Mean Reversion in Randomized Controlled Trials: Implications for Program Targeting and Heterogeneous Treatment Effects

Eligibility criteria for interventions can induce an Ashenfelter Dip, and subsequent mean reversion results in improvement over time even absent the intervention. We investigate these dynamics for a food-as-medicine program to treat diabetes, where eligibility required elevated hemoglobin A1c (HbA1c). Both treatment and control groups experienced significant improvements in HbA1c, resulting in an estimated null effect. When we predict improvement using baseline characteristics, we find that subjects who are unlikely to improve on their own appear to benefit from the program. Our findings have implications for program targeting and estimating heterogeneous treatment effects.