Investigating neighborhood adaptability using mobility networks: a case study of the COVID-19 pandemic

Human mobility in urban areas is a complex dynamic interaction significantly influencing social and economic well-being. Studies have shown its correlation with social interactions, health, productivity, and economic prosperity. Analyzing mobility patterns through network science offers valuable insights. Previous research has utilized mobility networks to study the effects of the COVID-19 pandemic, observing topological changes due to non-pharmaceutical interventions (NPIs). This study aims to understand how various environmental and demographic factors shaped neighborhood adaptability to NPIs during the COVID-19 pandemic in New York City. Adaptability is defined as a neighborhood's ability to adjust behavior in response to NPI constraints. The study uses a dynamic network analysis approach, constructing weekly mobility networks between census block groups (CBGs) based on aggregated point-of-interest (POI) visit patterns from January 2019 to December 2020. Node and ego-network features were computed for each neighborhood to capture the dynamics of local mobility and relationships with neighboring CBGs. The dissimilarities between feature vectors of the same weeks in 2019 and 2020 were analyzed, broken down by socioeconomic groups. Combining mobility network metrics with census data and COVID-19 test results helped reveal how neighborhood characteristics predict changes in mobility networks and behavior. The study uses established network metrics and the Huff gravity model to simulate mobility flows, allowing analysis of adaptability under various hypothetical scenarios. The researchers expected to find that centrality metrics and geographic attributes significantly predict neighborhood adaptability to shelter-in-place orders, confirming prior research and revealing the importance of geographic attributes such as access to amenities.

Existing literature highlights the relationship between human mobility and various socioeconomic outcomes. Studies have established correlations between mobility patterns and social interactions, health, productivity, and regional economic resilience. Researchers have increasingly utilized network science methodologies to analyze mobility data, offering critical insights into complex interactions. Several studies have explored the impact of COVID-19 on human mobility and the effects of NPIs using mobility networks. For instance, some research used inter-county mobility networks to demonstrate topological changes resulting from NPIs aimed at curbing COVID-19 spread. However, this research extends this work by focusing specifically on the topological changes within the mobility networks during the pandemic using established network metrics, such as centrality indicators, to analyze the changing landscape of interactions and connections. This offers valuable insights into urban adaptability during public health crises. In contrast to existing research, the researchers' goal was to analyze the topological changes in mobility networks during the pandemic using established network metrics to rapidly discern and analyze the alterations in interactions and connections, providing valuable insights into urban adaptability during a public health crisis. The study also leverages the Huff gravity model to simulate mobility flows under various POI densities.

The study utilized several datasets: SafeGraph mobility and places data providing weekly aggregated visits from CBGs to POIs from January 2019 to December 2020; Google COVID-19 community mobility reports showing mobility trends across POI categories; COVID-19 case and infection rates data from Johns Hopkins University, estimated weekly cases per CBG using a weighted average based on the population of a CBG in each ZCTA and its corresponding COVID-19 case rate; and United States Census data providing demographic features at the CBG level. Mobility patterns were modeled as weighted directed networks for each week, with CBGs as nodes and visits between CBGs as edges. Ego-network-based node features (e.g., degree, clustering coefficient) were used to compute dissimilarity between paired weekly networks from 2019 and 2020, using the Canberra distance. Centrality metrics (betweenness, in-degree, out-degree, self-visit ratio) were analyzed to demonstrate temporal changes in the topological importance of CBGs. CBGs with high weekly new COVID-19 cases were defined as hotspots, and CBGs frequently interacting with hotspots were identified as bridge CBGs. The Huff gravity model was used to simulate mobility flows under various POI densities, specifically focusing on grocery stores in Staten Island to assess the impact of increased POI density on visits to hotspot CBGs. The model considers distance between CBGs and POIs and POI areas. The analysis time frame was from March 22nd to June 8th, 2020, during the first wave of the pandemic in NYC.

The CBG-level dissimilarity analysis revealed that CBGs showing the most significant shifts in mobility patterns (top dissimilarity quartile) were concentrated in Manhattan and had predominantly high-income, high-education, and largely white populations. These residents were better positioned to adapt to NPIs through remote work options. The analysis of centrality metrics showed that betweenness centrality differed significantly between socioeconomic groups. High-income CBGs had higher betweenness before the pandemic, while less affluent CBGs gained higher betweenness scores during the first wave, suggesting a shift in network connectivity. Affluent CBGs were more successful at reducing mobility compared to less affluent ones. The self-visit ratio analysis revealed higher self-visits among high-income CBGs during the first wave. Analysis of COVID-19 hotspots and bridge CBGs revealed that many bridge CBGs were in lower income and education quartiles, indicating a link between socioeconomic status and interaction with hotspots. Staten Island showed a distinct response, with residents traveling longer distances to POIs due to limited access to essential amenities. The Huff gravity model simulation in Staten Island showed that increasing grocery store density to match other boroughs significantly reduced visits to hotspot CBGs (47% and 23% reduction under Manhattan and Queens' grocery store densities, respectively).

The study's findings address the research question by demonstrating that neighborhood adaptability to COVID-19 NPIs is significantly influenced by both socioeconomic and geographic factors. The results highlight the importance of considering the spatial distribution of amenities when assessing community resilience. The less affluent and less educated neighborhoods demonstrated less adaptability to the policy interventions aimed at reducing their mobility level. The study demonstrates that communities with similar socioeconomic and demographic features may show different mobility responses based on their neighborhoods’ urban structure. The findings confirm prior research indicating that socioeconomic status plays a role in pandemic response, but they also reveal a crucial interaction with neighborhood-level geographic characteristics. The significant reduction in visits to hotspot CBGs under simulated increased POI densities underscores the potential for urban planning interventions to mitigate pandemic spread.

This study contributes to the understanding of neighborhood adaptability during pandemics by highlighting the interplay between socioeconomic and geographic factors. The results underscore the importance of considering both when designing effective interventions. Future research could extend the simulation analysis by incorporating more POI categories and physical characteristics of neighborhoods into the model. Furthermore, future studies could analyze the effects of hypothetical increases in amenities using flow graphs between larger geographical areas, such as boroughs.

The study's limitations stem from the aggregated nature of the datasets used. The inability to capture specific trip motivations and transportation modes limits the analysis of essential versus non-essential trips and their relationship to virus transmission. The POI dataset's limited coverage of workplaces could also affect the accuracy of the gravity model simulation. The utopian nature of the recommendations might require additional governmental support to incentivize private sector involvement. Despite these limitations, the study provides valuable insights into the complex relationship between mobility, demographics, and pandemic response.