Source-linked AI summary
Influence of sociodemographic characteristics on human mobility
Maxime Lenormand, Thomas Louail, Oliva G. Cantu-Ros, Miguel Picornell, Ricardo Herranz, Juan Murillo Arias, Marc Barthelemy, Maxi San Miguel, Jose J. Ramasco
TL;DR
Many studies using ICT data lack demographic information needed to describe mobility patterns. This paper analyzes geolocated credit-card transactions from Madrid and Barcelona, finding differences in travel distance and timing by gender, age, and occupation. It also notes that further data are needed to assess whether children help explain gender differences in mobility.
Problem
Many prior ICT-based studies lack information about individuals’ gender, age, or occupation, although these characteristics are relevant to mobility and infectious-disease modeling.
Source: §I · p. 1Method
The authors analyze credit-card transaction data to compare consumption and mobility patterns by gender, age, and occupation.
Source: §I · p. 2Results
Women have shorter journeys and shorter inter-event times than men, while men make longer journeys; distance measures also differ by age and activity status.
Source: C. Mobility patterns · p. 9, §IV · p. 11Takeaways & Limitations
Credit-card records can be used to assess how sociodemographic characteristics relate to mobility and spending patterns.
Source: §IV · p. 10Further data are needed to assess whether gender differences in mobility relate to social characteristics such as having children.
Source: §IV · p. 11
Abstract
from arXiv · showhide
Human mobility has been traditionally studied using surveys that deliver snapshots of population displacement patterns. The growing accessibility to ICT information from portable digital media has recently opened the possibility of exploring human behavior at high spatio-temporal resolutions. Mobile phone records, geolocated tweets, check-ins from Foursquare or geotagged photos, have contributed to this purpose at different scales, from cities to countries, in different world areas. Many previous works lacked, however, details on the individuals' attributes such as age or gender. In this work, we analyze credit-card records from Barcelona and Madrid and by examining the geolocated credit-card transactions of individuals living in the two provinces, we find that the mobility patterns vary according to gender, age and occupation. Differences in distance traveled and travel purpose are observed between younger and older people, but, curiously, either between males and females of similar age. While mobility displays some generic features, here we show that sociodemographic characteristics play a relevant role and must be taken into account for mobility and epidemiological modelization.
I. INTRODUCTION
Geolocated ICT records have enabled high-resolution studies of human behavior, but many such studies lack demographic detail. This paper analyzes credit-card transactions from Madrid and Barcelona to examine consumption and mobility by gender, age, and occupation.
I. INTRODUCTION
- Geolocated ICT data have supported research on mobility distributions [6], urban hotspots [10], land use [14] [15], and comparisons across cities and countries [20].These data register when and where people interact with or through ICT tools.Source: §I · p. 1
- Many ICT-based studies lack information about individuals’ gender, age, or occupation, although these attributes matter for urban planning, transport demand, public health, and infectious-disease modeling.The paper notes that demographic factors can influence mobility as well as disease transmission.Source: §I · p. 1
- Earlier survey-based work reported gender differences in travel purposes and activities, while studies of social-network dynamics found behavioral differences by gender and age [59].The paper extends this line of work using credit-card transactions and additional demographic attributes.Source: §I · pp. 1–2
- The study examines consumption and mobility patterns by gender, age, and occupation using more than 40 million credit-card transactions from Spain’s two most populated provinces.The dataset includes transactions from customers in Madrid and Barcelona.Source: §I · p. 2
A. Dataset description
The study uses anonymized BBVA credit-card transactions from 2011, linked to customer demographics and residence postcodes and to business categories and locations. Its geographic scope is limited to transactions by residents of Madrid and Barcelona made within their respective provinces.
A. Dataset description
- The dataset contains 40 million transactions in 2011, each with an amount, timestamp, anonymized customer ID, and anonymized business ID.Customer records link to gender, age, occupation, and residence postcode; business records link to category and terminal coordinates.Source: A. Dataset description · p. 2
- The analysis includes only credit-card payments made within Madrid or Barcelona by individuals living in that same province.The two-province scope is also shown in Figure 1.Source: A. Dataset description · p. 2
- The provinces have similar population size, area, and business counts, but Madrid has more users and transactions; users represent about 8% of its census population versus 5% in Barcelona.Table I presents basic statistics for the two provinces.Source: A. Dataset description · p. 2, Table I · p. 2
III. RESULTS
Because the Barcelona and Madrid data have similar statistical features, the paper pools them for general analyses and separates them for mobility analyses. This balances statistical power with differences in provincial geography.
III. RESULTS
- The authors aggregate the two provinces for general analyses, then analyze them separately for mobility because their geographic shapes differ.The aggregation provides higher statistical power; equivalent Barcelona mobility results appear in the appendix.Source: §III · pp. 2–3
- The paper reports equivalent mobility results for Barcelona in appendix Figures S9–S15, although only Madrid figures are displayed in that section.The authors state that the two provinces’ statistical features are similar.Source: §III · pp. 2–3
- Figure 2 organizes descriptive statistics by gender, age, and occupation, showing group proportions and median transaction and spending measures.Its rows cover individual proportions, transactions per user, annual spending per user, and average spending per transaction.Source: Figure 2 · p. 3
A. General features
The dataset overrepresents men, middle-aged people, and employed people, so the authors normalize measures by group. Spending levels and business-category shares also vary across demographic groups.
A. General features
- Men, middle-aged people, and employed people are overrepresented relative to census data, so subsequent indicators are normalized by group.Employed people comprise about 80% of the dataset, twice their proportion in Spain.Source: A. General features · p. 3
- The three demographic distributions are interdependent; for example, age composition varies by occupation and occupation composition differs between men and women.There are more female than male homemakers.Source: A. General features · p. 3
- Women have more transactions and higher total spending than men, while men spend more per transaction; transaction counts and spending decline with age as average transaction amounts rise.Employed people and homemakers have higher transaction counts and spending than unemployed people, students, and retirees.Source: A. General features · pp. 3–4
- Women spend larger fractions on fashion, food/hypermarkets, health, and wellness/beauty, while men spend larger fractions on automotive, bars/restaurants, technology, and transport.Students spend larger fractions on bars/restaurants, fashion, sports/toys, and technology than other occupation groups.Source: A. General features · p. 4, Figure 3 · p. 4
- Restricting category-spending comparisons to employed individuals yields the same conclusions as the full-sample analysis.This check addresses potential bias from differing occupational composition by gender.Source: A. General features · p. 4
- The authors also display hourly spending across weekdays, Friday, Saturday, and Sunday, alongside the demographic composition of each spending-time cluster.Figure 4 presents these temporal and cluster-composition views.Source: Figure 4 · p. 5
B. Time evolution of the amount of money spent
Spending varies across the week, with Sunday activity lower and more concentrated in daytime hours. Clustering normalized spending profiles identifies contrasting morning/weekday and evening/weekend patterns.
B. Time evolution of the amount of money spent
- Spending is higher on weekdays and Friday and Saturday than on Sunday; Sunday activity runs 10am–7pm, peaking around 4pm.Weekday, Friday, and Saturday spending spans 8am–10pm, with peaks around noon and 7pm; a small nightlife peak around 11pm is also reported for the first three days.Source: B. Time evolution of the amount of… · p. 5
- K-means clustering of normalized spending profiles identifies two opposing patterns: one-third are more active in mornings and weekdays, while the other cluster spends more after 6pm and on weekends.Profiles are normalized by each user's total spending in 2011; the cluster count is selected using the highest pseudo-F statistic.Source: B. Time evolution of the amount of… · p. 5
- The morning/weekday cluster is over-represented by women, older people, homemakers, and retired individuals.Source: B. Time evolution of the amount of… · pp. 5–6
- Figure 5 plots inter-event-time probability densities overall and by gender, age, and occupation; its insets show Tukey boxplots and black points mark averages.Source: Figure 5 · p. 6
C. Mobility patterns
Mobility was characterized through time between transactions, inter-transaction distance, and radius of gyration. These measures vary by sociodemographic group, and gender differences in journey distance persist across filtered populations.
C. Mobility patterns
- ∆t averages about 5 days (median 2 days); ∆r decays rapidly beyond 40 km, while P(rg) peaks around 6 km before rapid decay.The paper uses ∆t, ∆r, and radius of gyration to characterize mobility; ∆r depends partly on transaction frequency, while radius of gyration gives a more holistic view around each user's center of mass.Source: C. Mobility patterns · pp. 6–7, Figure 6 · p. 7, Figure 7 · p. 8
- Men have longer average and median inter-event times than women, and inter-event time increases with age; unemployed people, students, and retirees have longer intervals than employed people and homemakers.Inter-event time is negatively correlated with the number of transactions per individual.Source: C. Mobility patterns · p. 7
- Women travel shorter distances and stay closer to their center of mass than men; both measures decrease with age, while retired, homemaker, and unemployed people travel shorter distances than others.These comparisons concern inter-transaction distance and radius of gyration.Source: C. Mobility patterns · pp. 7, 9, Figure 6 · p. 7, Figure 7 · p. 8
- Across ∆t, group differences in mean ∆r persist; mean distance rises for ∆t < 6, dips at 18–30, then peaks at 30–42, especially for active people.The authors suggest the valley and peak may reflect home-to-work or school commuting.Source: Figure 8 · p. 8, C. Mobility patterns · p. 9
- Normalized KS distances exceed 1.95 across demographic filters, indicating women have shorter inter-event times while men make longer journeys; gaps are larger for middle-aged and employed people.The paper uses 1.95 as the rejection threshold at level 0.001 for its normalized KS distance.Source: C. Mobility patterns · p. 9, Figure 9 · p. 9
- Across business categories, home-to-business distances are greater for men, decrease with age, and are higher for employed people and students; Food/Hypermarkets, Health, Wellness/Beauty, and Book/CD/Stationery distances are lower.Those four categories also have more transactions by women than by men, which the authors say partially explains women's shorter shopping distances.Source: C. Mobility patterns · p. 10, Figure 10 · p. 10
IV. DISCUSSION
Credit-card records reveal sociodemographic differences in both spending and mobility. Men, younger people, and active people travel shorter distances and stay closer to their centers of mass; the gender mobility difference remains difficult to explain.
IV. DISCUSSION
- Spending fractions vary by business type: women spend more in fashion, food, health, and beauty, while men spend more in automotive, restaurants, technology, and transport.The discussion also distinguishes inactive people’s morning and weekday activity from more active people’s after-work and weekend activity.Source: §IV · p. 10
- Men, younger people, and active people travel shorter distances and stay closer to their centers of mass than women, older people, and inactive people.The discussion identifies these as three main differences between groups.Source: §IV · pp. 10–11
- The gender mobility difference is the hardest to explain, and assessing whether children relate to it would require further data.The authors note that the difference is less pronounced among older people and students, and suggest that women with children may move differently from women without children.Source: §IV · p. 11
APPENDIX
The appendix shows histograms of the joint distributions of individuals across gender, age, and occupation.
APPENDIX
- Figure S1 plots the joint distribution of individuals by gender and age.The caption identifies the plot as a histogram.Source: Figure S1 · p. 14
- Figure S2 plots the joint distribution of individuals by gender and occupation.The caption identifies the plot as a histogram.Source: Figure S2 · p. 14
- Figure S3 plots the joint distribution of individuals by age and occupation.The caption identifies the plot as a histogram.Source: Figure S3 · p. 15
P(NbT)
The appendix figures describe spending and mobility measures, presenting distributions or averages across sociodemographic groups and business categories.
P(NbT)
- Figures S4–S6 show probability densities for transaction counts, total 2011 spending, and average spending per transaction, overall and by gender, age, and occupation.Figure S4 covers transactions per individual; S5 covers money spent per individual; S6 covers average spending per transaction per individual.Source: Figure S4 · p. 16, Figure S5 · p. 17, Figure S6 · p. 18
- Figure S7 shows employed individuals’ average spending fractions by business category, gender, and age; Figure S8 plots pseudo-F against cluster count for k-means on normalized hourly spending distributions.The Figure S8 caption specifies Euclidean distance.Source: Figure S7 · p. 19, Figure S8 · p. 19
- Figures S9–S11 show distributions of inter-event time, distance between consecutive transactions, and radius of gyration, overall and by gender, age, and occupation.Their insets show Tukey boxplots, with black points indicating averages.Source: Figure S9 · p. 20, Figure S10 · p. 21, Figure S11 · p. 22
- Figures S12–S13 show average radius of gyration over elapsed time by demographic group and KS distances between men’s and women’s mobility distributions by demographic characteristics.Figure S13 covers elapsed time, travel distance, and radius of gyration distributions.Source: Figure S12 · p. 23, Figure S13 · p. 23
- Figure S14 shows average residence-to-business distance by demographics and business category, while Figure S15 shows average transaction counts by gender and business category.Figure S14 specifies distances in kilometers, computed using Haversine distance.Source: Figure S14 · p. 24, Figure S15 · p. 24