Theory-based habit modeling for enhancing behavior prediction in behavior change support systems

Chao Zhang, Joaquin Vanschoren, Arlette van Wissen, Daniël Lakens, Boris E. R. de Ruyter, Wijnand A. IJsselsteijn

Abstract

Behavior Change Support Systems (BCSSs) often rely on intrusive self-reports to assess user habit strength. This study evaluates computational models derived from psychological theory as a passive alternative for behavior prediction. Results indicate that theory-based models significantly outperform traditional survey methods and provide a principled framework for feature engineering in behavioral health applications.

Introduction

The effectiveness of Behavior Change Support Systems (BCSSs) is often limited by their inability to sense the "invisible" cognitive states of users, such as habit strength. Habitual behaviors are characterized by automaticity and are triggered by environmental cues rather than conscious intention. To optimize interventions—avoiding redundant prompts for strong habits while providing necessary support for weak ones—systems must accurately estimate these states.

Traditionally, the Self-Report Habit Index (SRHI) has been the gold standard for measurement. However, surveys are prone to memory bias and can be burdensome for users. This research explores the feasibility of using theory-based computational models to estimate habit strength from passive sensor data.

Methodology

The researchers compared three approaches to predict next-day toothbrushing behavior:

  1. Survey Approach: Features derived from weekly self-reports on attitude and intention.
  2. Past-Behavior Approach: Simple statistics based on the frequency of past behavior.
  3. Theory-Based Approach: Computational models based on Hebbian learning principles, where habit strength ($HS$) is updated daily based on behavior performance and environmental consistency.
Computational modeling of habit strength (HS) showing acquisition and extinction phases based on the Miller et al. (2019) framework.

The study utilized data from two dental health interventions involving 111 participants. Objective behavior data was collected using manual toothbrushes equipped with Axivity AX3 accelerometers.

Results

The theory-based models achieved predictive accuracies of 68.6% and 76.1% across two studies, significantly outperforming survey-based models ($p < .01$).

Comparison of mathematical models (Miller et al., 2019 and Tobias, 2009) used to translate psychological theory into predictive features.

Key insights include:

  • Theory as Feature Engineering: Converting psychological theories into difference equations allows for the creation of "synthetic sensors" that estimate latent cognitive states.
  • Habit vs. Memory: Habit strength was found to be substantially more predictive (2.4x to 4.7x) than memory accessibility, reinforcing the importance of automaticity in established routines.
  • Performance Parity: While superior to surveys, theory-based models performed similarly to simple past-behavior averages for one-step-ahead prediction.

Discussion and Conclusion

This research demonstrates that passive, theory-based modeling is a viable and superior alternative to intrusive surveys for monitoring habit formation. While simple metrics are effective for short-term prediction, theory-based models offer a more robust foundation for understanding the underlying dynamics of behavior change. Future work should focus on integrating contextual cues and applying these models to a broader range of health behaviors.

Summary updated

Abstract

Psychological theories of habit posit that when a strong habit is formed through behavioral repetition, it can trigger behavior automatically in the same environment. Given the reciprocal relationship between habit and behavior, changing lifestyle behaviors is largely a task of breaking old habits and creating new and healthy ones. Thus, representing users’ habit strengths can be very useful for behavior change support systems, for example, to predict behavior or to decide when an intervention reaches its intended effect. However, habit strength is not directly observable and existing self-report measures are taxing for users. In this paper, building on recent computational models of habit formation, we propose a method to enable intelligent systems to compute habit strength based on observable behavior. The hypothesized advantage of using computed habit strength for behavior prediction was tested using data from two intervention studies on dental behavior change (\(N = 36\) and \(N = 75\)), where we instructed participants to brush their teeth twice a day for three weeks and monitored their behaviors using accelerometers. The results showed that for the task of predicting future brushing behavior, the theory-based model that computed habit strength achieved an accuracy of 68.6% (Study 1) and 76.1% (Study 2), which outperformed the model that relied on self-reported behavioral determinants but showed no advantage over models that relied on past behavior. We discuss the implications of our results for research on behavior change support systems and habit formation.
BibTeX
@article{zhang2022theory,
  author = {Chao Zhang and Joaquin Vanschoren and Arlette van Wissen and Daniël Lakens and Boris E. R. de Ruyter and Wijnand A. IJsselsteijn},
  title = {{Theory-based habit modeling for enhancing behavior prediction in behavior change support systems}},
  journal = {User Model. User Adapt. Interact.},
  year = {2022},
  doi = {10.1007/s11257-022-09326-x},
}
Figure