Extended Abstract
Combatting the rise of the drug epidemic is a central challenge of U.S. health care policy. A large clinical literature has demonstrated the effectiveness of offering incentive payments for healthy behaviors to those with substance use disorders, showing that incentives decrease substance use and medical costs. However, adoption of such programs has been limited, in part because most existing approaches are implemented in-person and are not scalable due to high costs and logistical complexity. This project evaluates a scalable incentive program delivered through a mobile application. The app provides a “turnkey" solution that enables remote monitoring of behavior; for example, drug tests can be administered in patients' homes, as patients submit selfie-videos showing them taking saliva drug tests, which are then verified by trained remote staff. We are evaluating impacts on patient substance use and treatment enrollment, as well as on healthcare costs and employment.
In addition to evaluating the program, our experiment is designed to investigate a key open question in the literature: how to dynamically adjust incentives in response to behavior. One way to do so is with escalating schedules that feature incentive amounts that increase as individuals comply with the behavior, and decrease with failures to comply. Escalating incentive schedules have good dynamic incentive properties, and are frequently tested in substance-use settings. However, escalating schedules may be poorly targeted, paying the largest incentives to individuals who are not struggling to abstain, and offering the smallest incentives to those who are struggling. De-escalating schedules can address the poor targeting of escalating schedules. De-escalating schedules feature incentive payments that increase when individuals fail to comply with the behavior. They hence offer larger incentives to those “at the bottom of the distribution” (i.e., who are struggling the most to abstain), which is a desirable feature in the addiction context, as the costs of substance abuse may be convex. While de-escalating schedules create a perverse incentive to “shirk” early on, the magnitude of this effect may be small if participants are not very forward-looking. Our experiment will test both escalating and de-escalating incentive schedules to shed light on their cost-effectiveness and distributional properties. We also use experimental variation to directly test whether participants respond to dynamic incentives (e.g., whether they respond differently to escalating vs. de-escalating schedules conditional on the reward for current behavior).