Infectiousness of places — Impact of multiscale human activity places in the transmission of COVID-19

The study investigates how human activity places at multiple scales contribute to the transmission of COVID-19. Motivated by concerns that dense and large settlements and specific indoor establishments may heighten infection risks, the authors aim to quantify transmission risks associated with macro-scale settlement characteristics (population size and density) and micro-scale establishments (11 types), and to examine their interactions. Using the COVID-19 pandemic as a natural experiment, the purpose is to generate causal evidence that can inform resilience planning and public health interventions while challenging the common belief that larger, denser cities are inherently riskier for disease spread.

Prior work has assessed epidemiological risks of places via mechanistic models, but these approaches are limited by difficult-to-calibrate interaction strengths. A growing empirical literature has used cross-country and cross-regional data to estimate impacts of non-pharmaceutical interventions (NPIs) on COVID-19 outcomes, often at a coarse level (e.g., stay-at-home orders, school closures) and sometimes finding strong effects for interventions like school closures and mobility restrictions. Urban scaling literature links city size and density to social phenomena, with some studies positing higher infection spread in larger cities. However, the granularity of establishment types and multiscale interactions (micro establishments vs. macro settlement form) have been less explored. This study advances the literature by (1) disaggregating micro-scale establishments into 11 types, (2) estimating causal effects using a DiD framework that helps net out concurrent behavioral changes, and (3) jointly analyzing macro-scale settlement characteristics and their interactions with micro-scale risks across four diverse countries.

Design: A difference-in-differences (DiD) approach implemented via two-way fixed effects models is used to estimate causal effects of establishment closures on transmission, controlling for other interventions and unit/day fixed effects. Analyses are conducted separately for Japan, the United Kingdom (UK), the United States (US), and Brazil during the first wave (March–August 2020). Data and Units: 906 spatial units: 45 prefectures (Japan), 234 local authority districts (UK), 308 metropolitan statistical areas (US), and 319 municipalities (Brazil). Units were selected based on data availability and population >100,000. Data include: daily infection cases (to estimate instantaneous reproduction number, R_t), government intervention timelines (11 establishment types and other NPIs: stay-at-home orders and gathering bans), and socioeconomic and spatial characteristics (population size, density, age structure, race composition where available, income, GDP per capita). Exposure/Interventions: Status of 11 establishment types—schools, childcare, offices, non-essential retail, restaurants, bars, entertainment venues, cultural venues, religious venues, indoor sports venues, outdoor sports grounds—coded as open/partial/closed (0/0.5/1). Additional controls include stay-at-home orders and gathering bans (indoor/outdoor, small/large). Outcome: log(R_t) per unit-day. R_t is estimated from infection data; observations with high uncertainty (coefficient of variation >0.3) are excluded in some diagnostics. Model for micro-scale effects: log(R_{c,i,t}) = β_c X_{c,i,t} + γ_c Z_{c,i,t} + α_{c,i} + τ_{c,t} + ε_{c,i,t}, with unit and day fixed effects. Robust standard errors clustered at the unit level. Highly collinear establishment statuses (Kendall’s tau >0.95) may be combined to avoid multicollinearity. Parallel-trend assumptions are checked via event-study designs. Sensitivity analyses include removing units and varying covariate sets. Joint impacts: For any set of establishments P, joint effects are computed by summing the relevant coefficients and deriving standard errors using robust SEs and covariance terms. Macro-scale analysis: The unit fixed effects α̂_{c,i} from the DiD model (interpretable as intrinsic transmission speed absent interventions/behavioral changes) are regressed on settlement characteristics: population size, density, age structure (elderly proportion), race proportions (Black, Asian; UK and US only), income, and GDP per capita using linear regression. Outliers are excluded to ensure normal residuals. Variance inflation factors are checked (all <7). Interactions across scales: Maximum joint effects of establishment closures are compared between high vs. low population and density subgroups (split at country medians; alternative thresholds between 1st–3rd quartiles tested for robustness). This assesses whether the share of infections attributable to establishments varies by settlement size/density.

Micro-scale (establishments): Percentage reduction in R_t attributable to closing each establishment type (interpreted as share of infections linked to those establishments) varies by country. Significant reductions (95% CI) include: - Japan: Entertainment venues 53% (4% to 77%). - United Kingdom: Restaurants and cultural venues combined with indoor gathering ban 25% (5% to 41%); Indoor sports venues 43% (13% to 63%). - United States: Entertainment venues 17% (1% to 31%). - Brazil: Non-essential retail 20% (9% to 31%); Indoor sports venues 36% (27% to 43%). Across countries, closures of essential places (schools, childcare, offices) generally showed no statistically significant effect on reducing R_t. Joint impacts: The largest reductions in R_t are achieved by closing about 2–6 establishment types; additional closures beyond this range yield diminishing returns. Overall, closing the examined establishments can reduce R_t by roughly 27% to 75% across the four countries. Macro-scale (settlements): Larger settlement population size is associated with lower intrinsic transmission speed (unit fixed effects) in three of four countries, corresponding to roughly 2.0% (1.1%–3.2%) to 4.9% (2.3%–7.5%) reduction in R_t per one million increase in population. Density effects are mixed and not positively significant in any country; Brazil shows a negative density association. Cross-scale interactions: The proportion of infections attributable to the 11 establishment types is consistently higher in relatively small settlements than in large ones across all four countries (disparity ~3%–18%). Similar but less consistent patterns are found for low-density vs. high-density settlements (clearer in Japan and Brazil). This implies that in larger settlements, a greater share of infections likely occurs in general public spaces (e.g., streets, transit) rather than within the specific establishments analyzed.

The findings address the research question by quantifying how micro-scale establishments and macro-scale settlement traits shape COVID-19 transmission. Contrary to common beliefs, larger and denser settlements are not associated with higher intrinsic transmission; larger cities may benefit from better health infrastructure or more cautious behavior. At the micro-scale, non-essential establishments (sports, entertainment, restaurants) are generally more infectious than essential ones, though magnitudes vary by country, potentially reflecting behavioral, enforcement, and contextual differences. Policy relevance includes targeting high-risk non-essential establishments for closures to achieve substantial R_t reductions with fewer social costs, while recognizing diminishing returns beyond 2–6 establishment types. In larger cities, where a smaller fraction of infections is attributed to these establishments, strategies beyond closures—such as enhanced contact tracing, mask policies, ventilation improvements, and sanitation of public spaces—may be more impactful. The results contribute to a generalizable framework linking place types to transmission risk while highlighting cross-country heterogeneity and the need to consider settlement context.

This work develops a multiscale, data-driven understanding of the infectiousness of places by combining DiD-based causal estimates for 11 establishment types with analyses of settlement size, density, and cross-scale interactions across four countries. Key contributions include evidence that (1) specific non-essential establishments often account for substantial transmission, (2) larger settlement size correlates with lower intrinsic transmission, and (3) the share of infections attributable to analyzed establishments is smaller in larger settlements, implying more transmission in general public spaces. Future research should: (a) elucidate mechanisms behind size/density effects (infrastructure, behavior, demographics, partisanship); (b) extend analyses to later waves and more countries; (c) assess generalizability to other pathogens and modes of transmission; (d) incorporate non-linear and interdependent intervention effects; and (e) integrate mobility, contact network, and venue-specific compliance data to refine establishment risk estimates.

- Data scope: Four countries and the first pandemic wave (March–August 2020) limit generalizability; interventions were implemented close in time, yielding wider confidence intervals. - Enforcement and compliance: Particularly in the US and Brazil, loose enforcement and non-compliance may bias effect sizes downward or even yield counterintuitive signs. - Variation in intervention timing: Limited cross-city variation (e.g., UK retail closures) can make estimates sensitive to a few units. - Identification assumptions: While parallel trends are tested, exogeneity may be violated by unobserved, unit-specific, time-varying confounders (e.g., localized outbreaks that trigger both behavior changes and policy responses). - Model specification: Potential “table 2 fallacy” when estimating multiple closures in one model; coefficients may reflect direct rather than total causal effects if policies are interrelated. - Linearity assumption: Assumes additive, linear effects on log(R_t); real-world interactions among establishments and sequence effects may introduce non-linearities. - Pathogen specificity: Findings derived from SARS-CoV-2 may not generalize to diseases with different transmission modes and susceptible populations. - Spatial unit differences: Japan uses prefecture-level units due to data constraints; although robustness checks suggest minimal impact, unit heterogeneity could affect comparability.