Correcting for Transitory Effects in RCTs: Application to the RAND Health Insurance Experiment, joint with Mona Balesh Abadi and Farah Omran
The long run price elasticity of healthcare spending is critically important to estimating the cost of provision. However, even large scale randomized controlled trials may be confounded by transitory effects. This paper shows evidence of a `deadline effect’ — a spike in spending in the final year of the program — among participants of the RAND Health Insurance Experiment, long considered the definitive RCT in the field. The deadline effect is economically and statistically significant, with power to identify coming from random allocation to three- or five-year enrollment terms. The deadline effect interacts with the price elasticity: with participants who face lower coinsurance rates show larger spending spikes. Crucially, controlling for the price-deadline interaction yields significantly smaller estimates of the price elasticity in non-deadline years, which we argue is a reasonable approximation for the long run elasticity. This has important implications for public finance and the design of private/temporary subsidy programs, and contributes to an ongoing debate that cautions against naive interpretation of the treatment effects generated by RCTs.