Two-Stage School and Classroom Cluster Randomization
RCTClinical Trial
Standard cluster randomized evaluations that assign entire institutions to either treatment or control cannot measure intra-institutional information spillovers between treated and untreated peers.
Picture this
Instead of flipping a coin to decide if a whole school gets a program, evaluators flip two coins in stages: the first coin decides if a school gets the program, and the second coin chooses which specific classroom inside that school gets the lessons while leaving neighboring classrooms untreated to see if ideas leak across the hallway.
What the evidence says
Disentangled direct treatment impacts from intra-school spillovers, proving zero statistical spillover to untreated classrooms (0.011 SD knowledge gain, 0.022 SD attitude gain).
- Who was studied
- N = 4,599 human students in 138 classrooms across 69 public secondary schools in 21 Colombian cities.
- How
- Two-stage experimental design randomizing 69 schools into treatment or control, then randomizing classrooms within treatment schools into treated (46 classrooms) or untreated spillover (46 classrooms).
What to do
Structure institutional field trials with multi-stage sub-unit randomization to isolate direct treatment effects from localized peer contagion.
From the source
"Assignment to treatment is randomized both across 69 schools as well as within schools at the classroom level (in order to measure information spillovers within the school)."
Do Information Technologies Improve Teenagers’ Sexual Education? Evidence from a Randomized Evaluation in Colombia