Investigating Causality in Human Behavior from Smartphone Sensor Data: A Quasi-Experimental Approach

Fani Tsapeli, Mirco MusolesiarXiv 2015 · DOI · OpenReview

TL;DR

This paper introduces a novel quasi-experimental framework to infer causal relationships in human behavior using smartphone sensor data, moving beyond mere correlation. They demonstrate its effectiveness by showing that exercising and spending time outside the home/workplace causally reduce stress levels in students.

What Problem Does This Paper Address?

The ubiquitous nature of smartphones has led to an explosion of passively collected user data. While much research has focused on identifying correlations between smartphone-derived features and human behavior (e.g., stress, mood), understanding causation has remained a significant challenge. This is primarily because true causal inference often relies on randomized controlled trials (RCTs), which are ethically difficult or impossible to conduct for many human behaviors. Observational data, while abundant, is prone to confounding variables that can mask or mimic causal effects. This paper aims to bridge this gap by developing a method to infer causal links from this readily available observational smartphone data.

Methods

The core of the paper is a quasi-experimental framework for causal inference. This framework is built upon Rubin's counterfactual framework, which posits that a causal effect is the difference between what would have happened with and without a treatment. Since direct observation of both scenarios for the same individual is impossible, the method focuses on comparing treated and control groups that are similar in baseline characteristics (confounding variables).

Key methodological steps include:

  • Variable Definition: Identifying independent (treatment) variables (e.g., time spent exercising), dependent (outcome) variables (e.g., stress level), and confounding variables (e.g., personality traits, previous day's stress).
  • Unit Definition: Defining "units" as observations of individuals at specific time points, characterized by the defined variables.
  • Confounder Identification: Using Kendall rank correlation to identify variables that correlate with both the treatment and outcome variables.
  • Quasi-Experimental Design (Matching): Emulating RCTs by creating matched pairs of treated and control units. This involves defining treatment and control groups based on thresholds of the treatment variable (e.g., spending more or less time at university than average). They then use Genetic Matching to find control units that are as similar as possible to treated units based on confounding variables, aiming to minimize bias.
  • Average Treatment Effect (ATE) Estimation: Calculating the average difference in outcomes between matched treated and control units.
  • Balance Check: Using the standardized mean difference (SMD) to ensure that confounding variables are well-balanced between the treated and control groups after matching.

Datasets

The study utilizes the StudentsLife dataset. This dataset was collected from 48 students at Dartmouth College over 10 weeks and includes:

  • GPS location traces
  • Accelerometer data (activity classification)
  • Calendar data (deadlines)
  • Student responses to questionnaires about their stress levels and Big Five personality traits.

The paper does not introduce a new benchmark dataset but rather leverages an existing one to demonstrate their framework.

Key Insights and Ideas

  • Beyond Correlation: The paper strongly emphasizes the limitations of correlation analysis for understanding human behavior and highlights the necessity of causal inference for actionable insights.
  • Smartphone Data for Causality: Demonstrates that rich, passively collected smartphone data can be leveraged for sophisticated causal inference, not just correlational studies.
  • Exercise and Outdoors Time Reduce Stress: A significant finding is the strong positive causal effect of exercising and spending time outside home/work on stress reduction. This suggests that encouraging these activities could be a direct intervention for stress management.
  • University Time's Nuanced Impact: The time spent at university has a positive causal impact on stress, but only when it's lower than average, and this effect is not particularly remarkable. This suggests that while excessive university time might increase stress, moderate or lower-than-average time doesn't significantly alleviate it.
  • Personality Matters: The impact of these factors can differ based on personality traits. Extroverts benefit more from time spent outside, while neurotics benefit more from exercise. This points towards personalized interventions.

Key Results

  • Exercise and Outdoor Time: Spending more time exercising or outside the home/workplace causally reduces stress levels. For example, spending more time outside led to a ~3% reduction in stress for the general population, and this effect was amplified to ~14.45% when considering students who spent more time outside than average (a=0.15).
  • University Time: Spending less time at university than average resulted in a ~3.1% lower stress level for the general population. This effect was not statistically significant for extroverts.
  • Socialization Venues: While visiting socialization venues is correlated with stress, the study found no statistically significant causal link. This suggests that the benefit might come from general "time outside" rather than specific venue types.
  • Personality Subgroups:
    • Extroverts showed a stronger positive effect from spending time outside.
  • Neurotics showed a stronger positive effect from exercising.
  • Confounder Balance: The Genetic Matching method successfully balanced confounding variables, with standardized mean differences generally below 0.1, indicating minimal confounding bias.

Limitations and Future Work

  • Limited Sample Size and Representativeness: The study involved only 48 students, which is a small and likely unrepresentative sample. Extrapolating general conclusions about the causal impact of these factors on stress is therefore not feasible.
  • Inference Accuracy: Inferences about activities (e.g., assuming a visit to a sports center implies exercise) can be inaccurate due to noisy sensor data or misinterpretation of activities.
  • Uncertainty Quantification: Assessing and incorporating the uncertainty arising from inferring activities from sensor data is an area for future work.
  • Generalizability: The study focused on students, and the findings might not generalize to other populations.
  • Future Work: The authors suggest further investigation into the degree of uncertainty in sensor-based inferences and their incorporation into causal analysis. They also highlight the need for larger, more representative samples to draw broader conclusions.

Summary updated

Abstract

Smartphones have become an indispensable part of our daily life. Their improved sensing and computing capabilities bring new opportunities for human behavior monitoring and analysis. Most work so far has been focused on detecting correlation rather than causation among features extracted from smartphone data. However, pure correlation analysis does not offer sufficient understanding of human behavior. Moreover, causation analysis could allow scientists to identify factors that have a causal effect on health and well-being issues, such as obesity, stress, depression and so on and suggest actions to deal with them. Finally, detecting causal relationships in this kind of observational data is challenging since, in general, subjects cannot be randomly exposed to an event. In this article, we discuss the design, implementation and evaluation of a generic quasi-experimental framework for conducting causation studies on human behavior from smartphone data. We demonstrate the effectiveness of our approach by investigating the causal impact of several factors such as exercise, social interactions and work on stress level. Our results indicate that exercising and spending time outside home and working environment have a positive effect on participants stress level while reduced working hours only slightly impact stress.
BibTeX
@misc{tsapeli2015investigating,
  author = {Fani Tsapeli and Mirco Musolesi},
  title = {{Investigating Causality in Human Behavior from Smartphone Sensor Data: A Quasi-Experimental Approach}},
  year = {2015},
  eprint = {1505.02818},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1505.02818},
}
Figure