LucasHazen12/rct-analysis-kenya-enterprise

Cluster-RCT analysis of a Kenyan enterprise program. Covers power analysis, ITT, IV/2SLS, matching estimators, randomization inference (500 simulations), Lee Bounds, and cost-benefit analysis. # rct-analysis-kenya-enterprise Cluster-RCT analysis of a Kenyan enterprise program. Covers power analysis, ITT, IV/2SLS, matching estimators, randomization inference (500 simulations), Lee Bounds, and cost-benefit analysis. # Cluster-Randomized Trial Analysis: Kenya Enterprise Program ## GPEC 464 — Designing Field Experiments | UC San Diego GPS ## Overview This project analyzes a two-round cluster-randomized trial of an entrepreneurship training program in rural Kenya. The analysis covers the full lifecycle of an RCT evaluation: power analysis at the design stage, balance testing, ITT and ToT estimation via IV and matching, randomization inference, heterogeneous treatment effects, Lee Bounds under differential attrition, and cost-benefit analysis. ## Research Question Does a village-level entrepreneurship training program increase monthly earnings for participants in rural Kenya? ## Data - **Design:** Cluster-randomized trial, treatment assigned at the village level - **Sample:** 804 individuals across 168 villages, 2 rounds (baseline + endline) - **Outcome:** Monthly earnings in Kenyan Shillings (Ksh) - **Treatment:** Binary — offered training (ITT) / received training (ToT) - **Covariates:** Baseline earnings, trusting, risk amount, age, urban status, highest grade completed - **Compliance:** One-sided non-compliance; uptake rate = 38.6% among offered ## Methods & Key Results ### Power Analysis (Pre-Treatment) - J = 168 villages, n ≈ 4.79 per cluster, ICC = 0.056 - MDE at 80% power: 1,028 Ksh/month without covariates; 982 Ksh/month with covariates - Covariate adjustment (R² = 0.088) provides only modest precision gain ### Balance Testing - Two-sample t-tests on six baseline covariates - No covariate significant at 5% level — randomization is balanced ### ITT Estimation - Simple OLS: 1,774 Ksh/month (p = 0.133) - ANCOVA (with baseline covariates): 1,744 Ksh/month (p = 0.121) - Insignificance driven by low uptake diluting the ITT across compliers and never-t …