Two women standing outside doing scientific field research

Narrowing the STEM Gender Retention Gap: Early Learnings from Research

The working paper discussed in the below article is authored by Matthew Baird, Nikhil Gahlawat, Rosie Hood, Paul Ko, and Silvia Lara.

Women are less likely than men to pursue degrees in Science, Technology, Engineering, and Mathematics (STEM). They are also less likely to work in STEM. As shared in our recent white paper, there is a large expansion in the gender gap that occurs between graduation with a STEM degree and employment one year later in the United States. In a new working paper, we conducted a deep dive into the gender gap in retention in STEM occupations that occurs between graduation and one year later in the US. Our early findings indicate that men are 17.1 percentage points more likely than women to transition from STEM education to STEM employment one year later. Using regression analysis and decompositions, we showed that the entire gender gap can be explained by differences in a few observed factors. 

STEM majors as a contributing factor

One factor that contributes to the gap is differences in which STEM majors the workers are graduating from. Engineering and computer science graduates are more likely to work in STEM than social science, biology, and psychology graduates, even though all these majors are part of STEM. Women are disproportionately in majors with lower transition rates, while men are in majors with higher transition rates. We estimate that, if men and women within STEM majors graduated from each field at the same rate, this would nearly halve the gender transition gap from 17.1 percentage points to 10.2 percentage points. 

Labor market conditions have an impact on STEM retention

Another factor that impacts the gender gap in STEM retention is differences in the labor market conditions. Among these factors, early insights from our whitepaper found that the proportion of STEM workers who are the same gender as the graduate is a key driver of STEM retention. On average, women graduate in markets where around one-third of the STEM workforce is the same gender as them, while men have around two-thirds. We estimate that a ten-percentage point increase in STEM workers who are the same gender as a new STEM graduate would increase the probability of that graduate transitioning into a STEM occupation by around 2 percentage points. This also would create a positive feedback cycle, as more women in STEM work would increase the rate of new female STEM graduates entering STEM, which in turn increases the proportion of women in STEM. The overall STEM retention gap narrows from 17.1 to 8.8 percentage points when controlling for this (the proportion of STEM workers the same gender in your market) as well as the other labor and market conditions we control for. 

Job search activity and STEM retention

The third main factor explaining the STEM retention gap is differences in job search activity. Graduates who view and apply for a higher fraction of STEM jobs are much more likely to end up working in STEM. Women graduating from STEM who are viewing job postings on LinkedIn view on average 39.2% of STEM postings, compared to 58.6% for men. Although we cannot attribute reasons for this discrepancy in our current research, the gap impacts STEM retention. Among STEM graduates applying to job positions on LinkedIn, 42% of applications are for STEM positions among women, but 63.8% among men. If male and female STEM graduates viewed and applied for jobs similarly, this preliminary research indicates that the STEM gender retention gap would narrow from 17.1 percentage points to 5.5 percentage points. 

However, our approach is not causal: our results do not imply that forcing men and women to view and apply for STEM positions at similar rates would eliminate the gap. Differences in preferences for STEM positions, as revealed by job search activity, explain at least some of the gap. The question remains about how much of this gap in search activity is attributable to beliefs about fit in STEM and how much can be shifted over time by changing market conditions and increasing workplace policies and STEM work culture to better support women in STEM fields. 

This working paper shows how important many observed factors are to explaining the STEM retention gender gap. Without controls, men are 17.1 percentage points more likely than women to transition from a STEM degree to a STEM occupation. However, with the controls, we estimate that instead, women are 1.8 more likely than men to be retained in STEM. While this doesn’t imply that we can easily solve the STEM gender gap through these mechanisms, it does offer insights into policies that universities, workplaces, and governments can take to move the needle in the right direction. These include encouraging women to participate in STEM majors that have higher transition rates into STEM work, demonstrating to women that they do have a place in STEM work and should apply for STEM jobs, and building a positive feedback cycle wherein more women are in STEM, encouraging the new generations to also participate.  



Gender identity isn’t binary and we recognize that some LinkedIn members identify beyond the traditional gender constructs of “man” and “woman.” If not explicitly self-identified, we have inferred the gender of members included in this analysis either by the pronouns used on their LinkedIn profiles, or inferred on the basis of first name. Members whose gender could not be inferred as either man or woman were excluded from this analysis.

STEM jobs are defined as those with a LinkedIn Skills Genome containing at least 1 STEM skill within the top ten skills for that occupation. STEM skills are those skills that members who hold STEM degrees are at least five times more likely to list than non-STEM-degree holders. STEM Degrees are identified according to the U.S. Department of Homeland Security’s STEM Designated Degree Program list of majors (based on the U.S. Department of Education’s National Center for Education Sciences definition of STEM fields). A full description of the methodology can be found in this technical note.

Regression analyses employed multivariate regression using ordinary least squares. Decompositions used the Blinder-Oaxaca decomposition using the R package “oaxaca”.