ points; for the oldest (16 years), the effect size was 1.099 IQ points. Conversely, the time lag between the tests was a significant moderator of the effect size (−0.031 IQ points per year, SE = 0.015, p =.033). This meta-regression indicated that at the smallest age gap (5 years), the effect size was 2.398 IQ points, whereas for the largest age gap (72.44 years), the effect size was a substantially smaller 0.317 IQ points. Note that this age gap is almost fully confounded with the age at the out- come test (r =.988), assessed as a moderator below….”

Age at intervention.

Age at early test and time lag between tests.

“…We analyzed results from the three most prominent quasi-experimental methods for testing the effects of education on intelligence. We defined intelligence as the score on a cognitive test; see below for consideration of how the test scores might relate to the underlying psychological processes. Each method implements a different approach to minimize effects stemming from selection processes. Full meta-analytic inclusion criteria are reported below, but first we describe each of the three designs, providing a canonical example of each..…. The first research design, which we label control prior intelligence, is a longitudinal study in which cognitive testing data are collected before and after variation in the duration of education. This allows the relation between education and the second test to be adjusted for by each participant’s earlier ability level. An example of this design is the study by Clouston et al. (2012), who analyzed data from three large U.S. and UK cohort studies, all of which had both an adolescent and a midlife cognitive test. Results indicated that completing a university education was linked to higher midlife cognitive ability, above and beyond adolescent intelligence….…The second design, policy change, relies on changes in educational duration that are, by all accounts, exogenous to the characteristics of the individuals. An example of this design is the study by Brinch and Galloway (2012), who used large-scale data from a 1960s educational reform in Norway. This reform increased compulsory education by 2 years; critically, it was staggered across municipalities in the country. This allowed the researchers to estimate the effect of an additional year of school on a later intelligence test, taken by males at entry to military service as part of Norway’s universal military draft. Under the assumption that the policy change affected intelligence only via increasing years of schooling, the authors used an instrumental-variables analysis to estimate the effect of 1 year of schooling on intelligence at approximately 3.7 points on a standard IQ scale (M = 100, SD = 15 in the population)….…The third design takes advantage of a school-age cutoff. These studies use regression-discontinuity analysis to leverage the fact that school districts implement a date-of-birth cutoff for school entry. The first study to use this method was by Baltes and Reinert (1969), but the most highly cited example is by Cahan and Cohen (1989), who, in a sample of over 12,000 children across three grades of the Israeli school system (between the ages of approximately 10-12 years), found that schooling exerted positive effects on all of 12 tests covering a variety of cognitive domains. These educational effects were around twice the effect of a year of age. The strict assumptions of this method are some- times not fully met (Cliffordson, 2010); methodological issues are discussed in more detail below….”

Their Evidence

“…In three separate unconditional random-effects meta- analytic models (one for each study design), we estimated the effect of 1 additional year of education on cognitive outcomes. For all three study designs, there was a significant effect of 1 additional year of education. For control prior intelligence, the effect was 1.197 IQ points (SE = 0.203, p = 3.84×10−09); for policy change, it was 2.056 IQ points (SE = 0.583, p = 4.23×10−04); and for school-age cutoff, it was 5.229 IQ points (SE = 0.530, p = 6.33×10−23). An overall model including all estimates from all three designs found an average effect size of 3.394 IQ points for 1 year of education (SE = 0.503, p = 1.55×10−11)….”

Types of Studies They Looked At

We are all familiar with comparisons showing that Black people earn 50% less than white peers and women earn 70 cents for every dollar that a man earns. However, the most popular tools used to combat disparities in the workplace have produced almost no measurable results. The average impact of corporate diversity, equity, and inclusion (DEI) training is zero and some evidence suggests that the impact can become negative if the training is mandated.One of the most important developments in the study of racial inequality has been the quantification of the importance of pre-market skills in explaining differences in labor market outcomes between Black and white workers. In 2010, using nationally representative data on thousands of individuals in their 40s, I estimated that Black men earn 39.4% less than white men and Black women earn 13.1% less than white women. Yet, accounting for one variable-educational achievement in their teenage years--reduced that difference to 10.9% (a 72% reduction) for men and revealed that Black women earn 12.7 percent more than white women, on average. Derek Neal, an economist at the University of Chicago, and William Johnson were among the first to make this point in 1996: “While our results do provide some evidence for current labor market discrimination, skills gaps play such a large role that we believe future research should focus on the obstacles Black children face in acquiring productive skill.”

Roland Fryer, "It’s time for data-first diversity, equity, and inclusion,"Fortune, June 20, 2022, https://fortune.com/2022/06/20/data-first-diversity-equity-inclusion-careers-black-workers-gender-race-bias-dei-roland-fryer/

It’s time for data-first diversity, equity, and inclusion

I have spent hundreds of hours in diversity training over the past two decades-from descriptions of federal anti-discrimination laws to academic-style seminars on the perils of implicit bias, microaggressions, or misgendering.

Advocates of this kind of training have their hearts in the right place. We are all familiar with comparisons showing that Black people earn 50% less than white peers and women earn 70 cents for every dollar that a man earns.

However, the most popular tools used to combat disparities in the workplace have produced almost no measurable results.

The average impact of corporate diversity, equity, and inclusion (DEI) training is zero and some evidence suggests that the impact can become negative if the training is mandated.

“Statistical Snapshots,” which describe how employee outcomes differ by demographic group, are another popular tool. These numbers cannot provide proof of bias. Simple averages often mislead and, importantly, crafting strategies based on misleading data often does more harm than good.

Some business leaders, in their determination to increase diversity, leap directly from observing raw disparities to removing some information from application forms, another common practice meant to make workplaces more equitable. However, hiding information on applications often leads to worse outcomes for those it was intended to help-likely because hiring managers use race itself as a proxy for the information they’re no longer allowed to see.

Our intuition for how to decrease race and gender disparities in the workplace has failed us for decades. It’s time to stop guessing and start using the scientific method. Remember when we thought that the Bubonic Plague was caused by a triple conjunction of Saturn, Jupiter, and Mars in the 40th degree of Aquarius?

Here is a three-step approach that can turn earnest intentions into good science.

Understand disparities

For decades, social scientists have shown that raw gaps in employment outcomes like hiring or wages-the type of data typically provided to C-suite executives-misstate the amount of actual bias in an organization. This data omits many factors that are key to personnel decisions, factors that often vary by group, owing to disparities in society at large. Business leaders can and should work to address inequality in their communities, but they should not mistake society-wide gaps for bias by their own employees.

One of the most important developments in the study of racial inequality has been the quantification of the importance of pre-market skills in explaining differences in labor market outcomes between Black and white workers. In 2010, using nationally representative data on thousands of individuals in their 40s, I estimated that Black men earn 39.4% less than white men and Black women earn 13.1% less than white women. Yet, accounting for one variable-educational achievement in their teenage years--reduced that difference to 10.9% (a 72% reduction) for men and revealed that Black women earn 12.7 percent more than white women, on average. Derek Neal, an economist at the University of Chicago, and William Johnson were among the first to make this point in 1996: “While our results do provide some evidence for current labor market discrimination, skills gaps play such a large role that we believe future research should focus on the obstacles Black children face in acquiring productive skill.”

Recently, I worked with a network of hospitals determined to rid their organization of gender bias. The basic facts were startling: Women earned 33% less than men when they were hired and their wages increased less than men once on the job. Yet, accounting for basic demographic variables known about individuals prior to hiring, these differences decreased by 74%. A problem remained, but it was an order of magnitude smaller than the unadjusted numbers implied.

Find the root causes of bias

Social scientists tend to categorize bias into one of three flavors: preference, information, and structural. Preference bias is good old-fashioned bigotry. If company A prefers group W over group B then they will hire and promote them more even if they are similarly qualified.

Information bias arises when employers have imperfect information about workers’ potential productivity and use observable proxies, like gender or race, to make inferences (gender stereotypes are a classic example).

Structural bias occurs when companies institute practices, formally or informally, that have a disparate impact on particular groups, even when the underlying practices are themselves group blind. Employee referral programs can fall into this category.

Over the past fifty years, economists and other social scientists have developed brilliant ways of statistically distinguishing between different types of bias. Gary Becker, in his 1993 Nobel Prize acceptance speech, outlined one such statistical procedure known as the “outcomes test.” It operates by comparing the success rate of decisions across groups and then inferring whether different decision rules were used for different groups. For example, if women CEOs statistically outperform male CEOs, all else equal, that would suggest that a higher standard was applied to women in the selection process. This type of statistical test can be used for hiring, promotions, and attrition across an organization.

In 2013, collaborators and I developed a similar test to detect information-based bias. Our approach uses the insight that if employers have information-based biases at the time of hiring, but then learn more about an employee’s productivity once they are on the job, one would expect to see the returns to tenure within the company to be higher for the group that faced the initial bias. Using a nationally representative dataset of thousands of individuals, we found that there was a significant gap at the time of hiring for Black candidates relative to white peers but that, as predicted, Black candidates experienced a 1.1 percentage point higher return to tenure.

With the aforementioned hospital network, the data pointed to a structural bias in scheduling. Women and men who worked the same number of hours earned exactly the same wage, but men worked more hours due to how the company assigned schedules, not women’s desire to work less.

This is the key step that is missing in every DEI initiative I have seen in the past 25 years: a rigorous, data-driven assessment of root causes that drives the search for effective solutions. In other aspects of life, we would not fathom prescribing a treatment without knowing the underlying cause. Hiding information on resumes when information bias is present is as effective as using alcohol baths to treat fever.

Evaluate

We must rigorously evaluate what works and what doesn't. The old cardiac test-where you “feel it in your heart”-is not good enough. Once we know where potential biases exist, determine what caused them, and curate a set of solutions to test, we must meticulously evaluate what’s working and what’s not.

Solutions that yield measurable results can be substantiated into company policy, while those that don't should be discarded. In the case of the hospital network, once a small change was made to the structure of their scheduling, gender differences were reduced. Despite countless hours spent in training and seminars, their results were unchanged for years. The solution was hidden in plain data.

This will seem heretical to some-but it barely scratches the surface of what's possible with a data-first approach to diversity, equity, and inclusion. More corporate leaders should be trying to solve diversity challenges in the same way they solve problems in every other aspect of their business: through intelligent use of data, rigorous hypothesis testing, and honest inference about what is working.

The surviving descendants of the old elite have prospered. By 2010 they were again richer and more educated than the Chinese average, according to data gathered by an international group of academics. Adding new evidence from cities to previous work, which looked at rural areas alone, the authors now find that the elite’s grandchildren have even out-earned Communist Party members.To measure the initial impact of China’s reforms on inequality, the authors compiled data on land ownership in 1950 from archived records. Unsurprisingly, inequality of land holdings, measured by a Gini coefficient, fell sharply after land reform, from 0.5 to 0.1—close to perfect equality. To measure what has happened since then, the authors used a survey of 36,000 Chinese residents from 2010. It recorded earnings and education, as well as social class. The party created hereditary class labels in 1950, in part to punish the old guard. These allowed the researchers to distinguish between old elites and everyone else. The authors found that elites born before 1940 were 7% likelier than their contemporaries to have finished secondary school. Their stigmatised children were 3% less likely to have done so than others their age. By 2010 the children of old elites earned 5% less than other Chinese. But things flipped back. Descendants of the old elite born between 1966 and 1990 were 6% more likely to finish high school than their contemporaries. In 2010 they earned 12% more than other Chinese. They even earned 2% more than party members. The researchers found that the old elite’s grandchildren are more enterprising and work longer hours than the descendants of those who had lower social standing. Although the elite’s capital was destroyed 70 years ago, their social capital has endured.

The authors found that elites born before 1940 were 7% likelier than their contemporaries to have finished secondary school. Their stigmatised children were 3% less likely to have done so than others their age. By 2010 the children of old elites earned 5% less than other Chinese.

But things flipped back. Descendants of the old elite born between 1966 and 1990 were 6% more likely to finish high school than their contemporaries. In 2010 they earned 12% more than other Chinese. They even earned 2% more than party members.

The researchers found that the old elite’s grandchildren are more enterprising and work longer hours than the descendants of those who had lower social standing. Although the elite’s capital was destroyed 70 years ago, their social capital has endured.

Economist Staff, “The grandchildren of China’s pre-revolutionary elite are unusually rich,”The Economist, June 9, 2022, https://www.economist.com/graphic-detail/2022/06/09/the-grandchildren-of-chinas-pre-revolutionary-elite-are-unusually-rich

The grandchildren of China’s pre-revolutionary elite are unusually rich

The land ownership system of feudal exploitation by the landlord class shall be abolished.” So read China’s agrarian reform law of 1950. Land was seized from the better-off and given to poor farmers, whose share of farmland rose from 14% in 1947 to 47% in 1954. Liu Shaoqi, who was president during the Mao era, called it the “most thorough reform in thousands of years of Chinese history”.

Hundreds of thousands of landlords were murdered. Tens of millions of people died in a famine when farms were collectivised. Yet the surviving descendants of the old elite have prospered. By 2010 they were again richer and more educated than the Chinese average, according to data gathered by an international group of academics. Adding new evidence from cities to previous work, which looked at rural areas alone, the authors now find that the elite’s grandchildren have even out-earned Communist Party members.

To measure the initial impact of China’s reforms on inequality, the authors compiled data on land ownership in 1950 from archived records. Unsurprisingly, inequality of land holdings, measured by a Gini coefficient, fell sharply after land reform, from 0.5 to 0.1—close to perfect equality.

To measure what has happened since then, the authors used a survey of 36,000 Chinese residents from 2010. It recorded earnings and education, as well as social class. The party created hereditary class labels in 1950, in part to punish the old guard. These allowed the researchers to distinguish between old elites and everyone else.

Women...now make up 44% of college-educated workers in science, technology, engineering, and mathematics (STEM) occupations. However, a closer look at the underlying data reveals divergent patterns of gender (in)equality within STEM. Men continue to make up the vast majority of degree recipients in STEM fields that are spatially and mathematically intensive — geoscience, engineering, economics, mathematics, computer science, and physical science — while women are overrepresented in life sciences, psychology, and social sciences....In 2017, 49% of all bachelor’s degrees in science and engineering were awarded to women. This number masks substantial gender disparities across fields, however. Women earned two-thirds of bachelor’s degrees in life sciences, psychology and social science fields — 62% in biology, 78% in psychology, and 61% in social science. In contrast,women were only slightly over one quarter of degree recipients in the combined math-intensive fields of geoscience (39%), engineering (22%), economics (32%), mathematics (42%), computer science (19%), and the physical sciences (40%)....Gender disparities within science, technology, engineering, and mathematics bachelor’s degrees have become more pronounced since 2000...Gender differences carry over from education to the STEM workforce.

A leaky pipeline reduces the share of women rising up the ranks of career hierarchy in academia. Women were 45% of junior full-time faculty but only 33% of senior full-time faculty in 2019. However, the pipeline leaks more in fields where women are overrepresented in the lower ranks. In psychology, for instance, the share of women was 68% among junior full-time faculty and 53% among senior faculty, a 15 percentage-point difference. With a 6 percentage-point difference, pipeline leakage is smallest in computer science: the share of women was 25% among junior full-time computer science faculty and 19% among senior faculty. What explains the differences in pipeline leakage by field? It is possible that the smaller number of women who choose to enter the male-dominated fields may be highly positively selected — especially motivated — compared to those who enter the more female-friendly fields. This is consistent with research showing that a large number of female entrants into STEM fields has a negative impact on gender equality in that field in terms of career outcomes. In psychology and the social sciences, it may be that the higher degree of female representation in the lower ranks indicates a shift in the preferences of men: the number of junior male faculty has declined since 1981. This may be due to occupational “tipping”, whereby men start to exit an occupation when the fraction of women in that occupation rises above a certain threshold.

What this Means:

As a smaller proportion of women enter geoscience, engineering, economics, mathematics, computer science, and physical science at the undergraduate level, the career pipeline in these fields will become even smaller. Less gender diversity could be detrimental to the quality and relevance of knowledge generated in these fields, by narrowing the range of ideas generated and reducing success in problem solving. We should then expect the same outcome in life sciences, psychology and social sciences if women continue to saturate them while men exit. A widening of the uneven gender distribution between STEM fields helps perpetuate the gender pay gap. This, in turn, can have negative consequences for gender equality outside the labor market if, for example, lower female-male wage ratios erode women’s decision-making power in the family and reduce women’s well-being.

The share of women holding academic positions has increased substantially but also varies greatly by field. In 2019, women comprised 39% of all academic positions held by science, engineering and health doctorate holders at 2- and 4-year colleges or universities. This reflects a substantial increase from just 9% in 1973 and 22% in 1993. Gender gaps vary noticeably by fields. Women held only 19% of academic positions in engineering, 26% in physical and geosciences, 26% in mathematics and statistics, and 23% in computer science. This is in contrast to 60% of positions in psychology, 47% in life sciences, and 42% in social sciences including economics. Among academic position types, women are more likely than men to hold “other” full-time positions (instructor, adjunct, lecturer, research associate, administrator) as well as post-doc and part-time positions.

The STEM fields in which women are underrepresented typically pay higher salaries than fields with high shares of women. For example, the median salary in 2019 for full-time workers holding the highest degree in their field was $105,000 for computer scientists and $100,000 for engineers, both male-dominated occupations. Among female-heavy occupations, median salary was $74,000 for psychologists and $64,000 for biological, agricultural and environmental scientists.

Gender differences carry over from education to the STEM workforce. While the college-educated STEM workforce had almost reached gender parity in 2019, with a female share of 44%, large gender gaps remain between types of STEM jobs. Although only 29% of the science and engineering (S&E) workforce in 2019 was female, women were 57% of S&E-related workers (e.g. nurse, dental hygienist). The representation of women coincides with the patterns of gender balance in STEM degrees. Women scientists and engineers comprise 48% of biological, agricultural and environmental life scientists, 81% of psychologists, and 64% of social scientists; and much smaller shares of geoscientists (30%), engineers (16%), economists (24%), mathematical scientists (42%), computer scientists (24%), and physical scientists (35%). Even within the S&E-related workforce, women comprise 70% of health-related occupations — such as nurses, pharmacists, and physician assistants — but only 21% of computer and information systems managers.

Female attrition along the STEM educational hierarchy is driven by the very disciplines that, paradoxically, seem to attract large numbers of women at the entry level. In general, women earned a smaller share of doctoral degrees than bachelor’s degrees in 2017, reflecting the well-known leaky pipeline in women’s progress up the career hierarchy. In STEM fields, the pipeline is leakiest in life science, psychology and social science fields, which are female-dominated at the undergraduate level — the female share of degree recipients in these fields was 58% at the doctoral level compared with 66% at the bachelor’s level in 2017. In contrast, the four fields with the lowest female shares among bachelor’s degrees recipients — geoscience, engineering, economics, and computer science — have higher female representation among PhD recipients (see here).

Gender disparities within science, technology, engineering, and mathematics bachelor’s degrees have become more pronounced since 2000. The number of women earning bachelor’s degrees in life sciences, psychology and social sciences grew faster than the number of men since 2000. These fields, which started out with high female shares at the turn of the century, became even more disproportionately represented by women at the undergraduate level. Similarly, even as the numbers of women in many of the STEM fields that started out with a disproportionate representation of men have been increasing, the number of men earning bachelor’s degrees in several of these fields grew faster than the number of women — this trend was most pronounced in computer science and mathematics, while engineering was an exception (see chart).

While women earn close to half of all STEM bachelor’s degrees, there is wide variation in women’s representation across fields. In 2017, 49% of all bachelor’s degrees in science and engineering were awarded to women. This number masks substantial gender disparities across fields, however. Women earned two-thirds of bachelor’s degrees in life sciences, psychology and social science fields — 62% in biology, 78% in psychology, and 61% in social science. In contrast, women were only slightly over one quarter of degree recipients in the combined math-intensive fields of geoscience (39%), engineering (22%), economics (32%), mathematics (42%), computer science (19%), and the physical sciences (40%).

Prathi Seneviratne, "Are Women Reaching Parity with Men in STEM?"Econofact, April 15, 2022, https://econofact.org/are-women-reaching-parity-with-men-in-stem

Are Women Reaching Parity with Men in STEM?

Women have made tremendous gains in education and now make up 44% of college-educated workers in science, technology, engineering, and mathematics (STEM) occupations. However, a closer look at the underlying data reveals divergent patterns of gender (in)equality within STEM. Men continue to make up the vast majority of degree recipients in STEM fields that are spatially and mathematically intensive — geoscience, engineering, economics, mathematics, computer science, and physical science — while women are overrepresented in life sciences, psychology, and social sciences. If anything, these gender disparities appear to be widening slightly over time at the undergraduate level, with potential implications for gender differences in employment, pay, academia, and beyond.

The Facts:

Women have made dramatic inroads in education since the 1990s. From making up less than a third of the college-educated workforce in the early 1990s, women have come to surpass men — making up 52% of the workforce with at least a college degree in 2019. The representation of college-educated women in STEM occupations has also been trending upwards since the early 1990s.

Jack Mountjoy and Brent Hickman, "The Returns to College(s): Estimating Value Added and Match Effects in Higher Education," University Of Chicago, January 2020, https://bfi.uchicago.edu/working-paper/the-returns-to-colleges-estimating-value-added-and-match-effects-in-higher-education/

Overall the paper does look at admitted “elite” students choosing less selective schools however I don’t see a carveout for those students and STEM selection, "...While Figure 2 shows that admission portfolio controls nearly eliminate imbalances in observable student potential acrosscollege treatments, a lingering concern with the matched applicant approach is that it relies on “the small share of students who make what is a very odd choice” (Hoxby, 2009): enrolling in a less selective college when admitted to a more selective alternative. Table 2 directly investigates the nature of this identifying variation in our sample, and finds that such choices are actually far more common than the concern supposes. The first section of Table 2 shows that among students in our identifying sample, i.e. those admitted to at least two colleges, only 55% end up enrolling in their most selective option, with the average admission portfolio spanning a selectivity range of 116 SAT points (SD 78). The second section of Table 2 shows that this phenomenon is not confined to students choosing solely among less selective institutions. Of the 27% of students in our identifying sample who are admitted to UT-Austin, the most selective campus in our students’ choice sets, only 58% choose to enroll there. And the remainder are not simply sidestepping en masse over to the other flagship: conditional on being admitted to UT-Austin and not being admitted to Texas A&M, still only 60% of students choose UT-Austin, meaning 40% forgo it in favor of a less selective, non-flagship school. Very similar patterns also arise among Top Ten Percent students in our sample, who are guaranteed admission to every Texas public university they apply to. Only 30% of them end up enrolling at UT-Austin, and 25% at Texas A&M, leaving fully 45% who decline to exercise their flagship entitlement and instead fan out across the 28 non-flagship campuses in our sample...."

Vox write up of Catalist data (note report attached)“…Biden’s share of votes by Latinos decreased by 8 percentage points compared to Hillary Clinton’s, and his share of votes by Black people decreased by 3 percentage points. Biden’s share of white voters without a college degree improved about 1 percentage point as compared to Clinton. Biden’s share of white voters with a college degree improved by about 4 percentage points from Clinton’s….”

Andrew Prokop, "A new report complicates simplistic narratives about race and the 2020 election,"Vox, May 10, 2021, https://www.vox.com/platform/amp/2021/5/10/22425178/catalist-report-2020-election-biden-trump-demographics

A new report complicates simplistic narratives about race and the 2020 election

In the days and weeks after presidential election results come in, commentators attempting to figure out what happened with voter demographics are often in a fog — forced to rely on unreliable exit polls. More rigorous analysis simply takes longer.

Now, Catalist, a Democratic data firm, has put out a report on “What Happened in 2020,” authored by Yair Ghitza and Jonathan Robinson, which makes a serious attempt to answer that question. The report is superior to the exit polls because it’s based in their research for what’s known as a “voter file.” Basically, they’ve put together a large database of turnout information about actual voters, assembled from state or local records about who actually showed up.

That information has its limits — it’s a secret ballot, so we don’t know who specific people voted for. But this information can be supplemented with precinct-level vote results, census information, and survey findings. The Catalist authors put all this together to estimate how different demographics voted. Some of their general findings about what changed since 2016:

Biden’s share of votes by Latinos decreased by 8 percentage points compared to Hillary Clinton’s, and his share of votes by Black people decreased by 3 percentage points.
Biden’s share of white voters without a college degree improved about 1 percentage point as compared to Clinton.
Biden’s share of white voters with a college degree improved by about 4 percentage points from Clinton’s.

But the report doesn’t just look at the narrow question of what changed in each group; it helps us understand the two parties’ coalitions as a whole. For instance, it drives home the fact that the non-college white population, while declining as a share of the electorate, remains quite large (they were about 44 percent of all voters in 2020), and Trump won 63 percent of them. And though Biden performed worse among Black voters than Clinton or Obama, he still won about 90 percent of that group.

One note here is that, for the sake of allowing easy year-to-year comparisons, these percentages are calculated by looking only at votes cast for either the Democratic or Republican nominees, rather than for third-party candidates. This is called the two-way vote share — Catalist’s report focused on this metric, so all of this article’s references to demographic vote percentage will use it as well.

The non-college white vote: Joe Biden’s underrecognized electoral achievement

Much attention has been paid to the shift of white college-educated voters toward Biden and Latino voters toward Trump in 2020, which makes sense — as a table from Catalist shows, those were particularly dramatic shifts.

This was crucial to Biden’s success for two main reasons. First, white non-college voters continue to make up a very large chunk of the electorate. Catalist estimates that their share of the electorate has been shrinking, but they still were 44 percent of all voters in 2020. And second, they make up an even larger share of the electorate in some key swing states.

Furthermore, the shorthand of punditry often says Democrats win because of a coalition of nonwhite and college-educated white voters. But the below chart makes the point that non-college whites actually still make up an estimated 32 percent of overall votes cast for Biden. In comparison, Black and Latino voters together also comprised 32 percent of the overall votes for Biden. More than half the Trump coalition, in contrast, is comprised of non-college-educated white voters (and Trump’s coalition is 85 percent white overall).

Latino voters: Biden’s better-recognized problem

Biden’s worsened performance among Latino voters as compared to past Democratic nominees has been much discussed since the election, and Catalist puts some numbers on just what happened there — the group estimates that Hillary Clinton won 71 percent of Latino voters in 2016, but Biden won 63 percent in 2020. This 8-point drop is the most dramatic shift in a four-year period among the major racial or ethnic groups seen in Catalist’s data.

This comes in the context of a massive increase in Latino turnout overall — while turnout was up across the board, it was particularly up among Latino (and Asian) voters, in what’s just an incredible increase in a four-year period.

The turnout surge also raises a question — did Trump improve because previously voting Latinos switched from backing Democrats to backing Republicans, or were his gains mainly among new voters? Catalist doesn’t provide a definitive answer, but the authors say, “we see suggestive evidence that both of these factors played some role.”

In contrast to claims that Democrats’ worsening performance with Latinos was mainly concentrated among Cuban Americans in Florida, the report finds that the decline was broader — but that the drops were smaller in key states that Biden barely won, like Arizona and Georgia.

Black voters’ support is crucial for Democrats — is it dropping?

Biden won an estimated 90 percent of the two-way vote among Black voters in 2020 — which is both an overwhelming win, and the worst performance for a Democrat among Black voters in over a decade. Clinton won 93 percent of the two-way vote among Black voters in 2016, and Obama won 97 percent in 2012.

The authors make that point that despite this drop, “Black voters are key to Democratic success” — without the large majorities of Black voters Biden won, he couldn’t have been competitive nationally or in key Electoral College states like Georgia, Pennsylvania, or Michigan.

Furthermore, the report estimates that turnout among Black voters increased from 2016 by 14 percent, a higher rate of increase than among white college voters (13 percent) or white non-college voters (11 percent). So it can be true both that Biden did worse than Clinton among Black voters and that Black voters remain a core pillar of the Democratic coalition.

White college-educated voters have moved toward Democrats (but many still vote for Trump)

So how does Joe Biden win if he barely improves on Clinton’s performance among non-college-educated white voters and does worse than she did among Black, Latino, and Asian voters?

It’s because of those gains among white voters with a college degree — where he built on and expanded the improvement Clinton had made. In 2012, Obama won an estimated 46 percent of these voters, in 2016 Clinton won 50 percent of them, and in 2020, Biden won 54 percent of them.

Further, according to the report, about 58 percent of white women with a college degree voted for Biden (compared to 55 percent voting for Clinton in 2016 and 50 percent voting for Obama in 2012).

This is true as far as the trend goes, and it’s one of the most important factors explaining why Biden won and Clinton didn’t. However, it’s also important to stress that 46 percent of college-educated whites still did vote for Trump. Democrats don’t exactly dominate among this group just yet, and it remains to be seen whether their gains here will prove durable.

One overall takeaway from the report, though, is that an electoral majority is often a fragile thing these days. Lose a few points among a few groups and it could slip from your grasp. And some apparent trends continue from one election to the next — while others can surprisingly reverse