Behavioral changes during the COVID-19 pandemic decreased income diversity of urban encounters

Cities generate productivity and innovation by fostering dense social connections through physical encounters. Prior work shows that network diversity predicts economic growth and recovery and that integrated community networks build resilience to shocks. The COVID-19 pandemic and associated mobility restrictions threatened both the quantity and quality of urban encounters, compounding concerns about inequality and segregation. Large-scale mobility and digital trace data have been used to study co-location and the diversity of encounters, revealing that mobility behavior (beyond residence) explains a substantial portion of urban segregation and that people tend to visit places aligned with their socioeconomic status. Compared to traditional, residence-based segregation measures, mobility-based analyses provide a richer view of income segregation in cities. Despite many studies on short-term mobility changes during early pandemic stages and on longitudinal transportation shifts, less is known about the long-term impacts of the pandemic on the income diversity of urban encounters. This study addresses that gap by examining how the experienced income diversity of encounters changed before and during the pandemic across three years in four large U.S. cities, and by identifying behavioral drivers of any long-term changes, with implications for policy as cities move beyond the pandemic.

The paper situates its contribution within several strands of literature: (1) Urban social connections and diversity: Prior empirical research links diversity of networks with economic growth, recovery, and community resilience. (2) Mobility-based measurement of encounters and segregation: Studies using CDRs, credit card data, and social media quantify co-locations at POIs and show that mobility behavior accounts for a large share of urban segregation; people largely visit venues aligned with their SES, occasionally visiting higher-income places. (3) COVID-19 mobility and behavior: Numerous works document short-term mobility reductions during early lockdowns, socioeconomic disparities in responses, disease spread implications, and longitudinal changes (e.g., less public transport use, more work-from-home, increased online delivery). However, the long-term effects on the diversity (quality) of encounters, beyond mere quantity, remained underexplored, motivating this study.

Data and setting: The study analyzes anonymized, privacy-enhanced GPS location data from Spectus for >1 million mobile devices across four U.S. CBSAs (Boston, Dallas, Los Angeles, Seattle) over more than three years (pre- and during pandemic). A places dataset of 433,000 verified POIs was obtained from the Foursquare API, with robustness checks using ReferenceUSA Business Historical Data. Post-stratification ensured representativeness across regions and income levels. Socioeconomic status (SES) assignment: Each user’s home census block group (CBG) was inferred by Spectus from nighttime locations (10 p.m.–6 a.m.) weekly. Users were assigned to one of four SES quantiles based on the median household income of their home CBG from the 2016–2020 ACS. Robustness to the number of income bins was verified. Data filtering and visit attribution: Only users with >300 minutes observed daily were included. Stays between 10 minutes and 4 hours were extracted using Sequence Oriented Clustering and matched to the nearest POI within 100 meters. Robustness to temporal filters and the 100 m spatial threshold, and to the POI dataset choice, was confirmed. Diversity measures: For each place a, compute the proportion of total time spent by each income quantile q, T_qa. Place-level experienced income diversity D_a = 1 − Σ_q (T_qa)^2, with D_a ∈ [0,1], where 1 indicates perfectly even exposure across the four income groups. For each individual i, compute the proportion of time at place a, T_ia, derive exposure to income quantile q as T_iq = Σ_a T_ia T_qa, then individual-level diversity D_i = 1 − Σ_q (T_iq)^2. Metrics were computed on rolling 2-month windows and deseasonalized using 2019 monthly trends. Results were robust to alternative diversity metrics (e.g., entropy). Counterfactual simulations: To disentangle behavioral drivers of decreased diversity, the authors constructed hierarchical counterfactual mobility datasets using 2019 month-matched data as a baseline and removed visits to match (i) reductions in total activity time; (ii) changes in traveled distance distributions by income quantile (7 distance bins); and tested category controls. Ten simulation runs per scenario yielded highly stable estimates (very small standard errors). Differences between observed and counterfactual diversity isolated the contribution of each factor. Behavioral model fitting: The Social Exploration and Preferential Return (Social-EPR) model was fitted across periods to estimate parameters capturing exploration versus return behaviors. The social exploration parameter σ_c measures the probability that, when exploring a new place, an individual visits a venue where their income group is not the local majority. Spatial heterogeneity modeling: CBG-level mean experienced income diversity D_CBG^INC = (1/N_CBG) Σ_{i∈N_CBG} D_i and changes ΔD_CBG relative to 2019 were modeled using linear regressions with PUMA fixed effects. Covariates included: (P_CBG) place subcategory preferences (2019), (M_CBG) mobility metrics (radius of gyration, average traveled distance), and (R_CBG) residential sociodemographics (population density, median income, age/race composition, public transit use), all standardized. Adjusted R^2 and coefficient significance were reported across months and cities, with full diagnostics in Supplementary materials. Policy stringency analysis: The OxCGRT COVID-19 Stringency Index at the state level was correlated with CBSA-level decreases in experienced diversity (ΔD_CBSA), with additional checks including cases/deaths and ARIMA-type models to account for autocorrelation.

- Despite recovery of aggregate mobility metrics (visits per day, time at POIs, unique POIs) to pre-pandemic levels by late 2021, experienced income diversity of encounters at places (D_a) and by individuals (D_i) remained below pre-pandemic levels in all four cities. - Magnitude and persistence: Diversity fell by roughly 30% during the first lockdown (April 2020) and remained about 10% below 2019 levels even by late 2021. Initial sharp declines were followed by partial, not full, recovery. - Place category heterogeneity: All major POI categories in Boston showed decreased diversity short- and long-term. The largest declines were in Museums, Leisure, Transportation, and Coffee; Grocery stores showed the smallest declines and maintained visit volumes during early waves. - Behavioral drivers (counterfactual decomposition): • Reduction in total activity time explained ~50% of the diversity loss during the first wave (short term), dropping to ~2% by late 2021 once mobility recovered. • Changes in traveled distances had small negative effects on diversity; changes in dwell time across major activity categories had negligible additional effects. • Microscopic behavioral shifts—reduced social exploration and altered place subcategory preferences—accounted for the remaining loss and dominated in later stages. - Social exploration decreased: The Social-EPR model’s social exploration parameter σ_c declined relative to 2019 across all cities, indicating lower probability of exploring places where one’s income group is not the local majority, thereby reducing experienced diversity. - Shift in POI subcategory preferences: During the pandemic, hardware stores, big box stores, and grocery stores became more frequently visited; gyms, movie theaters, and American food restaurants were visited less. Within major categories (e.g., restaurants), specific subtypes shifted (e.g., more fast food/donut, fewer American sit-down). - Spatial and sociodemographic heterogeneity: • Variance explained for baseline D_CBG^INC was high (≈60–70%), consistent with prior work. • Variance explained for changes ΔD_CBG^INC was lower (max R^2≈0.31 during outbreak months; R^2≈0.11 by Oct 2021), suggesting more homogeneous long-term decreases regardless of area characteristics. • During outbreak months, greater decreases were associated with higher population density, higher share of working-age residents (25–64), higher public transit reliance, and larger movement ranges (radius of gyration). - Policy trade-off: Strong, significant negative correlations between the OxCGRT Stringency Index and ΔD_CBSA across cities (ρ in [−0.9, −0.73], p<0.01) indicate a trade-off between containment measures/outbreak intensity and income diversity of encounters. Notably, diversity remained depressed in late 2021 even at low stringency levels in several cities, implying lingering behavioral effects.

The study directly addresses whether and how the COVID-19 pandemic altered the income diversity of urban encounters. It shows that, although people resumed pre-pandemic mobility volumes by late 2021, the diversity of their in-person encounters remained depressed, indicating that the pandemic changed not only how much people move but where and with whom they co-locate. Counterfactual analyses and behavioral model fits reveal that initial declines were driven by overall activity reductions, but persistent losses stem from decreased social exploration and shifts in place-level preferences. These behavioral changes, in combination with policy stringency and outbreak dynamics, reduced inter-income mixing at urban venues. The findings underscore the importance of diverse urban encounters for social capital, resilience, and economic growth: sustained reductions in diversity could, over time, erode weak ties and increase segregation and polarization. The observed strong trade-off between public health stringency and encounter diversity highlights the need for policy designs that mitigate social fragmentation while managing disease spread, and for targeted interventions to reinvigorate cross-income interactions as cities transition beyond the pandemic.

This work provides three main contributions: (1) Empirical evidence that experienced income diversity of urban encounters in four major U.S. cities declined substantially during the pandemic and remained below pre-pandemic levels into late 2021, despite recovery in aggregate mobility. (2) Identification of behavioral mechanisms—especially reduced social exploration and shifts in place subcategory preferences—that sustained lower diversity beyond the initial activity reductions. (3) Demonstration of a strong negative association between COVID-19 policy stringency/outbreak intensity and encounter diversity. Implications include the need for urban mobility and public space policies that actively promote cross-income encounters and restore social exploration, such as fare-free transit and investments in inclusive public spaces that lower the cost and barriers for inter-neighborhood interactions. Future research should extend the framework to other demographic dimensions (e.g., race), incorporate richer semantics of encounters (purpose, interaction quality), and evaluate the causal impact of specific interventions on restoring diverse urban mixing.

- Sampling and measurement biases: Despite robustness checks and post-stratification, biases may persist due to uncertainties in mobile data collection algorithms (e.g., location sampling frequency and timing). - Encounter semantics: The data cannot distinguish the purpose or type of co-visitation (e.g., silent co-presence vs. social interaction), so diversity measures serve as proxies for meaningful encounters and bounds on experienced diversity. - Scope: The analysis focuses on income-based diversity and does not directly evaluate other socioeconomic or demographic dimensions (e.g., racial diversity), though methods could be extended using ACS data. - Data access: Mobility data are proprietary and accessible via agreements, which may limit full reproducibility at the raw data level (though aggregated-code is shared).