'Nevertheless She Persisted?'

Having female peers -- even just a few of them -- can increase a woman’s odds of making it through her Ph.D. program in the natural sciences, technology, engineering or math, says a new working paper from the National Bureau of Economic Research.

Based on a sample of gradate students enrolled in STEM programs at public institutions in Ohio, women in Ph.D. cohorts with no female peers were about 12 percentage points less likely than their male peers to earn a doctorate within six years. One standard deviation in the share of female students in a cohort increased a woman’s chance of this "on time” graduation by about five percentage points. Much of that difference was attributable to a reduction in the dropout rate within the first year of graduate school. The findings were most pronounced in programs that are typically male-dominated.

This idea isn’t new. Advocates of women in STEM have long emphasized the importance of gender-inclusive environments, along with support networks of peers and mentors, in sealing the so-called leaky pipeline to the Ph.D. But the new study adds a quantitative dimension to the discussion.

It also adds nuance to the idea that male-dominated STEM fields, such as physics, are chillier to women than are fields such as biology, where there are many more women. Here, the idea is that cohort peer composition matters, not just composition of the field in general. So biology's overall climate for women may not matter to your success if you're the only woman you see. Conversely and especially, according to the study, physics' overall climate for women may not matter if you're surrounded by other women.

“A female-unfriendly climate is one cause of underrepresentation in STEM that resonates for many female scientists,” reads “Nevertheless She Persisted? Gender Peer Effects in Doctoral STEM Programs,” available in full from NBER here. “Unfortunately, the climate in these fields has been difficult to quantify empirically and researchers have consequently struggled to estimate the impact of environment on the gender gap in STEM.”

Using peer gender composition as a quantitative proxy for program climate, authors Valerie K. Bostwick, a postdoctoral researcher in economics at Ohio State University, and Bruce Weinberg, a professor of economics and public administration at Ohio State, analyzed a new data set linking administrative transcript records from all public universities in Ohio to data from UMETRICS. The Big Ten initiative provides information on the research environment -- source, timing and duration of funding -- for students who are supported by federal research grants. The key difference between this data set and others commonly used to study graduate students, such as the national Survey of Doctorate Recipients, is that it doesn’t just include completers. So the set allows for the longitudinal observation of program environments and what the study calls “drop-out behavior.”

Bostwick and Weinberg restricted their sample to STEM Ph.D. cohorts starting graduate school from 2005 to 2009, as their primary dependent variable was the probability of completing the program within the desirable target of six years. Extremely small programs were excluded, leaving them an estimation sample of 2,541 students in 33 programs at six Ohio universities.

The average cohort was approximately 17 students and 38 percent women. And while the overall finding was that women in cohorts with no female peers were 12 percentage points less likely than their male peers to graduate within six years of initial enrollment, the gap closed in highly female cohorts. For each additional 10 percent female students in a cohort, a woman’s chance of graduating on time increased by 1 percentage point. One standard deviation, or 21 percentage point, increase in the share of female students increased the probability of graduating within six years for women relative to men by 5 percentage points.

A key finding, the researchers say, is the data on the first year of programs. Women in cohorts with no female peers are 10 percentage points less likely to make it to the second year of a doctoral program than their male peers. That is the same as saying that women in cohorts with no female peers are 10 percentage points more likely to drop out in the first year, the paper says. It attributes the first-year influence of the cohort peer group to the fact that students are primarily engaged in coursework -- and each other -- at this time, rather than research.

Repeat: the first year matters. There was no evidence of any differences in financial support due to peer gender composition, but there was a small effect on grades.

Bostwick said this week that the approach allowed her and Weinberg to highlight not only that “some programs systematically have fewer women and lower female graduation rates,” but that even within those programs, “the share of women in your particular year or cohort matters.”

Asked about the real-world implications of her research, Bostwick said the most direct solution would be "to increase recruitment efforts to attract qualified women, limiting cohorts with very low shares of women." At the same time, she said, institutions and their faculties should be “striving to make the environments in their programs more inclusive and female-friendly.” That might mean formal training or “just having more supportive and understanding mentors.”

Weinberg said that some programs, including many in biology, "typically have large shares of women, and other programs, like many in engineering, typically have large shares of men." But in both cases, the programs tend to be small, "so if one or two more women than usual accept or turn down a program, it can have a meaningful impact on the gender mix of a cohort."

Brendan Price, an assistant professor of economics at the University of California, Davis, who saw Bostwick present the paper last year, said hers is a “careful study that uses rich data and a clever natural experiment."

Doctoral programs are a "small-numbers game, and for women pursuing Ph.D.s, sheer chance can easily make the difference between having several female peers or being the lone woman out,” he added. So possible lessons for STEM doctoral programs, especially those in traditionally male-dominated fields, include trying to enroll a “critical mass of female students” in each class. (The paper doesn't identify what a critical mass might be, only that the largest gender peer effects were observed in programs that were male-dominated, or over 62 percent male.) And when they can’t, Price said, these programs might need to “go to greater lengths to ensure that every student feels respected and supported."

Dick Startz, a professor of economics at the University of California, Santa Barbara, who (disclosure alert) supervised Bostwick’s dissertation when she was a student there, said he found her arguments “pretty convincing.” NBER working papers aren’t peer reviewed, but Startz said it wouldn’t have trouble becoming so.

As far as policy implications, Startz said the paper suggests that traditionally male-dominated fields should move beyond “tokenism in attracting more women students.” While it also suggests that departments should ask themselves why women need female peers, “introspection doesn’t work so well,” he said. So perhaps departments should also ask their female students these questions, then think and act accordingly.

Underscoring the finding about the first year in particular, Joshua Hawley, an associate professor of public affairs and education at Ohio State, said the paper suggests that institutions need to ensure that “high-quality advisers are available to all students from the first term.” Currently, however, many departments only assign doctoral advisers after students attain Ph.D. candidacy, following their comprehensive exams.

Changing that formula “may or may not be optimal” for departments, Hawley said, “but my belief is that careful initial assignment to a research adviser from Day No. 1” will serve male and female students alike, and possibly reduce attrition rates.