The temporal trajectories of habit decay in daily life: An intensive longitudinal study on four health-risk behaviors

Abstract

Habits are cue-behavior associations learned through repetition that are assumed to be relatively stable. Thereby, unhealthy habits can pose a health risk due to facilitating relapse. In absence of research on habit disruption in daily life, we aimed to investigate how habit decreases over time and whether this differs by four health-risk behaviors (sedentary behavior, unhealthy snacking, alcohol consumption and smoking). This 91-day intensive longitudinal study included four parallel non-randomized groups (one per behavior; N = 194). Habit strength was measured daily with the Self-Report Behavioral Automaticity Index (11805 observations) and modelled over time with constant, linear, quadratic, cubic, asymptotic, and logistic models. Person-specific modelling revealed asymptotic and logistic models as the most common best fitting models (54% of sample) and suggest the time for successful habit disruption to occur to range from 1 to 65 days. Multilevel modelling indicated substantial between-person heterogeneity and suggested initial habit strength but not the disruption process to vary by behavioral group. Findings suggest that habit disruption typically follows a decelerating negative trend, but that it is a highly idiosyncratic process with multiple potential temporal trends. Recommendations for analyzing habit strength time series are discussed, emphasizing the role of person-specific modelling and data visualization.

HabitWalk: A Micro-Randomized Trial to Understand and Promote Habit Formation in Physical Activity

“…For each participant, we modelled the trajectory with five models: constant, linear, quadratic, asymptotic, and generalized additive (GAMs) (for further details on growth models' specification, refer to supplemental materials, Table S3). The first four models come from previous habit tracking studies (Edgren et al, 2024;Fournier et al, 2017;Keller et al, 2021;Lally et al, 2010). GAMs were introduced as they allow fitting smoothed functions by combining multiple low-level functions, ultimately providing a flexible approach to model non-linear trends that account for multiple and complex bends.…”