Source-linked AI summary
Gender gaps in urban mobility
Laetitia Gauvin, Michele Tizzoni, Simone Piaggesi, Andrew Young, Natalia Adler, Stefaan Verhulst, Leo Ferres, Ciro Cattuto
TL;DR
Gendered mobility remains poorly measured, despite its importance for understanding women’s and girls’ access to urban opportunities. The paper combines anonymized mobile-phone traces with socioeconomic, transport, and open data in Santiago, Chile, finding that women move less and that mobility gaps are associated with socioeconomic conditions and transport access. The study provides a large-scale basis for examining gendered mobility patterns, while its spatial measure is bounded by a defined distance around points of interest.
Problem
Women’s mobility needs are often overlooked because traditional studies have limited observations, time spans, spatial resolution, or self-reported data, while mobility is multidimensional.
Method
The study analyzes anonymized Call Detail Records from a large cohort over 3 months and combines mobility measures with socioeconomic, transport, and points-of-interest data.
Results
Women visit fewer locations than men and are more localized; access to public and private transport reduces mobility differences across socioeconomic contexts.
Takeaways & Limitations
Combining high-resolution telecommunications data with other datasets can elucidate relationships among gender, mobility, and poverty and identify gendered mobility needs.
Takeaways & Limitations
The spatial analysis uses a distance of 2 standard deviations on either side of each point-of-interest location.
Abstract
from arXiv · showhide
The use of public transportation or simply moving about in streets are gendered issues. Women and girls often engage in multi-purpose, multi-stop trips in order to do household chores, work, and study ('trip chaining'). Women-headed households are often more prominent in urban settings and they tend to work more in low-paid/informal jobs than men, with limited access to transportation subsidies. Here we present recent results on urban mobility from a gendered perspective by uniquely combining a wide range of datasets, including commercial sources of telecom and open data. We explored urban mobility of women and men in the greater metropolitan area of Santiago, Chile, by analyzing the mobility traces extracted from the Call Detail Records (CDRs) of a large cohort of anonymized mobile phone users over a period of 3 months. We find that, taking into account the differences in users' calling behaviors, women move less than men, visiting less unique locations and distributing their time less equally among such locations. By mapping gender differences in mobility over the 52 comunas of Santiago, we find a higher mobility gap to be correlated with socio-economic indicators, such as a lower average income, and with the lack of public and private transportation options. Such results provide new insights for policymakers to design more gender inclusive transportation plans in the city of Santiago.
1 Introduction
Urban mobility is gendered, but women’s and girls’ mobility needs are often overlooked because available data are limited and traditional studies have restricted coverage. This study uses large-scale mobile-phone traces and complementary datasets to quantify gender disparities in Santiago and relate them to socioeconomic and transport conditions.
- Motivation: Women’s mobility needs are rarely incorporated into urban and transportation planning, partly because robust data on their daily mobility are lacking.Gender-disaggregated data gaps can leave urban planning gender-blind and limit interventions to bridge mobility inequalities.
- Evidence gap: Traditional mobility studies often rely on few observations, short time spans, limited spatial resolution, or self-reported data.Observed mobility differences also remain difficult to attribute to physical sex differences versus socially constructed factors such as household roles.
- Evidence gap: Mobility is multidimensional, so no single dataset or approach is sufficient to provide decision-makers with a complete picture.Relevant data may also be collected by corporations and remain unavailable to researchers and policymakers.
- Study design: The study examines gendered urban mobility in Santiago, Chile, using anonymized mobile-phone mobility traces from a large cohort over 3 months.The analysis combines CDR-based mobility measures with socioeconomic indicators, transport-network information, and points of interest.
- Study objectives: Its objectives are to quantify gender disparities in Santiago residents’ mobility and identify socioeconomic and transport factors associated with mobility inequalities.The study also maps gender differences across comunas and identifies points of interest appearing more often along women’s or men’s trajectories.
2 Results
Across Santiago, women showed lower mobility than men across distance, location diversity, and temporal distribution, with wider gaps among lower-income groups and incomplete equalization from transport access. Gender differences also varied by visited place type, especially around hospitals, malls, and taxi stops.
- Data and approach: 418,624 individuals were analyzed from about 2 billion anonymized Call Detail Records collected over three months.The records were collected between May 1 and July 30, 2016.
- Gender inequalities in mobility: 8.58 locations was the average difference between men and women across all unique locations visited, with women visiting fewer.For core locations accounting for users’ daily activity, the difference was 2.02 locations.
- Gender inequalities in mobility: 1.09 Km was the shorter average radius of gyration for women, indicating more spatially localized movement than men.Women also had lower Shannon entropy by 0.26, distributing trips among fewer preferred locations while men spread trips more evenly.
- Socioeconomic differences: 2.50 locations was the core-location gap for the most deprived socioeconomic group, compared with 1.66 for the wealthiest ABC1 group.The gender gap in mobility widened as socioeconomic status decreased, although a smaller gap remained among the wealthiest users.
- Visited place types: ρF/ρM showed gender imbalances near hospitals, malls, and taxi stops, while ratios approached 1 at larger spatial distances.Spatially perturbing POIs moved the ratios toward unity and into the 95% confidence interval across a broad bandwidth range.
3 Discussion
The study shows that women’s mobility in Santiago differs from men’s across spatial, temporal, and destination patterns, with gender gaps widening amid socioeconomic inequality. Combining anonymized mobile-phone traces with demographic, transport, and POI data links these differences to income, employment, transport access, and visitation purposes, while highlighting important data and privacy limitations.
- Data integration: Point-of-interest geographic databases can expose gender differences in visited locations and relate them to categorized spatial features.The resulting patterns can suggest hypotheses for further research and inform potential interventions.
- Mobility patterns: Women visit fewer locations than men and distribute their time among a smaller set of preferred locations, making their movements more localized.The study attributes reduced mobility partly to women’s greater concentration at their most visited locations.
- Socioeconomic disparities: Gender mobility inequality widens with income inequality: affluent municipalities have smaller mobility gaps than more deprived areas.Income, employment, and gender mobility equality are positively correlated across Santiago’s municipalities.
- Transport access: Transport access reduces mobility differences across socioeconomic segments for men but significantly less so for women.Lower income continues to constrain women’s mobility even when public transport is available, supporting more gender-inclusive transport design.
- Visited locations: Women’s and men’s mobility differs in the types of locations they visit, with women visiting more places near hospitals, malls, and taxi stops.These patterns may indicate a greater burden of caring for family members or related individuals.
- Limitations: Mobile-phone mobility analysis is constrained by operator-dependent sample representativeness, gender- and age-related calling differences, unavailable age confounding, phone-ownership bias, and privacy concerns.The study controlled for some biases but could not address all potential confounding effects; it also adopted privacy-preserving strategies and analyzed data in aggregated form.
4 Materials and Methods
The study analyzes anonymized mobile-phone mobility traces from Santiago, combining user attributes with census, socioeconomic, transportation, and point-of-interest data. Mobility is aggregated spatially and compared across gender and socioeconomic groups while applying privacy safeguards.
- CDR data: The dataset covers three months of anonymized Call Detail Records from Santiago, enriched with gender, socioeconomic segment, and registered phone-line information.The study period was May–July 2016.
- CDR data: Filtering produced 418,624 users with identifiable homes, more than two visited locations, and at least 91 calls during the study period.The socioeconomic segment was known for 315,844 users.
- User attributes: Gender was recorded as female or male from subscription information, and females represented 51% of the user sample.The operator supplied the binary gender value based on information provided when users subscribed.
- Socioeconomic data: Socioeconomic groups were assigned by the phone carrier and consolidated into upper and lower segments using a reference household income I*.The lower segment comprises C3, D, and E; the upper segment comprises AB, C1a, C1b, and C2.
- Privacy protection: The researchers analyzed anonymized, spatially aggregated outputs within the operator’s systems and did not attempt to identify individuals or link records to third-party data.Phone numbers were hashed with SHA-3, and reports from towers with fewer than three unique phone numbers were excluded.
- Contextual datasets: The analysis incorporated census features, public-transport stops, and OpenStreetMap amenity and other point-of-interest data for Santiago’s comunas.Public-transport access was identified from GTFS stop coordinates, while POIs were collected from OpenStreetMap and additional sources.
Supplementary Material
Supplementary analyses assess sample representativeness, mobility distributions, calling activity, socioeconomic composition, municipal mobility gaps, and sensitivity across spatial and contextual specifications.
- Representativeness: CDR-based population, gender, and socioeconomic ratios correlate with census or survey baselines at 0.93, 0.79, and 0.88, respectively.These comparisons support the representativeness checks reported for the CDR sample.
- Mobility distributions: Women and men show similar ranges of radius of gyration, but women tend to have a smaller radius of gyration.The distribution is disaggregated by gender.
- Calling activity: Users’ average calling activity is computed hourly over the three-month study period and normalized separately for males and females.The supplementary distribution describes calls made or received during the day.
- Socioeconomic composition: Both genders are generally equally represented across the socioeconomic classes ABC1, C2, C3, D, and E.Users without an assigned socioeconomic class are labeled ND.
- Municipal comparisons: Municipal mobility gaps are plotted against the GSE ratio using entropy ratios and women-to-men ratios of locations covering at least 80% of activity.Point size represents comuna population.
- Visit-pattern sensitivity: Gender visit-pattern ratios are examined across kernel bandwidths, with the spatial-resolution bandwidth marked and taxi, hospital, and mall highlighted as the most imbalanced POI types at log(d) = −2.5.Additional figures test sensitivity to home location, distance from home, work location, and collective-taxi locations.
Text A: Controlling for differences in call activity by gender
The analysis tests whether gender differences in mobility could reflect unequal calling activity by controlling or resampling users’ call histories. Women continue to show lower entropy under these adjustments.
- Controls: The analysis controls correlations for call activity and population size when relating mobility differences to sociodemographic variables.It also evaluates sensitivity to alternative sampling of users’ call activity distributions.
- Down-sampling: After randomly removing 25% or 50% of men’s calls, women still have lower average entropy than men.The reported estimates are NSBF = 1.7 versus NSBM = 1.87 after 25% removal, and NSBF = 1.7 versus NSBM = 1.94 after 50% removal.
- Activity threshold: Restricting the sample to users who made at least 200 calls during three months retains 72% of the total sample.This is another sensitivity analysis for unequal calling activity.
- Equalized activity: With equalized activity from 200 randomly selected calls per user, women’s entropy remains lower: NSBF = 1.73 versus NSBM = 2.01.The gender difference remains statistically significant according to a Kruskal–Wallis test with p < 0.001.
Text B: Temporal stability of the entropy
A ten-fold temporal stability analysis evaluates whether users’ entropy measurements depend strongly on the selected time period. The resulting correlation indicates that entropy is stable over time.
- Stability procedure: The study divides the observation window into k = 10 folds and compares entropy averaged across nine folds with entropy measured on the held-out fold.This k-fold procedure is applied separately for each user.
- Stability result: The average Pearson correlation between the two entropy vectors is r = 0.81 ± 0.01, indicating temporal stability of the entropy metric.The comparison is reported across all users.
Text C: Sensitivity to the spatial resolution
The socioeconomic correlation with the gender mobility gap observed across comunas remains at the finer cell level. Cell-level analyses show negative associations between GSE ratio and both mobility metrics.
- The comuna-level correlation between socioeconomic variables and the gender mobility gap remains present at the cell level.
- r = −0.40 for the Pearson correlation between GSE ratio and R ˆ Nl at the cell level.The association is statistically significant at p < 10−25.
Text D: Robustness with respect to call activity
The gender differences in mobility persist when analyses are restricted to users with higher calling activity. Women visit fewer distinct locations, have lower mobility entropy, and show a smaller average radius of gyration than men.
- 254,586 users comprised the restricted sample requiring at least 3 calls per day on average.This stricter activity threshold was used to test sensitivity to the original user-selection criterion.
- Women display lower mobility entropy and visit fewer locations than men in the restricted sample.The overall gender differences remain observable after restricting users by calling activity.
- About 12 fewer distinct locations were visited by women than men on average (95% CI [11.57, 12.01]).
- The difference in core daily-activity locations was ∆ˆNl = 2.73 [2.68 −2.79].This metric considers distinct locations characterizing the core of users’ daily activity.
- Women’s average radius of gyration was 1.14 Km shorter than men’s (95% CI [1.10 −1.17]).
Text E: Sensitivity analysis of gender differences in visit patterns
Sensitivity analyses indicate that the reported gender differences in visit patterns are not explained by residence location, frequently visited work-related locations, or the analyzed facility categories. However, CDR spatial resolution limits claims to proximity at scales of about 1 km or more.
- CDR data and its spatial resolution allow inference only about proximity to a given POI, using POI densities over spatial scales of 1km or more.
- The study makes no claims about gendered visits to places or services that cannot be described at a finer spatial scale.
- Using each user’s inferred home tower, gender imbalances vanish for malls and hospitals and are drastically reduced for taxi stands.
- Visits near hospitals and malls remain strongly gendered for distant locations, while the taxi-POI imbalance is reduced.The hospital and mall effects grow slightly stronger when locations within 5km of inferred home are removed.
- Removing the most frequently visited non-home location, and top-k such locations, produces no significant change in the POI results.This robustness check treats the most frequently visited non-home location as a likely work location for employed individuals.
- The reported gender differences are associated with locations in the long tail of visits rather than frequently visited locations such as workplaces.
- The authors note that OpenStreetMap POI quality might influence results, although mall and hospital data were manually or officially checked.
- Spatial bias in taxi-POI reporting remains challenging to assess, while similar gender differences appear for shared-taxi collectivo POIs.