Unequal changes in commuting patterns across socio-economic strata in response to pandemic restrictions

Using anonymized mobile phone data from Bogotá, this study reveals that while pandemic restrictions reduced commuting across all income groups, lower-income populations rebounded faster to pre-pandemic levels than higher-income groups, a disparity driven by territorial characteristics and digital connectivity inequalities.

The Gist

Imagine the city of Bogotá as a giant, bustling beehive. Every day, thousands of bees (workers) fly out from their hives (homes) to the flower fields (offices) to gather nectar (work). This daily flight is called commuting.

Now, imagine a massive storm hits the hive: the COVID-19 pandemic. The beekeeper (the government) slams the gates shut, telling everyone to stay inside. But here's the twist: not all bees reacted the same way, and when the gates opened again, they didn't all fly back at the same speed.

This paper is a detective story that uses mobile phone data (like tracking the digital footprints of bees) to figure out exactly how different groups of people in Bogotá changed their travel habits during the storm.

Here is the story broken down into simple parts:

1. The Setup: A City Divided

Bogotá is a city with a very clear "map of wealth."

• The Rich (High SES): They live in the northern, greener parts of the city.
• The Poor (Low SES): They live in the southern, more crowded, and often more precarious areas.
• The Middle: They are somewhere in between.

The researchers wanted to see: When the storm hit, did the rich bees stay home more than the poor bees? And when the storm passed, who flew back to work first?

2. The Storm Hits (2020 Restrictions)

When the lockdown started, everyone stayed home. But the reason they stayed home was different.

• The Poor Bees: Many of them had to keep flying. They work in jobs that can't be done from home (like construction, cleaning, or selling food). Even though they wanted to stay safe, their jobs forced them to commute.
• The Rich Bees: They had a "superpower" called Remote Work. They could sit at their kitchen tables and do their jobs just as well as in the office. So, they stayed home much more easily.

The Analogy: Imagine a video game. The rich players have a "Pause Button" that lets them stop moving without losing their progress. The poor players are playing on "Hard Mode" where if they stop moving, they lose their game (their income).

3. The Aftermath (2021 Reopening)

When the city started to open up again, something interesting happened.

• The Poor Bees: They rushed back to the flower fields immediately. They couldn't wait. Their "stay-at-home" time was over the moment the restrictions lifted.
• The Rich Bees: They were slower to return. Even though the gates were open, many of them kept using their "Pause Button" (working from home) because they could.

The Result: The gap between the two groups grew. The poor people were back to their old, long commutes, while the rich people had discovered a new, more flexible way of living.

4. The "Wi-Fi" Factor

The researchers also looked at Internet Speed. They found a "Digital Divide."

• Rich Areas: Had super-fast, reliable Wi-Fi. This made working from home easy and comfortable.
• Poor Areas: Had slow, spotty internet. Trying to work from home here was like trying to run a marathon in heavy mud. It was frustrating and often impossible.

The Metaphor: Think of the internet as the fuel for the "Remote Work" car. The rich had a full tank of premium fuel, so they could drive smoothly from home. The poor had an empty tank or bad fuel, so they had to get back in their old, crowded buses to get to work.

5. The Big Lesson

The paper concludes that the pandemic didn't just pause the city; it highlighted and widened the cracks in society.

• Inequality: The rich got more flexible and comfortable. The poor got stuck with the same old struggles, just with less money and more risk.
• Infrastructure Matters: The study suggests that if we want to fix this, we need to invest in digital infrastructure (better internet for everyone). If the poor have good internet, they get the same "Pause Button" as the rich, which helps level the playing field.

Summary in One Sentence

The pandemic acted like a spotlight, showing us that while everyone was forced to stop, the rich had the tools to stay home comfortably, while the poor had to keep running, and when things opened up, the rich stayed home a bit longer because they finally had the internet speed to do so.

Technical Summary

Here is a detailed technical summary of the paper "Unequal changes in commuting patterns across socio-economic strata in response to pandemic restrictions."

1. Problem Statement

The paper investigates how exogenous shocks, specifically the COVID-19 pandemic, disrupt human mobility and exacerbate existing socio-economic inequalities. While previous studies have established that pandemics alter mobility, there is a lack of high-resolution empirical evidence regarding heterogeneous responses across socio-economic strata in developing regions (Global South).

• The Gap: Most mobility research relies on surveys (prone to bias) or data from developed nations. There is limited understanding of how digital infrastructure, spatial segregation, and income levels interact to shape commuting behavior during crises in Latin American cities.
• The Context: Bogotá, Colombia, serves as the case study. It is characterized by high urbanization, pronounced spatial segregation, a unique socio-economic stratification system (strata 1–6), and significant reliance on public transport.

2. Methodology

Data Sources

The study utilizes a multi-source dataset covering March, April, and May of 2019 (pre-pandemic), 2020 (strict restrictions), and 2021 (partial reopening):

1. Mobile Network Operator (MNO) Data: Anonymized eXtended Data Records (XDRs) from a major Colombian operator (approx. 25% market share).
   • Volume: ~7.6 billion connection events across 3 years.
   • Resolution: Network-level connection events (tower IDs, timestamps, user gender) rather than individual GPS trajectories.
2. Socio-Economic Data: Colombian government cadastral data classifying residential blocks into 6 strata (mapped to Low, Medium, High income).
3. Digital Infrastructure Data: Geolocated Speedtest® measurements from Ookla® to proxy broadband quality (download speeds).
4. Administrative Data: Bogotá's Local Planning Units (UPLs) and demographic variables (population density, gender ratio).

Data Processing & Feature Engineering

• Home/Work Inference:
  • Home: Inferred from the cellular tower with the highest connection frequency between 22:00 and 06:00.
  • Work: Inferred from weekday connections during business hours (09:00–17:00, with robustness checks for 07:00 and 08:00 start times).
  • Filtering: Users must connect to both home and work towers on multiple days to be included.
• Spatial Aggregation:
  • Data is aggregated into 56 geographic regions defined by the intersection of Administrative Units (UPLs) and Socio-Economic Strata (SES).
  • This creates "sub-UPL zones" with homogeneous SES characteristics.
• Key Metric: Non-Commuting Fraction ($p_{h,w,y}$):
  • Defined as the fraction of users associated with a home-work pair who did not commute on a specific day.
  • $p = 0$ implies full commuting; $p = 1$ implies everyone stayed home.
  • Changes are measured relative to the 2019 baseline.

Analytical Framework

• Descriptive Analysis: Comparison of commuting reductions and recovery rates across residential and workplace SES groups.
• Segregation Index: A metric ($S_\alpha$) adapted from [54] to measure the degree of socio-economic mixing at workplaces (0 = perfect mixing, 1 = complete segregation).
• Multivariate Regression: An Ordinary Least Squares (OLS) model with bootstrapped confidence intervals (500 samples) to identify predictors of commuting changes.
  • Dependent Variable: Change in non-commuting fraction relative to baseline.
  • Independent Variables: Period (2021 vs. 2020), Residential/Workplace SES, Urban/Rural vocation, Population, Gender Ratio, Commuting Distance, Absolute Home Internet Speed, and Log-ratio of Home-to-Work Speed.
• Robustness Checks:
  • Alternative working hour definitions (07:00, 08:00, 09:00 start times).
  • Machine Learning validation using Random Forest and AdaBoost regressors with SHAP values and permutation importance.

3. Key Results

A. Heterogeneous Commuting Responses

• Restriction Period (2020): Commuting declined by ~30% across all residential SES groups. However, workplace SES revealed a sharp gradient: high-SES work areas saw a ~40% reduction in worker presence, while low-SES work areas saw only a ~10% reduction. This suggests low-income workers (often essential workers) were less able to stop commuting.
• Recovery Period (2021):
  • Low-SES Groups: Rebounded quickly to pre-pandemic commuting levels.
  • High-SES Groups: Maintained higher "stay-at-home" behavior (remote work) even after restrictions eased.
  • Commuting Distance: High-SES individuals significantly reduced their commuting distances, whereas low-income workers maintained longer commutes, indicating less flexibility in job location or transport options.

B. Spatial Segregation Dynamics

• Pre-Pandemic: High-SES workplaces were less segregated (mixed workforce) than low-SES workplaces.
• Pandemic Effect: Segregation increased across all categories, but the sharpest rise occurred in high-SES workplaces. This indicates that during the pandemic, high-SES areas became more socio-economically homogeneous as low-income workers were unable to access these remote-work-friendly environments.

C. Role of Digital Infrastructure

• Digital Divide: High-SES areas had median download speeds >50 Mbps, while low-SES areas had 10–20 Mbps.
• Regression Findings:
  • Absolute home internet speed was not a significant predictor of commuting changes.
  • Relative Connectivity (Log-ratio of Home/Work speed) was highly significant.
  • Interpretation: When home connectivity is superior to or equal to workplace connectivity, commuting decreases (facilitating remote work). Conversely, if workplace connectivity is superior, commuting increases. This highlights that the relative quality of digital access, not just absolute speed, drives mobility decisions.

D. Predictors of Mobility Change

The multivariate regression (Median $R^2 \approx 0.30$) identified the following key drivers:

1. Period: The strongest predictor; 2021 showed significantly higher commuting propensity than 2020.
2. SES: Both residential and workplace SES significantly shape behavior. Low SES correlates with smaller reductions in commuting (i.e., more people commuting).
3. Commuting Distance: Longer distances correlate with larger reductions in commuting (likely due to the high cost/time of long commutes).
4. Digital Infrastructure: The home-to-work speed differential is a critical factor.
5. Urban Vocation: Urban workplaces saw greater reductions than non-urban ones.

4. Key Contributions

1. Empirical Evidence from the Global South: Provides high-resolution, mobile-phone-based evidence of pandemic mobility dynamics in a major Latin American city, filling a gap left by studies focused on the Global North.
2. Disentangling Residential vs. Workplace Effects: Demonstrates that the nature of the employment location (SES of the workplace) is a fundamental driver of mobility variation, often outweighing individual residential characteristics.
3. Digital Infrastructure as a Mobility Determinant: Establishes that disparities in digital connectivity (specifically the relative speed between home and work) are a significant predictor of commuting behavior, suggesting that digital infrastructure is a critical component of urban mobility resilience.
4. Methodological Rigor: Combines large-scale mobile data with administrative and speedtest data, utilizing robust statistical methods (bootstrapping, ML validation) to handle aggregation biases and infer causality proxies.

5. Significance and Policy Implications

• Inequality Amplification: The pandemic did not just disrupt mobility; it amplified pre-existing inequalities. Low-income workers were forced to continue commuting (exposing them to health risks) and could not adopt remote work, while high-income groups utilized digital infrastructure to isolate.
• Infrastructure Investment: The findings suggest that investing in digital infrastructure (broadband) in low-income and peripheral areas is not just a social good but a mobility policy tool. Improving home connectivity relative to work could enable remote work options for lower-income groups, reducing long commutes and improving well-being.
• Urban Planning: The results support "proximity-based" urban planning (e.g., the "15-minute city" concept) to reduce the necessity of long commutes, which disproportionately affect lower-income populations.
• Data for Good: The study validates the utility of private sector data (MNO and Ookla) for public policy analysis, provided privacy is maintained through aggregation.

Limitations

• Data Bias: Mobile data excludes non-phone users (potentially the most vulnerable).
• Inference Limits: Cannot distinguish between remote work, unemployment, or leave for those staying home.
• Causality: Regression identifies associations, not causal mechanisms (e.g., industry sector composition is a confounding variable).
• Generalizability: Findings are specific to Bogotá's spatial and policy context.