A prevailing assumption in urban studies is the 'cosmopolitan mixing hypothesis,' which posits that large, diverse cities foster socioeconomic interaction due to increased population density and diverse spaces. This hypothesis, however, contrasts with the observation that substantial socioeconomic segregation persists in many large cities. Existing measures of segregation often rely on static residential data, failing to capture real-world interactions in various locations. This study aims to address this gap by developing a new measure of exposure segregation that uses mobile phone mobility data to quantify the socioeconomic diversity of everyday encounters. The research seeks to empirically test the cosmopolitan mixing hypothesis and explore the underlying mechanisms driving socioeconomic segregation in different urban environments. The importance of this study lies in its potential to inform urban planning and policy interventions aimed at promoting social integration and reducing inequality.

The research draws upon existing literature on economic segregation, highlighting its significant costs, including reduced economic mobility, increased health problems, and political polarization. While various reforms are aimed at reducing segregation, the cosmopolitan mixing hypothesis suggests that urbanization itself can be a powerful mitigating factor. However, the study notes that the counter-hypothesis, suggesting that large cities facilitate self-segregation due to increased choice of like-minded communities, has not been rigorously tested due to methodological limitations. Previous studies using GPS data have made strides but have faced challenges in accurately ascertaining individual socioeconomic status (SES) and capturing the dynamic nature of real-world encounters. This paper builds on these previous efforts, leveraging improved methodologies and larger datasets to address these limitations.

This study utilizes de-identified GPS location pings from mobile phone data to construct a dynamic network of individual exposures. The researchers infer individual SES based on the monthly rent value of their inferred home location (determined from nighttime location pings). A dynamic network is then created, representing 1.6 billion exposures among 9.6 million individuals. Each edge in this network represents an exposure event—two individuals being in close proximity (within 50 meters) for a short period (5 minutes). Exposure segregation is then measured as the correlation between an individual's SES and the average SES of those they encounter. This approach contrasts with traditional static measures of segregation, which rely on residential data and thus fail to reflect real-world interactions. The study uses linear mixed-effects models to estimate exposure segregation, accounting for potential bias from sparse data. The researchers analyze exposure segregation at both the metropolitan statistical area (MSA) and county levels, comparing large and small MSAs to test the cosmopolitan mixing hypothesis. The analysis further explores factors influencing exposure segregation, including tie strength (frequency of encounters), location type (e.g., restaurants, parks), and the spatial distribution of city hubs.

The study's primary finding is that exposure segregation is significantly higher in large MSAs compared to smaller ones, contradicting the cosmopolitan mixing hypothesis. The ten largest MSAs exhibit 67% higher exposure segregation than small MSAs. This relationship is robust to controlling for confounding factors and holds for both population size and density. The analysis reveals that exposure segregation is lower than conventional static measures because encounters outside individuals' home tracts introduce more socioeconomic diversity. However, even the component of segregation within home tracts is higher in larger cities. Furthermore, stronger ties between individuals are associated with higher segregation. Location type also significantly impacts segregation; places like religious organizations exhibit higher segregation than stadiums. This variation is explained by the socioeconomic differentiation of spaces. Large MSAs offer a greater variety of leisure options, and these choices are more socioeconomically differentiated, leading to higher segregation within specific locations like restaurants. The study introduces a 'bridging index' to measure the extent to which city hubs (e.g., shopping centers) bridge socioeconomically diverse neighborhoods. A high bridging index is strongly associated with lower exposure segregation. This suggests that urban design can play a crucial role in mitigating socioeconomic segregation.

The findings directly challenge the cosmopolitan mixing hypothesis, demonstrating that larger cities, while diverse, do not necessarily promote greater socioeconomic mixing. Instead, the increased choice and differentiation of spaces in large cities allow for and even encourage self-segregation. The study's introduction of a dynamic measure of exposure segregation provides a more nuanced understanding of socioeconomic interactions in urban settings than static measures based solely on residential patterns. The strong negative correlation between the bridging index and exposure segregation highlights the importance of urban design in fostering integration. Strategically locating hubs in bridging positions between diverse neighborhoods can significantly reduce segregation. This has implications for urban planning policies aimed at promoting social equity and reducing inequality. Future research should explore the mechanisms by which specific urban design features affect social interaction and further investigate how to optimize urban spaces to maximize social integration.

This research demonstrates that large cities, contrary to popular belief, exhibit higher exposure segregation than smaller cities. The availability of differentiated spaces catering to specific socioeconomic groups contributes to this increased segregation. However, the study also identifies the potential of urban design to mitigate this effect through the strategic placement of bridging hubs. Future research could focus on developing more refined measures of exposure segregation and investigating the long-term social and economic implications of exposure segregation in different urban contexts.

The study acknowledges limitations in its methodology. The use of physical proximity as a proxy for social interaction may not fully capture the nuances of social connections. While the core findings are robust under various thresholds, future research might benefit from using richer data sources to better capture social ties. Additionally, the representation of certain subpopulations (e.g., the homeless) may be uneven, potentially affecting the generalizability of the results. Finally, the use of housing consumption as a proxy for SES may not fully capture the complexity of socioeconomic status.