Randomized Saturation Design for Network Spillovers
Evaluating active labor market policies without accounting for neighborhood social network spillovers can misestimate treatment impacts due to information sharing or local displacement among untreated peers.
Picture this
Imagine dropping food packets into select households in different neighborhoods—some neighborhoods get food for 20% of homes, while others get food for 90% of homes. By varying the density of aid across distinct neighborhoods, researchers can measure how much neighbors share food with each other without confusing it with the direct effect on the recipients.
What the evidence says
Moderate transport subsidy saturation (40%) generated positive indirect spillovers on untreated neighbors, raising formal employment by 6.2 percentage points (p < 0.10) and permanent employment by 6.4 percentage points (p < 0.05); high saturation (90%) produced negative spillovers on permanent work (-6.8 percentage points, p < 0.01).
- Who
- N = 3,052 young job-seekers across 178 geographic clusters in Addis Ababa, Ethiopia.
- How
- Randomized saturation design varying cluster-level treatment proportions (20%, 40%, 75%, and 90%) across geographic clusters located over 2.5 km from the city center.
What to do
Vary the geographic density of treatment assignment across intervention clusters to isolate indirect peer effects and social network information transmission from direct policy impacts.
From the source
"We document a positive indirect effect on formal and permanent work among control individuals in clusters with 40 percent saturation. We also document that untreated individuals in clusters with 90 percent saturation are 5.6 percentage points less likely to be in permanent employment than individuals in pure control clusters."
Anonymity or Distance? Job Search and Labour Market Exclusion in a Growing African City