Comparative Perspective of Human Behavior Patterns to Uncover Ownership Bias among Mobile Phone Users

Ayumi Arai, Zipei Fan, Dunstan Matekenya, Ryosuke ShibasakiISPRS Int. J. Geo Inf 2016 · DOI · OpenReview

TL;DR

This paper tackles the critical issue of ownership bias in mobile phone Call Detail Record (CDR) data, which often misrepresents the general population. By comparing CDR-derived routines with those from diary surveys, the authors reveal that CDRs tend to underrepresent students and overrepresent working males and housewives, leading to skewed insights into human mobility and behavior.

What Problem Does This Paper Address?

Understanding human mobility is crucial for urban planning and public policy. Mobile phone Call Detail Records (CDRs) offer a rich source of spatiotemporal data for this purpose. However, a significant challenge arises from "ownership bias": CDR data only reflects the behavior of mobile phone users, who may not be representative of the entire population. This is particularly problematic in developing countries where mobile phone ownership varies significantly across demographic groups (e.g., by gender and age). Without accounting for this bias, analyses based on CDRs can lead to misleading conclusions about societal issues. This paper aims to quantify and understand this discrepancy by comparing the behavior patterns derived from CDRs with those of the general population.

Methods

The study employs a multi-faceted approach:

  • Topic Modeling (LDA): A Latent Dirichlet Allocation (LDA) topic model is adapted to analyze CDR data. Each user's CDRs are treated as a "document," and individual call records (timestamp and location) are treated as "words." This model extracts "topics" that represent typical calling patterns and routines. The model is extended to incorporate both temporal (time of day, day of week) and spatial (location) information.
  • Routine Extraction and Comparison:
    • CDR Data: The adapted LDA model is used to extract typical routines from CDRs of mobile phone users.
  • Diary Survey Data (Mobile Phone Users): A field survey (SPACE) collected diary data from mobile phone users, capturing their activities and attributes. This data is used to profile the principal components of mobile phone users and validate the routines extracted from CDRs.
  • Diary Survey Data (General Population): A separate diary survey (Person Trip data) of the general population in Dhaka is used to establish a baseline for comparison. This data includes demographic information and daily activities.
  • Statistical Analysis: Proportions of different demographic groups and their activities are calculated and compared between the general population and mobile phone users.

Datasets

  • Call Detail Records (CDRs): CDRs from August 2013 were obtained from a leading mobile network operator in Bangladesh. A random sample of 5000 unique user IDs was used.
  • SPACE Survey Data: A field survey conducted from November 2013 to January 2014 in Greater Dhaka, collecting one-day diary records and personal attributes from 922 mobile phone users.
  • Person Trip (PT) Data: Diary survey data of the general population in Greater Dhaka, collected in 2009 by the Japan International Cooperation Agency (JICA), involving 75,000 individuals.

Key Insights and Ideas

  • Distinct Behavioral Archetypes: The analysis reveals two dominant archetypes among mobile phone users in Dhaka:
  • Working Males: Primarily engaged in income-generating activities outside the home.
  • Housewives: Primarily engaged in household tasks, spending most of their time at home.
  • Student Underrepresentation: A significant finding is the stark underrepresentation of students in CDR data. While students form a substantial portion of the general population, their presence in mobile phone user data is minimal. This highlights a critical bias when using CDRs to study younger demographics.
  • Spatial and Temporal Routine Extraction: The adapted LDA model effectively captures routine behaviors by considering both where people are and when they are there, linking these to distinct "topics" or activity patterns.
  • The Power of Combined Data: The study demonstrates the value of combining CDR data with traditional survey data. CDRs provide a large-scale, dynamic view, while surveys offer crucial demographic context and ground truth for interpretation, helping to uncover biases.

Key Results

  • Dominant Mobile User Groups: The study found that working males and housewives constitute the majority of mobile phone users in Dhaka.
  • Behavioral Patterns:
    • Working males exhibit patterns of spending most of the day outside the home, with peaks in activity around 8 a.m. and 7 p.m.
  • Housewives show a strong tendency to be at home, with near 100% probability throughout the day.
  • Students, though a significant group in the general population, are poorly represented in CDRs.
  • Ownership Bias Quantification: Table 7 shows that while working males and housewives are well-represented in both general and mobile phone user populations, students are significantly underrepresented in mobile phone user data (4% vs. 32% in the general population).

Limitations and Future Work

  • Limited Complexity of Lifestyles: The authors acknowledge that the activity patterns analyzed are not overly complex, potentially due to strong social norms in Dhaka. Applying the methodology to areas with more diverse lifestyles might require further analysis.
  • Cost of Data Acquisition: Conducting large-scale field surveys is expensive.
  • Future Work:
    • Census Data Utilization: The authors propose using census data for future studies to reduce data acquisition costs and provide a broader demographic context.
  • Broader Application: The approach to identifying device domain-specific bias is suggested to be extendable to other areas and data types.
  • Deeper Analysis: Further analysis on the relationship between principal population group features and their daily routines is recommended.

Summary updated

BibTeX
@article{arai2016comparative,
  author = {Ayumi Arai and Zipei Fan and Dunstan Matekenya and Ryosuke Shibasaki},
  title = {{Comparative Perspective of Human Behavior Patterns to Uncover Ownership Bias among Mobile Phone Users}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2016},
  doi = {10.3390/ijgi5060085},
}
Figure