Women account for only 35% of global STEM graduates, a share unchanged for a decade. We use administrative microdata from centralized university admissions in ten systems to deliver the first crossnational decomposition of the STEM gender gap into a pipeline gap (academic preparedness) and a choice gap (first-choice field conditional on eligibility). In deferred-acceptance platforms where eligibility is score-based, we isolate preferences from access. The pipeline gap varies widely, from -19 to +31 percentage points across education systems. By contrast, the choice gap is remarkably stable: high-scoring women are 25 percentage points less likely than men to rank STEM first.

This paper studies how social influences on academic performance evolve during transition to university when students encounter new social environments and publicly observable signals of academic ability. Using administrative data from a large, elite, and linguistically and culturally heterogeneous public university in Uganda and exploiting random assignment of students to dormitories, we examine how exposure to coethnic peers and peers with scholarship status affects academic performance. We find that greater exposure to peers from similar linguistic and cultural backgrounds improves academic achievement early in university, especially for students graduating from homogeneous high schools, but these effects fade over time. In contrast, exposure to scholarship peers generates persistent gains in academic performance that increase in magnitude throughout students' university experience. Our setting allows us to distinguish between peers' underlying academic ability and publicly observable signals of ability since scholarship status is visible, while underlying academic preparation is not. We find that exposure to scholarship peers improves academic performance regardless of those peers' underlying ability, while exposure to high-ability peers without a scholarship signal has no detectable effect. These results suggest that students respond more strongly to observable signals of academic standing than to peers' underlying academic preparation. Taken together, our findings indicate that the sources of academic influence evolve over time. Familiarity appears most important during the initial transition to university, while observable signals of academic standing become increasingly important as students progress through university.

This paper analyze a gender-based affirmative action policy implemented in 2020 for STEM programs at Uganda’s largest public university. Pushing more women in STEM majors by relaxing the admission criterion may lead to a potential mismatch. Improved gender balance, on the other hand, can generate positive peer effects for the entire cohort, especially women. Linking different administrative datasets on student’s applications, admissions and college academic performance, we empirically assess the effects of the policy induced changes in gender composition, on students’ higher education outcomes. We first establish that the policy successfully increased women’s access to STEM majors, resulting in a 9-percentage-point rise in female enrollment in traditionally male-dominated fields. We document that women (men) entering treated majors after the policy have, on average, lower (higher) high school test scores than their counterparts in untreated majors. Despite this, the policy led to improved college GPA and reduced the likelihood of failing courses in the first year for both men and women in the treated majors. Even among the top high school applicants whose admission status were not affected by the policy, we find that both men and women in treated majors experienced significant GPA improvements. Women’s GPA rose by 8%, compared to a 4% increase for men. Our results suggest that increased gender diversity, brought by the policy, had positive spillover effects beyond its direct impact on enrollment.

This study examines how employers select employees for training and employee demand for training. We incentive-compatibly elicit employers' beliefs about which of their employees it would be socially optimal to train, as well as employers' preferences over which of their employees they choose to train. We will then investigate if employers' selection of workers is individually rational or driven by behavioral biases. Finally, we will measure employees' self-selection into training and its alignment with employers' selections. To ensure incentive compatibility of employer and employee choices, we provide employees from a sample of metalworking SMEs in Kampala, Uganda, with free, high-quality skills training. In addition, we conduct practical skills tests to measure employee metal working skills before and after training. AEA RCT registration (Data analysis stage)

We study the effect of Uganda's policy that made chemistry and physics compulsory in addition to biology and mathematics, which were already mandatory for all students in junior high school. We examine the impact of this policy on performance in national exams, the number of girls entering STEM fields at advanced levels, and the persistence of both boys and girls in STEM at university. Using test scores in national exams, university applications, and school-level registrations, our approach exploits pre-policy variation in the school-level share of candidates sitting for national chemistry and physics exams.