Molly Offer-Westort

Response adaptive experimental designs use algorithmic treatment assignment procedures to optimize design as experiments unfold. In addition to improving outcomes over time, these designs are more effective than conventional randomized control trials for best arm identification and policy learning problems. However, uptake of these designs in research and policy settings where objectives are inferential is still limited. I discuss considerations for design decisions and procedures for analysis for applied researchers; I present non-contextual and contextual applications to Facebook Messenger studies.