Unpacking Habit With Bayesian Mixed Models: Dynamic Approach to Health Behaviors With Interchangeable Elements, Illustrated Through Multiple Sun Protection Behaviors

Abstract

Analytics for behavioral habit typically model one behavior at a time, despite the fact that habit often involves multiple cooccurring behaviors, such as food choices and physical activities, where interrelated behaviors are often equally recommended. We propose a novel Mixed-Effects Dynamic hAbit model (MEDA) to simultaneously model multiple related, habitual behaviors. As an illustrative example, MEDA was applied to real-time assessments of sun protection (sunscreen, shade, hat, and protective clothing) reported twice daily by first-degree relatives of melanoma patients who are themselves at an elevated risk of skin cancer. MEDA aims to explicate habits in sun protection under varying environmental cues (e.g., sunny and hot weather). We found consistent between-group differences (e.g., men responding to weather cues more consistently than women) and interactions between cooccurring behaviors (e.g., being in shade discourages sunscreen wearing, more so in men than women). Moreover, MEDA transcends conventional methods to address longstanding challenges—how cue to action and volitional choices differ by groups or even by individual persons. Such nuances in interrelated habitual behaviors are relevant in numerous other applications, such as recommended dietary or physical activity behaviors. These methods best inform personalized behavioral interventions targeting individual preferences for precision behavioral intervention.

Keywords

• habit; sun protection; bayesian hierarchical models; variance heterogeneity; conditional probability