Inferring Mobility of Care Travel Behavior From Transit Origin-Destination Data

Daniela Shuman, Awad Abdelhalim, Anson Stewart, Kayleigh Campbell, Mira Patel, Inés Sánchez de Madariaga, Jinhua ZhaoarXiv 2022 · DOI · arXiv

TL;DR

This paper introduces a novel, data-driven workflow to identify "mobility of care"—trips related to caregiving and household maintenance—using transit fare card data rather than traditional surveys. By combining name-based gender inference with geospatial analysis of Washington D.C.’s transit network, the researchers demonstrate that women disproportionately perform these complex, time-consuming journeys, offering transit agencies a scalable blueprint for more equitable service planning.

What Problem Does This Paper Address?

Public transit systems have historically been designed around the "peak-hour commuter"—the person traveling from home to a central business district and back. This model often ignores Mobility of Care (MoC): the umbrella concept for trips associated with the "reproduction of life," such as escorting children to school, grocery shopping, or taking elderly relatives to appointments.

Existing research, largely based on small-scale surveys, shows that women perform the lion's share of these trips. However, surveys are expensive and provide only a snapshot in time. There is a significant "evidence gap" in using automated data (like fare card taps) to track these behaviors. The challenge is that fare card data is usually anonymous and doesn't explicitly label a trip as "dropping off a child." This paper seeks to bridge that gap by inferring demographics and trip purposes from big data.

Methods

The authors developed a multi-stage workflow applied to the Washington Metropolitan Area Transit Authority (WMATA) system:

  1. ODX Inference: They used an existing "Origin, Destination, and Transfer" (ODX) model to turn raw fare card "taps" into complete journeys, even when riders only tap in (common on buses).
  2. Name-Based Gender Inference: Since WMATA has a database of registered SmarTrip cards, the authors used first names to infer gender via the Genderize.io API and US Social Security records. They validated this against a smaller set of self-reported data, achieving 94% accuracy.
  3. Geospatial POI Mapping: They identified "Places of Interest" (POIs)—daycares, schools, and grocery stores—and developed an algorithm to find the specific bus stops most likely to serve those locations within a 400m radius.
  4. Trip-Chaining & Accompaniment Analysis:
    • They looked for "trip-chaining" (e.g., boarding a second bus shortly after alighting near a daycare).
  5. They identified "accompaniment patterns" where two cards tap in rapid succession (within 30 seconds), specifically when one card uses a student, senior, or disability fare product.

Datasets

  • WMATA SmarTrip Data (Q1 2019): A massive dataset containing ~80 million stages and ~70 million journeys from 3 million unique cards.
  • Registered User Data: A subset of 1.9 million cards with name information used for gender inference.
  • DC Open Data: Geolocation coordinates for daycares, schools, and grocery stores in the D.C. metro area.
  • Validation Set: A WMATA customer survey of ~5,000 respondents with self-reported demographics.

Key Insights and Ideas

  • The "Double Burden" in Data: The study found that even among full-time employees (who are overrepresented in the registered card data), women still perform significantly more MoC trips. This suggests that employment doesn't equalize the caregiving travel burden.
  • The Complexity of Care: MoC trips are not just "errands"; they are statistically more "inconvenient." They involve more transfers and longer in-vehicle times compared to direct commutes.
  • Validation of the "Pink Tax" on Time: By showing that women are more likely to trip-chain near POIs, the paper quantifies how caregiving adds layers of time and complexity to a woman's daily transit experience.
  • Policy as a Tool: The authors note that WMATA’s policy of free bus-to-bus transfers within two hours is a vital (if perhaps unintentional) support for MoC, as these trips often require multiple short legs.

Key Results

  • Gender Parity Deviation: In an equally sampled group of men and women, women were 10% to 15% more likely than men to be found at MoC-related stops (daycares/schools) during morning drop-off hours.
  • The Accompaniment Gap: Women were significantly more likely to exhibit consistent accompaniment patterns with children, seniors, or people with disabilities. As the frequency of accompaniment increased, the likelihood of the caregiver being a woman rose sharply.
  • Travel Burden: MoC trips were found to be, on average, 10 minutes longer in-vehicle than non-MoC trips between the same origin-destination pairs.
  • Volume: The geospatial analysis identified approximately 2 million MoC trips out of 54 million multi-modal journeys made by active users.

Limitations and Future Work

  • Selection Bias: The study relied on registered fare cards. These users are more likely to be full-time employees receiving transit subsidies. This likely excludes lower-income riders or those in the informal economy who may perform even more MoC trips.
  • Binary Gender Constraints: The name-based inference is limited to a binary (male/female) classification, which does not account for non-binary identities or the nuances of gender expression.
  • The "Why" vs. the "What": While the data shows where and when people travel, it cannot capture the qualitative experience—such as the stress of traveling with a stroller on a crowded bus.
  • Future Directions: The authors suggest expanding the POI list (e.g., healthcare facilities) and using this workflow to test how specific service changes (like "open stroller" policies) specifically benefit caregivers.

Summary updated

Abstract

There are substantial differences in travel behavior by gender on public transit. Studies have concluded that these differences are largely attributable to household responsibilities typically falling disproportionately on women, leading to women being more likely to utilize transit for purposes referred to by the umbrella concept of "mobility of care". In contrast to past studies that have quantified the impact of gender using survey and qualitative data, we propose a novel data-driven workflow utilizing a combination of previously developed origin, destination, and transfer inference (ODX) based on individual transit fare card transactions, name-based gender inference, and geospatial analysis as a framework to identify mobility of care trip making. We apply this framework to data from the Washington Metropolitan Area Transit Authority (WMATA). Analyzing data from millions of journeys conducted in the first quarter of 2019, the results of this study show that our proposed workflow can identify mobility of care travel behavior, detecting times and places of interest where the share of women travelers in an equally-sampled subset (on basis of inferred gender) of transit users is 10% - 15% higher than that of men. The workflow presented in this study provides a blueprint for combining transit origin-destination data, inferred customer demographics, and geospatial analyses enabling public transit agencies to assess, at the fare card level, the gendered impacts of different policy and operational decisions.
BibTeX
@misc{shuman2022inferring,
  author = {Daniela Shuman and Awad Abdelhalim and Anson Stewart and Kayleigh Campbell and Mira Patel and Inés Sánchez de Madariaga and Jinhua Zhao},
  title = {{Inferring Mobility of Care Travel Behavior From Transit Origin-Destination Data}},
  year = {2022},
  eprint = {2211.04915},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2211.04915},
}
Figure