Experimental Design
The study uses a two-stage randomized evaluation design (a randomized controlled trial with a waitlist rollout), embedded in the EPAG training program in urban Liberia.
1) Recruitment and sample
• Young women were recruited in nine target communities around Monrovia (and Kakata), applied in person, completed a registration form (including their preferred track), and took a basic literacy/numeracy screen.
• Eligibility targeted women age 16–27, with basic literacy/numeracy, not recently enrolled in school, and residing in the target communities.
• About 2,106 were eligible; 1,991 were interviewed at baseline and entered the random assignment procedure.
2) Two interventions (tracks)
• Participants could receive either Job Skills (JS) training or Business Skills (BS) training. The intervention package included ~6 months of classroom training + ~6 months of follow-up services.
3) Stage 1 assignment: preference + capacity + random reassignment
• At application, women stated a preferred track (JS or BS).
• Because JS slots were limited, about 30% of JS-preferring applicants were reassigned to BS. This reassignment was done by randomly drawing application forms (manual random draw).
• This creates three “candidate groups”:
• G1: preferred JS and assigned JS
• G2: preferred BS and assigned BS (“BS chosen”)
• G3: preferred JS but reassigned to BS (“BS reassigned”)
4) Stage 2 assignment: randomized rollout (treatment vs waitlist control)
• Within each candidate group (G1–G3), women were randomly assigned to start in Round/Cohort 1 vs Round/Cohort 2:
• Cohort 1 = “treatment” (offered training immediately)
• Cohort 2 = “control” (waitlisted; started ~13 months later, after the first follow-up survey).
• The cohort randomization was done by computer, 2/3 to cohort 1 and 1/3 to cohort 2, stratified by experimental group.
• This yields six experimental groups (T1/C1, T2/C2, T3/C3) corresponding to (G1–G3) × (early vs late start), with sample sizes:
• T1=437, C1=312; T2=540, C2=304; T3=243, C3=155.
5) Measurement and estimation approach
• Surveys were collected at baseline (pre-disclosure of randomization results), first follow-up (after cohort 1 finished, before cohort 2 began), and second follow-up (immediately after cohort 2 finished).
• We report intention-to-treat (ITT) effects based on initial assignment, not actual participation.