Re-conceptualizing SMART designs as a hybrid of randomized and regression discontinuity designs: opportunities, cautions

Abstract

ABSTRACT Though previously unacknowledged, a SMART (Sequential Multiple Assignment Randomized Trial) design uses both regression discontinuity (RD) and randomized controlled trial (RCT) designs. This combination structure creates a conceptual symbiosis between the two designs that enables both RCT- and previously unrecognized, RD-based inferential claims. In Phase 1, based on a quantitative cut-point, a SMART design yields successful or unsuccessful participants after the initial RCT. In Phase 2, interventions are randomly assigned to successful and unsuccessful, Phase 1 subgroups, thereby yielding both RCT and RD results. Like RCT-based inferences, many but not all newly recognized, RD-based SMART inferences are potentially unbiased. In particular, comparisons between RCT and RD outcomes are potentially problematic. Principles from the within-study comparison literature are used to inform SMART-based research, thereby introducing otherwise ignored inferential and procedural possibilities. Awareness of RD designs embedded within SMARTs encourages previously unrecognized inferential claims, though potential internal validity threats and limitations must be carefully weighed. Finally, new, SMART-like sequences of quasi-experiments and experiments can be tested and embedded logic applied.