From crisis to new routine: shifts in urban shopping mobility and socioeconomic inequality during and after the COVID-19 shock

When the COVID-19 pandemic swept through New York City in 2020, it did more than empty the subways and shutter offices. It quietly rewired the way millions of residents chose where to shop, according to a new study that tracked four years of urban mobility data across the city. The research, published in Frontiers in Big Data, shows that New Yorkers’ revealed store-selection patterns changed dramatically during the pandemic shock and, perhaps most strikingly, had not snapped back to their pre-pandemic baseline even by 2021, when vaccinations and reopenings were well underway. The crisis, in other words, did not just disrupt a routine. It seeded a new one.

The study, led by Yilun Xu of Harvard University’s Laboratory for Innovation Science together with Mohsen Bahrami and Alex Pentland of MIT, focused on visits to general merchandise and department stores, a retail category that includes department stores, warehouse clubs, supercenters, and dollar-type chains. The team pooled 282 stores across New York City and analyzed anonymized smartphone location data from more than 5,500 census block groups, the small statistical neighborhoods of roughly 600 to 3,000 people defined by the United States Census Bureau. By combining this mobility panel with detailed place data and census demographics, the researchers could reconstruct, block group by block group, how residents weighed different store attributes when deciding where to spend their shopping trips.

The analytical engine behind the study is a modernized version of the Huff gravity model, a probabilistic framework first proposed in 1964 that estimates the likelihood a customer will choose one store over another based on the store’s attractiveness and its distance from the customer. The researchers extended the model using the multiplicative competitive interaction formulation, which allows multiple attractiveness factors to be considered simultaneously, each with its own calibrated exponent. Six factors made the cut: the Haversine distance between a customer’s home block group and the store, the store’s floor area, the count and Shannon-entropy diversity of points of interest surrounding the store, the socioeconomic similarity between the customer’s neighborhood and the store’s neighborhood, and the customer’s prior revealed loyalty to the store’s chain.

Calibrating such a model on massive mobility data is a formidable optimization challenge, and the team turned to metaheuristics to solve it. They used particle swarm optimization, an algorithm inspired by the flocking behavior of birds, to fit the six store-selection weights for each block group in each year, and then wrapped that procedure inside a simulated annealing algorithm to tune the normalization interval and search-space boundaries. The normalization step, which mapped every attractiveness factor into a common interval, was critical for eliminating zero observations in the multiplicative model and for making the calibrated exponents comparable across factors as different as square footage and fractional loyalty measures. The fitted model reproduced observed visit distributions with chain-level correlations of 0.964 and 0.956 in the pre-pandemic years 2018 and 2019, confirming that the approach captured shopping behavior faithfully in stable times.

Then came 2020, and the model revealed a shock. Kolmogorov-Smirnov tests comparing the distributions of calibrated weights across consecutive years found no statistically significant changes between 2018 and 2019, but all six parameter distributions shifted significantly with the pandemic’s onset. The revealed weight associated with store area surged by roughly 79 percent from 2019 to 2020, suggesting that New Yorkers gravitated toward larger stores that could accommodate social distancing under capacity constraints. Chain loyalty jumped by about 44 percent, indicating that residents made fewer exploratory visits and concentrated their shopping within previously selected chains, a pattern the authors attribute to heightened search costs under elevated health risk. The revealed weight on the count and diversity of nearby points of interest also increased, consistent with residents trying to minimize the number of trips by bundling multiple errands into multipurpose visits.

The most counterintuitive finding concerned distance. The revealed weight associated with customer-store distance fell by about 23 percent during the pandemic shock and by nearly 35 percent from 2019 to 2021, meaning that observed shopping patterns became markedly less sensitive to how far a store was from home. This echoes earlier survey research showing that customers during the pandemic prioritized effective social distancing regulations over proximity, even accepting longer waits. In a dense city where stores cluster tightly together, the result suggests that the pandemic loosened the gravitational pull of geography itself, at least temporarily, in favor of store characteristics and familiarity.

Crucially, the pandemic-era patterns did not revert to their 2019 baseline by 2021. Although the magnitudes of change from 2020 to 2021 were smaller than the initial shock, most parameter distributions in 2021 remained more similar to their pandemic values than to their pre-pandemic ones. The authors are careful to interpret 2021 as an early recovery and reopening period rather than a fully post-pandemic equilibrium, but within the available observation window, the persistence is unmistakable. The distance effect, in particular, continued to weaken through 2021, while the emphasis on store area remained elevated or even intensified in some communities, such as those with higher education levels and median incomes, which showed a further 20 percent increase in attention to store size during the recovery year.

The study also uncovered deep socioeconomic heterogeneity in how different communities responded. Residents of block groups ranking in the top 5 percent for Asian resident proportion showed the strongest increase in chain loyalty, followed by Hispanic/Latino communities and older age groups, possibly reflecting variations in perceived search costs. African-American and Hispanic/Latino communities shifted toward stores surrounded by a diverse range of facilities during the recovery period. White communities placed greater revealed weight on the sheer number of nearby points of interest as restrictions eased. An unsupervised K-means clustering analysis, which partitioned the city into five socioeconomic subregions using all demographic attributes combined, confirmed these patterns and showed remarkable stability across random initializations, with Adjusted Rand Index values near 1.0. A difference-in-differences analysis treating the 2019-to-2020 transition as the treatment period confirmed that the pandemic was associated with significant shifts in store area, chain loyalty, and distance weights across the population and its subgroups.

The team subjected their findings to an unusually thorough battery of robustness checks. A fit-stratified sensitivity analysis showed that the main pandemic-shock and non-reversion findings persisted after excluding the block groups with the weakest model calibration, and remained directionally consistent across all quartiles of the fit distribution. A format-level analysis separating department stores from general merchandise and warehouse club formats confirmed the direction of the store area, point-of-interest, and distance effects in both categories, though it revealed that the chain-loyalty increase was disproportionately concentrated in department stores during the acute 2020 shock. A lagged specification of chain loyalty, which computed loyalty from the prior year’s visits to rule out mechanical concentration effects, retained roughly 92 percent of the population-level effect for the 2019-to-2021 comparison. Sampling-bias sensitivity analyses, spatial autocorrelation diagnostics, and estimator-stability tests across ten random seeds all reinforced the qualitative picture, though the authors caution that precise point estimates carry meaningful uncertainty.

The implications ripple outward from retail to urban planning and public policy. For retailers, the results suggest that store scale, chain familiarity, and the surrounding amenity environment may become more decisive under conditions of health risk and mobility uncertainty, while the enduring gravitational pull of proximity weakens. For urban planners, the findings highlight that crisis-induced changes in shopping mobility may have heterogeneous effects across socioeconomic communities, potentially reshaping access to retail opportunities and everyday services in ways that are not uniform across the city. And for scientists studying how societies absorb shocks, the study offers a template for quantifying not just whether behavior changes during a crisis, but whether the change calcifies into a new routine, one trip pattern, one store choice, and one neighborhood at a time.

Subject of Research: From crisis to new routine: shifts in urban shopping mobility and socioeconomic inequality during and after the COVID-19 shock

Article Title: From crisis to new routine: shifts in urban shopping mobility and socioeconomic inequality during and after the COVID-19 shock

Article References: Xu, Y., Bahrami, M., & Pentland, A. (2026). From crisis to new routine: shifts in urban shopping mobility and socioeconomic inequality during and after the COVID-19 shock. Frontiers in Big Data, 9, Article 1776172. https://doi.org/10.3389/fdata.2026.1776172

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1776172

Keywords: crisis, routine, shifts, urban, shopping, mobility, socioeconomic, inequality, during, COVID-19, shock, scientific research