The dissertation's premise is that physical activity is a powerful protective factor against disease, yet a large share of the population falls short of recommended levels, and that conventional population-based (nomothetic) intervention design may be the wrong lens. Park's central hypothesis is that people's responses to walking prompts are simultaneously dynamic (time-varying), context-dependent, and idiosyncratic — but predictably so within each individual — and that these patterns would be largely invisible to classic population statistics, requiring instead idiographic (within-person) Bayesian modeling.
The method is an optimization trial framed as a system-identification experiment: an open-loop design in which several a priori intervention strategies (notifications and adaptive daily step goals) are presented to all participants and alternated over time, in a manner akin to a micro-randomized trial. It is "open-loop" because each person's exposure does not depend on their prior responses. Real-time Fitbit data drove a custom mobile app and an automated decision system that estimated each participant's dynamic state and tailored the intervention accordingly. On the analytic side, Park used mixed-effects models, Bayesian regression, and machine-learning models (including multilayer perceptrons) to interrogate the nomothetic, idiographic, and dynamic structure of response, choosing Bayesian models for their fit to N-of-1-style designs where interventions are experimentally varied within individuals over time.
The key reported result is that it was feasible to identify individualized states in which a person reliably increased steps in the three hours after support (versus no support in the same state) for 91% of participants with sufficient data (40 of 44), or 83% under an intent-to-treat framing (40 of 48); the dissertation further notes that up to 98% could be reached if need, opportunity, and receptivity were allowed to be considered independently. Park frames this as strong justification for the next step in the research program — folding these idiographic insights into a real-time, at-scale control optimization trial driven by a personalized controller.
Honest context: the dissertation is organized as a scientific record spanning concepts, design considerations, a reprinted protocol chapter, analysis protocols, and visually summarized results, and it explicitly positions itself as a stepping stone rather than a finished intervention — the controller-driven JITAI is named as future work. Its results are about the feasibility of identifying actionable individualized states, not a demonstrated long-term increase in physical activity; the latter is the explicit aim of the subsequent trial. The work also documents the substantial systems engineering — a dedicated study app, real-time data pipeline, and automated per-person decision-making — required to make idiographic experimentation possible at all.