Source-linked AI summary
Understanding Behavioral Dark Patterns of High BMI Individuals
Manjeet Yadav, Prasenjit Karmakar, Suchetana Chakraborty
TL;DR
Existing research provides limited temporal and culturally diverse evidence about everyday behaviors associated with BMI. This paper analyzes four weeks of passive smartphone sensing and EMA data from 453 students across eight countries using correlations and an HMM. Higher BMI is associated with more soda, alcohol, processed meat, and food-delivery-app use, while overweight and obese participants show recurring transitions back to unhealthy routines after exercise.
Problem
Existing studies rely heavily on self-reports or limited sensing and often use single-country or self-selected cohorts, limiting temporal and broader-population evidence about BMI-related behavior.
Method
The study analyzes four weeks of passive smartphone sensing and EMA data from 453 university students across eight countries using Spearman correlations and a Hidden Markov Model.
Results
Higher BMI is associated with more frequent soda, alcohol, and processed-meat consumption, greater food-delivery-app use, and recurring returns to unhealthy routines after exercise.
Takeaways & Limitations
The findings identify dietary, digital, and temporal behavioral patterns associated with body weight, including routines that make weight loss challenging for overweight and obese individuals.
Takeaways & Limitations
The findings should be supported by larger, more diverse cohorts from general adult populations across several countries, with country-stratified analysis.
Abstract
from arXiv · showhide
Understanding how everyday behaviors influence body weight is essential for designing effective and personalized health interventions. Existing studies largely rely on self-reported questionnaires or limited sensing modalities, making it difficult to capture the temporal dynamics of daily behavior. In this work, we analyze the DiversityOne dataset, comprising four weeks of passive smartphone sensing and ecological momentary assessments collected from 453 university students across eight countries. We extract behavioral features spanning dietary habits, physical activity, screen time, and smartphone usage, and investigate their associations with self-reported Body Mass Index (BMI). Beyond feature-level analysis, we employ Hidden Markov Models (HMMs) to uncover latent behavioral patterns. Our analysis reveals that higher BMI is associated with more frequent consumption of soda, alcohol, and processed meat. We further reveal that overweight and obese individuals spend longer periods in food delivery apps and are more likely to transition back to unhealthy eating and drinking routines after starting to exercise. In contrast, normal-weight individuals lead a more balanced lifestyle. These findings highlight key behavioral patterns that make weight loss particularly challenging.
1 Introduction
The paper addresses gaps in understanding how everyday behaviors relate to BMI across cultures and over time. Using DiversityOne smartphone sensing, EMA data, correlations, and HMMs, it identifies dietary and digital behaviors associated with BMI and routines that may make weight loss challenging.
- Motivation: Existing studies rely heavily on self-reported questionnaires and food diaries, which provide sparse snapshots of everyday behavior.These approaches are also described as expensive to administer and prone to recall bias.
- Motivation: Smartphone-based body-weight research remains underexplored and often uses single-country or self-selected health-app cohorts, limiting generalizability.The paper also notes systematic bias in self-reported dietary information among individuals with higher BMI.
- Approach: DiversityOne provides approximately four weeks of passive sensing and EMA data from 453 university students across eight countries and four continents.Participants reported meals, snacks, activities, and contextual information alongside passive sensing streams.
- Research questions: The study asks which lifestyle choices consistently associate with BMI across cultures and how obese individuals’ routines make weight loss challenging.The questions cover diet, physical activity, digital engagement, and daily behavioral routines.
- Findings: Alcohol, soda, and processed-meat consumption are associated with higher BMI, while obese participants spend more time in food-delivery apps and may relapse to junk eating after exercise.Alcohol consumption per meal rises from 3.5% among normal-weight individuals to 9.8% among obese individuals.
2 Dataset Preparation
The study prepares a multinational smartphone-sensing and EMA dataset by selecting participants with BMI categories and at least one week of data, then correcting key logging problems before behavioral analysis. Daily dietary, digital, and physical features are aggregated for correlation and temporal modeling.
- Dataset: The analysis selects 453 participants from a four-week, eight-country university study who contributed at least one week of data and reported BMI categories.The selected BMI groups include 42 underweight, 326 normal-weight, 70 overweight, and 15 obese participants.
- Dataset: Participants used iLog, which continuously collected 26 smartphone modalities, and completed EMAs covering activity, location, social context, mood, and eating patterns.The dataset combines passive sensing with detailed dietary self-reports and BMI data.
- Data quality: The preparation pipeline created globally unique participant identifiers because 182 IDs were reused across countries.Country and participant ID were combined to resolve cross-country identifier collisions.
- Data quality: Daily step estimates were corrected by summing positive consecutive differences and capping accumulation at five steps per second.This addressed implausible totals caused by device restarts and sensor resets, including some days exceeding 2 million steps.
- Data quality: Screen sessions were split at midnight, sessions longer than 180 minutes were discarded, and app usage was estimated only while the screen was on.These corrections addressed calendar misassignment, artifact sessions, and inflated polling counts.
- Feature construction: Daily dietary, digital, and physical features were aggregated across varying sampling rates to form behavioral sequences for analysis.The features include EMA-derived dietary measures and iLog-derived screen and app usage measures.
3 Observations
The analysis identifies dietary, digital, and physical behaviors associated with BMI, then uses an HMM to characterize recurring routines across BMI groups. Higher-BMI participants show heavier eating, shorter-lived exercise, and relapse toward unhealthy states, whereas normal- and underweight participants show more sustainable transitions.
- 3.1 RQ1 – Behavior vs BMI: Dietary behaviors show the strongest BMI associations, with higher-BMI participants reporting more soda, alcohol, and processed meat consumption.Healthy and unhealthy snack features were excluded from temporal analysis because their associations were inconsistent, while late-night meals showed weak associations.
- 3.1 RQ1 – Behavior vs BMI: Food-delivery app use rises from 14% among normal-weight participants to 18% among overweight and 21% among obese participants.Food-delivery app use had the strongest digital association with BMI (ρ=0.219, q<0.001).
- 3.2 RQ2 – Routine vs BMI: The HMM identified four latent behavioral states from daily dietary, digital, and physical features, with labels assigned post hoc from estimated emission means.The temporal analysis excluded healthy and unhealthy snacks, late-night meals, and walking based on the preceding average-behavior analysis.
- 3.2 RQ2 – Routine vs BMI: Normal- and underweight participants predominantly occupy inactive S1, occasionally exercise in S2, and transition from healthy S0 to unhealthy S3 only 7% of the time.S1 occupancy is 63% with 84.68% recurrence; exercise days mainly return to inactivity or healthy eating.
- 3.2 RQ2 – Routine vs BMI: Overweight and obese participants remain in heavy-eating S0 with 47.46% recurrence, then often relapse from short-lived exercise S2 to S0 or unhealthy S3.Transitions from S2 go 24% to S0 and 22% to S3, while S3 returns to heavy eating 18%.
4 Limitations and Future Work
The study is observational and constrained by imbalanced BMI groups, a university-student sample, and unmodeled country-specific food environments. Future work should use larger, more diverse adult cohorts and add contextual information such as mood and mental well-being.
- Limitations: The analysis is observational and uses a cross-country dataset of participants with normal and high BMI.
- Limitations: BMI groups are highly imbalanced, with only 15 obese participants, limiting statistical power for obesity-specific behavioral patterns.The dataset includes 42 underweight, 326 normal-weight, 70 overweight, and 15 obese individuals.
- Limitations: Because all participants are university students, findings may not generalize to children or older adults.Their routines, living environments, and smartphone usage may differ from those of the broader adult population.
- Limitations: The analysis overlooks dietary and digital behaviors shaped by local food environments, cultural practices, and country-specific food-delivery availability.
- Future Work: Future work should use larger, more diverse general-adult cohorts across countries and examine contextual factors such as mood and mental well-being.
5 Conclusion
Using four weeks of sensing and EMA data from 453 participants across eight countries, the study links higher BMI with unhealthy dietary choices, food-delivery app use, and recurring relapse patterns after exercise.
- Higher BMI is most strongly associated with soda, alcohol, and processed-meat consumption and increased food-delivery app use.
- Normal-weight individuals generally follow a sustainable cycle of healthy eating and periodic exercise.
- Overweight and obese individuals exhibit a recurring behavioral dark pattern in which short-lived exercise is followed by relapse into unhealthy eating and drinking.This recurring pattern makes weight loss challenging.