Not Now, Ask Later: Users Weaken Their Behavior Change Regimen Over Time, But Expect To Re-Strengthen It Imminently

Title of the Paper

Not Now, Ask Later: Users Weaken Their Behavior Change Regimen Over Time, But Expect To Re-Strengthen It Imminently

Paper Information

• Domain: Human-Computer Interaction (HCI) and Behavior Change Technologies
• Keywords: Behavior change, productivity interventions, habit formation, user retention, online distraction, intervention difficulty, user preferences
• Conference: CHI ’21 (Conference on Human Factors in Computing Systems)
• DOI: 10.1145/3411764.3445695

Research Background and Problem

• Problems and Challenges:

  • In behavior change interventions aimed at controlling online distraction habits, do users consistently adhere to higher difficulty interventions?
  • If users abandon these interventions, will they attempt them again or permanently give up?
  • How can behavior change systems design adaptive intervention programs as user preferences evolve over time?

• Significance:

  • Online distractions impact work efficiency and the achievement of long-term goals.
  • Existing productivity tools face user retention issues (e.g., high dropout rates) and require optimization of intervention models.

• Motivation and Related Work:

  • Users are often overly optimistic about their behavior change goals but tend to gradually reduce intervention difficulty or terminate interventions over time.
  • Current research lacks systematic exploration of how user intervention preferences dynamically change.
  • The reasons for long-term user attrition in behavior change tools (e.g., HabitLab) have not been thoroughly studied.

Solution

• Proposed Method:

  • Utilize the online behavior change platform HabitLab to study users’ choices and changes in intervention difficulty over time.
  • Conduct three studies to explore:
    1. Temporal trends in user preferences for intervention difficulty.
    2. The impact of prompting users to select intervention difficulty on retention rates.
    3. User preferences regarding the frequency of prompts.

• Innovations:

  • Provide quantitative analysis of user preferences shifting from initially high difficulty to lower difficulty or no intervention.
  • Identify trade-off strategies between prompt frequency and user retention.
  • Reveal contradictions between users’ actual choices and expectations (e.g., optimistic restart behavior).

• Implementation Steps and Key Techniques:

  • Record and analyze HabitLab user behavior data, including initial preferences, subsequent intervention choices, and prompt interval parameters.
  • Use Cox proportional hazards regression models to analyze the effect of prompt frequency on user attrition.
  • Apply interaction designs such as customized prompt frequency options to observe user choices regarding prompts.

Research Findings

• Specific Results:

  1. Users initially choose intervention difficulties categorized as "medium" or "high," but these preferences significantly decrease to "low difficulty" or "no intervention" over time.
  2. User preferences are not static, necessitating dynamic adaptation of behavior change systems to meet evolving user needs.
  3. Low-frequency prompts (e.g., displayed once every 25% of visits) significantly improve user retention while having limited impact on the accuracy of predicting user preferences.
  4. Despite weakened preferences, most users express optimism about resuming higher difficulty interventions in the future (optimistic restart).

• Comparison with Existing Solutions:

  • HabitLab’s dynamic prompting and intervention adjustment methods outperform traditional static intervention designs.
  • This study is the first to quantify the temporal changes in user intervention preferences within behavior change systems.

• Experimental or Evaluation Results:

  • User revisit data shows that 73% of users choose the "no intervention" mode by their 200th visit.
  • Excessive prompt frequency (increasing to 100% of visits) leads to significant user attrition.
  • Users accept low-frequency prompts (e.g., weekly or daily notifications) while maintaining over 90% prediction accuracy.

• Limitations and Future Directions:

  • Limitations:
    • Subjective user ratings of intervention difficulty may vary due to individual differences.
    • Data analysis focuses on HabitLab’s user base, and generalizability to other tools remains to be validated.
  • Future Directions:
    • Explore more precise intervention designs based on psychological models (e.g., cognitive dissonance theory).
    • Develop more effective commitment devices to help users sustain higher difficulty behavior changes over the long term.
    • Extend research to other domains, such as health behavior change and intervention optimization in education.

Brief Explanation of Output Format

• Content is presented in Markdown format, making it suitable for direct use in literature notes or academic writing.
• Key information is distilled with accurate and complete sources, enabling rapid understanding of the main logic and contributions of the research.