Time-sensitive RCTs in behavioral public policy: a pragmatic framework using sequence methods, personalization, and reinforcement learning

This article presents a pragmatic framework for time-sensitive analysis of behavioral RCTs using sequence methods and Markov modeling. The focus is not methodological novelty but translation: we map common policy questions to appropriate temporal tools, provide a reporting checklist for transparency, and show how estimates become implementable rules for booster timing, triage, and exit. We position sequence analysis alongside multi-state hazards, HMMs, SMART/MRT, and g-methods, and we introduce an openly documented R package, sequenceRCT, that operationalises the end-to-end workflow with uncertainty quantification and reproducible outputs. A simulated illustration demonstrates interpretation and decision use, with ablations and a counterfactual booster vignette. We extend the framework to personalized interventions–where state-specific individual treatment effects are difficult to detect–and to reinforcement learning, where sequence-derived state spaces, empirical kernels, and off-policy evaluation support safe policy learning. We conclude with a staged validation agenda on existing datasets.