Estimating Sense of Agency from Behavioral Logs: Toward a Just-in-Time Adaptive Intervention System

Ryunosuke Togawa, Roberto Legaspi, Yasutaka Nishimura, Akihiro Miyamamoto, Bo Yang, Eriko Sugisaki, Kazushi Ikeda, Nao Kobayashi, Yasushi Naruse

Abstract: Sense of agency (SoA) is the subjective experience that one’s volitional action caused an event. It has been widely investigated that SoA plays a significant role in shaping and moderating human cognition, emotion, and behavior. Thus far, there is no method that accounts for recognizing SoA in real-time in complex natural settings. To this end, we estimated SoA from logs of human behavior phenotypes, which include heart rate readings, physical activities, smartphone-usage time, application usage and GPS. Using smartphone app and wearable device, we collected user behavioral logs and SoA ground truths (\(n=25\) , over 4,000 data points), and built and compared multiple models to estimate the SoA. Our results showed that models with daily behavioral logs as independent variables significantly outperformed the baseline model that contains demographic and personality information, confirming the predictive utility of daily behavioral logs for SoA estimation. We then discussed the viability of this model in developing just-in-time adaptive interventions.