TL;DR
This paper introduces a novel way to measure racial segregation by analyzing smartphone GPS data to capture individuals' actual daily exposures, rather than just their residential neighborhoods. They find that experienced segregation is significantly lower than traditional measures suggest, particularly in denser, wealthier cities, and that time spent outside the home is a major driver of this difference.
What Problem Does This Paper Address?
Traditional measures of racial segregation primarily focus on where people live, using indices like residential isolation. While useful, these measures have limitations: they assume uniform exposure within a neighborhood and ignore the diverse places people visit throughout their day (work, shopping, recreation). This can lead to an overestimation of segregation, as individuals might live in segregated areas but interact with diverse populations elsewhere. This paper aims to address this by developing and applying a new measure, "experienced isolation," that accounts for individuals' actual movement patterns and exposures across their daily activities.
Methods
The core innovation is the use of large-scale Global Positioning System (GPS) data from smartphones to construct a new segregation metric. Instead of relying on static residential data, the researchers track the locations of millions of smartphone users over several months. They define "experienced isolation" as the degree to which individuals from one group (defined by their home geohash7's racial composition) are exposed to individuals from another group in the places they actually visit.
Key methodological steps include:
- Geohash7 Grids: Dividing geographic space into small, consistent grid cells (approximately 500 ft x 500 ft) for precise location tracking.
- Activity Space Definition: Inferring an individual's "activity space" by aggregating all the geohash7s they visit over time.
- Exposure Calculation: Estimating the proportion of individuals from a "White" group (WDs) encountered in any given location visited by an individual.
- Leave-One-Out Estimation: A statistical technique to mitigate small-sample bias when estimating exposure in specific locations by excluding the individual's own pings from their exposure calculation.
- Weighting: Adjusting the sample to better represent the US population by weighting devices based on their home geohash7's demographic composition relative to census data.
- Decomposition: Analyzing experienced isolation by time of day and by type of location (e.g., home, outdoor spaces, retail, restaurants) to understand where and when segregation is most or least pronounced.
Datasets
- Global Positioning System (GPS) Data: Anonymized location pings from approximately 5% of the US population, collected from smartphones between January and April 2017. This dataset includes device identifiers, timestamps, and geohash7 locations, as well as inferred home locations.
- US Census Data (2010): Used to impute demographic information (race) to geohash7s and to weight the GPS sample to be more representative of the US population.
- InfoUSA and OpenStreetMap (OSM): Used to identify and geolocate various establishments and features of interest (restaurants, parks, schools, etc.) to categorize visited locations.
- Chetty et al. (2020) Income Mobility Data: Used to correlate experienced segregation with economic mobility at the Metropolitan Statistical Area (MSA) level.
Key Insights and Ideas
- "Experienced" vs. "Residential" Segregation: The fundamental idea is that where you live doesn't fully capture where you go and who you encounter. This distinction is crucial for understanding the lived experience of segregation.
- The Power of the Activity Space: Individuals spend significant time outside their home neighborhoods, and these "away-from-home" locations are key drivers of reduced experienced segregation.
- Commercial Spaces as Integration Hubs: Places like restaurants, bars, and retail stores are particularly effective at breaking down segregation, offering much higher exposure to diverse groups than residential areas.
- Urban Density and Diversity: Denser, wealthier, and more educated cities with better public transit tend to have lower experienced segregation, suggesting that urban infrastructure and amenities can foster greater intergroup contact.
- Income Mobility Link: Experienced segregation is negatively correlated with income mobility, hinting at a potential relationship between diverse interactions and opportunities for upward mobility.
Key Results
- Lower Experienced Isolation: The average experienced isolation across MSAs is 0.46, significantly lower than the average residential isolation of 0.61. This means individuals experience substantially less isolation than residential measures would suggest.
- High Correlation, Different Levels: Experienced and residential isolation are highly correlated across MSAs (Pearson correlation of 0.86), meaning cities that are residentially segregated also tend to have higher experienced segregation. However, the level of experienced segregation is consistently lower.
- Time of Day Matters: Experienced isolation is lowest during the middle of the day and highest during the morning and evening, aligning with typical daily routines.
- Location Matters: Experienced isolation is much lower in locations outside of home tracts (average 0.21) compared to within home tracts (average 0.63), which is even higher than residential isolation. This highlights the integrating effect of time spent away from home.
- Commercial vs. Residential: Experienced isolation in outdoor spaces, retail, and restaurants/bars is significantly lower (ranging from 43.5% to 50.3% of baseline isolation) than in home neighborhoods.
Limitations and Future Work
- Geographic vs. Sociological Segregation: The study measures geographic segregation (co-presence in the same space) rather than sociological segregation (actual interaction). The authors acknowledge that time spent in the same location doesn't guarantee interaction.
- Proxy for Race: The study uses home geohash7 demographics as a proxy for individual race, which is a simplification. While they show robustness to an alternative imputation strategy, it's not a direct measure of individual race.
- Sample Representativeness: While efforts were made to weight the sample, it's noted that the GPS data may slightly oversample more educated and wealthier areas.
- Data Sharing: The raw device-level location data cannot be shared due to privacy concerns and data-sharing agreements.
Future work could explore:
- Directly measuring actual interactions rather than just co-presence.
- Refining race imputation methods or using alternative data sources for individual-level race.
- Investigating the causal mechanisms behind the observed correlations between experienced segregation, urban characteristics, and income mobility.
- Decomposing segregation by specific types of interactions within locations.
Summary updated