Operational Dosage: The Impact of Capacity Constraints on RCT Evaluation скачать в хорошем качестве

Operational Dosage: The Impact of Capacity Constraints on RCT Evaluation

Hannah Li (Columbia University) https://simons.berkeley.edu/talks/han... Bridging Prediction and Intervention Problems in Social Systems We study RCTs evaluating service interventions—proactive outreach by teachers, medication adherence support from healthcare providers, or social worker home visits—where treatment is delivered by capacity-constrained resources. When participants share finite service capacity, adding more participants can reduce the timeliness or intensity of service that others receive, introducing interference and hidden variation in treatment that we term "operational dosage." Using queueing theory, we show that treatment effects are both capacity- and sample-size-dependent, and can decrease once sample size exceeds a critical threshold. Consequently, statistical power in service intervention RCTs can peak at intermediate sample sizes, contradicting conventional power calculations. Simulations calibrated to a tuberculosis intervention trial in Kenya demonstrate that high-capacity/small-sample designs can achieve the same power as low-capacity/large-sample designs. Our results highlight the importance of capacity selection in experiment design and provide a mechanism for replication failures and implementation challenges at scale.