Missing data in microrandomized trials: Challenges and opportunities

The vision of leveraging digital technologies to deliver real-time psychological interventions in everyday settings is realized via just-in-time adaptive interventions (JITAI) - an intervention design that guides the use of rapidly changing information about a person's internal states and contexts to decide whether and how to intervene in daily life. Microrandomized trials (MRTs) were developed as an experimental design to address scientific questions about how to best construct JITAIs, enabling scientists to investigate whether, what type, and under what conditions, intervention delivery can promote behavior change. However, missing data present challenges to the ability of MRTs to inform the development of JITAIs. This article articulates the multiple sources of missing data that can manifest in MRT studies, discusses how such missing data can impact (1) bias, (2) variance, and (3) the future implementation of JITAIs, and discusses strategies for both minimizing missing data in an MRT design and handling missing data when they occur. The overarching goal is to provide a conceptual framework that will guide future investigators in anticipating missing data and making informed decisions to manage them. Throughout, we illustrate concepts using existing data from the Mobile Assistance for Regulating Smoking (MARS) study. MARS (n = 99) involved a 10-day MRT that included up to six randomizations per person per day.