Going All In: Simultaneously Breaking Down Barriers for Women in the STEM Workforce

The intervention is a 24-month online learning experiential program in India implemented by an Indian Ed-tech start-up. It aims to select, train, and nurture first-year disadvantaged women in STEM for top engineering jobs. Selected participants receive a 100% scholarship covering the entire program fee, along with a US$1200 one-time stipend for students to cover their expenses. The program typically starts in March every year, and the final selected students are announced in Feb every year. Applicants and program participants are engineering students who are in the first year of their degrees. This program has been in operation for 4 years. This study will evaluate two new cohorts of the program: 2024-2026 and 2025-2027. The study, spanning 2023-2028, aims to evaluate the WE program's efficacy in enhancing participants' technical and higher-order skills, ultimately influencing their labor market outcomes once these women graduate college.

2024-01-01

2026-08-31

(1) Data analysis skills, measured via a screen-recorded coding assessment with trained rater evaluations; (2) Teamwork and collaboration, measured via survey-based workplace scenarios; (3) Career aspirations and orientation, measured via survey responses about post-college plans and career path preferences; (4) Professional development and labor market readiness, measured via LinkedIn profile data.

Primary outcomes are organized into four pre-specified families. Family 1 (Data Analysis Skills): a summary composite index of data analysis competency constructed from trained rater evaluations of screen-recorded coding sessions, capturing ability to structure analytical tasks, work iteratively with data, exercise independent judgment, and communicate findings. Family 2 (Teamwork): measures of teamwork orientation and collaborative decision-making from survey responses to workplace scenarios, capturing whether the program shifts participants toward team-efficient versus self-reliant behaviors. Family 3 (Career Aspirations): measures of career plans and preferences from survey responses about post-college plans and career path rankings, capturing STEM orientation, role preferences, and willingness to pursue high-growth trajectories. Family 4 (LinkedIn): measures of professional development and labor market readiness from LinkedIn profile data, capturing investment in professional identity, signaling of technical competencies, and professional network development.

Whether participants have a LinkedIn profile; marriage expectations and timing; secondary career plans; process measures from the coding assessment.

Secondary outcomes include: whether participants report having a LinkedIn account, certainty and preferred timing of marriage, secondary career plans, and coding assessment process measures (including skill ratings, patterns of AI tool usage, and time allocation across tasks).

We evaluate the sixth cohort (2024–2026) of a national online training program for women engineers in India. The program is highly oversubscribed, with the number of students reaching the final selection stage being double the program's intake. We leverage this oversubscription to implement a randomized evaluation. After baseline surveys and skill assessments, 450 applicants from the final round were randomly assigned to treatment (N=225, selected into the program) and control (N=225, not selected). Endline data collection is scheduled for March–June 2026 and includes an online survey (~15 min), an individual coding assessment with screen recording (~45 min), and LinkedIn profile data.

Not available

Randomization done by researchers using statistical software

Individual Student

No

450 individuals (not clustered)

450 students

225 students treatment (selected into the program), 225 students control (not selected)

With 225 students per arm (N=450), 80% power, and a 5% significance level, the MDE without controls is 0.26 SD. Baseline measures include coding ability, technical score, aptitude score, English proficiency, IQ (Raven's Progressive Matrices), EQ (Reading the Mind in the Eyes), confidence, and Big Five personality traits. Accounting for an 85% response rate and baseline controls (estimated R² of 0.30–0.40 for skill-based outcomes using lagged measures), the MDE is approximately 0.22–0.23 SD for coding assessment outcomes and 0.25 SD for survey-based outcomes.