Degrees Alone Won’t Be Enough for Women in STEM
Degrees Alone Won’t Be Enough for Women in STEM
Of the many problems and controversies plaguing Uber lately, none have such an intense response from the company as the recent account that sexist management practices that former Uber engineer Sarah J. Fowler encountered during her yearlong tenure. Her harrowing account of the indifferent to outright hostile response to issues raised about treatment of female engineers at the company shows, in many ways, the worst-case scenario of a workplace culture that both causes and is caused by an atmosphere where few women work or feel welcome.
Uber’s case is dramatic, and perhaps extreme, but it is endemic among a much broader trend: the underrepresentation of women in Science, Technology, Engineering, and Mathematics (STEM) professions. These jobs represent a relatively small share of a labor market that has overcome significant amounts of gender inequality in recent decades, but the ascendancy of Google, Facebook, and other tech companies has radically expanded demand (and compensation) for STEM workers and researchers. And while women still outnumber men in many “STEM-related” fields, especially nursing and some other medical fields, there is a worryingly clear predominance of men in engineering, computer science, and high-paying medical specialty fields.
Some of this gap manifests in educational attainment; women comprise about half of STEM graduates, but 57% percent of all college graduates overall. Between different STEM concentrations, the ratio varies widely: chemistry graduation rates are near gender parity, but fewer than one in four computer science graduates are women. Nor are these ratios set in stone: chemistry reached its current balance over several decades, in step with other non-STEM studies like law and business, while the computer science ratio for women actually dropped significantly when men flooded into the field starting around 2000.
This notion, that the imbalance of qualified STEM graduates accounts for the disparity in STEM employment, is known as the “pipeline problem,” and it applies not just to women, but also to African Americans, Hispanics, and other ethnic groups. The idea that underrepresented groups would be hired equally, if they really were enough qualified candidates out there, is a favorite of high profile employers, since it conveniently absolves them of blame.
To figure out how much this difference matters, I ran a series of models on data from the American Community Survey’s Public Use Microdata Sample, including linear probability models, logistic regressions, and propensity score matching analysis. Though gender is already essentially randomly distributed throughout the population, I also controlled for demographic factors such as race, age, marital status; labor market factors like English language fluency or possession of an advanced degree; and state of residence. Though the results differed greatly on some factors, three clear trends surfaced:
1) For those without STEM degrees, women were less likely to work STEM jobs than men.
2) Possessing a STEM degree was the single greatest predictor of all tested variables for whether or not an individual works in a STEM job.
3) STEM degrees increased the likelihood of working a STEM job by a greater predictive factor for men than for women. This is best demonstrated in the matching results below, where males and females were each matched with their most similar counterparts, one with a STEM degree and one without. The Average Treatment Effect on the Treated (ATT) is the resulting difference in probability of working a STEM job, which can be directly attributed to the STEM degree. The treatment effect is only 63% as strong for women as for men.
So what causes this difference? Some of it is almost certainly a spillover of factors affecting women in the labor market generally, including childcare expectations and maternity leave. Some is old-fashioned sexism creating hostile work environments for women. What cannot go unnoticed is the effect of widespread implicit bias: the behaviors and expectations that insidiously shape assumptions about women’s competency in STEM. These implicit biases are not just a matter of men believing themselves superior to women—at least one study has shown that, given nothing but the physical appearance of the candidate, male and female hirers are both more likely to hire men to do mathematical tasks. What’s more, women can internalize stereotypes about their lower aptitude for math and science, causing them to underperform their own standards in a phenomenon known as “stereotype threat.”
Policymakers looking to close this gap will have to accept there is not a gender problem in STEM, rather there are countless gender issues that in aggregate result in severe underrepresentation. Trying to pin down one problem—pipeline or otherwise—supports the false notion that there is a silver bullet to this quandary.
In reality, any steps that move closer to parity need to be continuously seized and built upon in a virtuous cycle: Enforcement of strong anti-discrimination and anti-harassment labor laws can mitigate the most egregious cases, like what Fowler encountered. Paid family leave or subsidized childcare services might help women stay in more competitive jobs, like those on the cutting edge of science and engineering. Governments and businesses alike can do more to showcase women in STEM as role models, with mainstream cultural touchstones like the film Hidden Figures spreading the message widely. STEM employers must continuously and honestly report their workforce diversity data, as Uber has only recently decided to do, to evaluate when interventions are helping and when they are not.
But of course, even with all this, parity in educational outcomes will be critical to stability and balance. The pipeline problem may not adequately explain every “leak,” but the benefits of education in the field are undeniable. Getting more girls interested in studying STEM, and more young women graduating with degrees, will continue to fill the candidate pool, while also developing a sense of passion and belonging that will extend into rewarding, intellectually challenging, and financially lucrative careers. From there, it may yet be possible to reach a world where Ms. Fowler’s experience sounds like a horror story from a quaint and antiquated past, rather than a painful exposé of the inequality we live with today.
 The matching analysis, performed using the R Matching package, works by first calculating the propensity of each individual to be in the treatment group (in this case, having a STEM degree) based on the other factors in the model. It then matches each member of the treatment group with their most similar counterpart in the control group (those without STEM degrees) by one of a number of possible weighting functions. In this case, I use the Mahalanobis distance to determine which observations are most similar. Then, when the treated population are all paired off with their most similar counterpoint in the control group, the two groups are compared in terms of their outcome (proportion with a STEM job). The difference is the ATT. For more, see the documentation for the R package.
Bryan Baird is a second-year MPP student at Georgetown's McCourt School for Public Policy, with a focus on technology policy and data. He received his BS from Washington University in St. Louis in 2012, where he studied systems science & engineering and political science. At Georgetown, he has been an active member of the Institute of Politics and Public Service, serving as a member of their inaugural class of student strategy teams.