A global analysis of the impact of COVID-19 stay-at-home restrictions on crime

Following the WHO’s declaration of COVID-19 as a public health emergency on 11 March 2020, governments imposed stay-at-home orders, travel bans, school and venue closures, and limits on gatherings to slow transmission. These policies drastically reduced mobility (retail/recreation trips in many countries dropped by over 80% at peak). This study investigates whether and how such stay-at-home restrictions affected urban crime globally. The authors analyze police-recorded daily incidents across 27 cities in 23 countries for six crime categories (assault, burglary, robbery, theft, vehicle theft, homicide), leveraging cross-city variation from voluntary recommendations (for example, Sweden) to highly enforced lockdowns (for example, Peru). The research is motivated by competing expectations: strain theories predict increases in crime due to stressors and social isolation, especially for domestic contexts and substance use, whereas opportunity and routine activity theories predict declines in crimes that depend on public-space interactions as movement is curtailed and guardianship rises. The goal is to assess overall impacts, heterogeneity across crime types and cities, and whether policy stringency and mobility changes explain differences in effect sizes.

Prior work noted short-term crime declines following COVID-19 restrictions in several jurisdictions, but effects varied across countries and crime types. Theoretical perspectives highlight different mechanisms: general strain and social stress could elevate motivations for certain offenses (for example, domestic violence, substance-related harms), while opportunity and routine activity theories anticipate reductions in public-space crimes (for example, theft, robbery) due to fewer suitable targets and increased guardianship. Opportunity structures are crime specific, implying that changes in theft opportunities may not mirror changes in assault. Early evidence suggested strong declines in property crimes tied to commercial areas and transit nodes, reduced residential burglary opportunities due to occupied homes, possible increases in commercial burglary, and decreases in public-place assaults with nightlife closures alongside concerns about domestic violence rising. Policing adaptations (for example, enforcing distancing, quarantine checks, border control) could also influence recorded crime patterns.

Design and data: The study uses an interrupted time series (ITS) approach treating COVID-19 stay-at-home restrictions as natural experiments. Daily police-recorded incidents were compiled for six crime categories (assault, burglary, robbery, theft, vehicle theft, homicide) from 27 cities in 23 countries. Crime category definitions referenced the International Classification of Crime for Statistical Purposes to maximize cross-site comparability. Some combined categories (for example, Seoul’s burglary with robbery; motor vehicle theft not distinguished from theft) were excluded from analyses to maintain comparability. Not all crime types were available in every city. Time series generally began in 2018 or 2019 and extended to May–September 2020. Treatment variable: A binary indicator equals 1 for periods when stay-at-home restrictions/recommendations were in effect in each locale and 0 otherwise (pre- and, where relevant, post-restriction periods). The step function estimates immediate level changes during the intervention period. Start and end dates were determined via local collaborators, the Oxford COVID-19 Government Response Tracker, and media sources. Modeling: City-level ITS models estimated Poisson generalized linear models with a logit-link function to capture level changes in daily counts after implementation. Models adjusted for seasonality (month/week/day-of-week dummies), autocorrelation (autoregressive and/or moving average terms based on ACF/PACF diagnostics and portmanteau Q tests), outliers/holidays with dummies, over-dispersion via scaling adjustment, and included average daily temperature (°C) and an annual population offset. Stationarity was checked with Dickey–Fuller tests. Very sparse daily outcomes (for example, homicide in some cities) were excluded to ensure model stability. Descriptive analyses: For each city and crime, pre/post averages were computed; 7-day moving averages were plotted and indexed to 100 on the first restriction day to visualize trajectories (mean trends across cities also plotted). Meta-analysis: Over 100 city-by-crime ITS effect sizes (incidence rate ratios, IRRs) were synthesized using random-effects meta-analyses to estimate grand mean effects for each crime category, accounting for heterogeneity (I² values). Meta-regression: Mixed-effects meta-regressions related city-by-crime effect sizes to policy stringency measures from the Oxford Tracker. The focal covariate was stay-at-home stringency (0 to 3, where 3 = do-not-leave-house with minimal exceptions). Additional models examined other containment policies (schools, workplaces, events, gatherings, public transport, internal travel), an overall stringency index, economic support, and Google mobility indices (retail/recreation, parks, residential, etc.). Given limited N per model, policy variables were entered separately. Sensitivity analyses assessed robustness to outliers (for example, Barcelona), differing crime categorizations (for example, domestic vs non-domestic assault availability), and restricting to cities with all crime categories.

- Overall effect: Across cities, crime declined by an average of 37% after stay-at-home restrictions were implemented. - Meta-analytic IRRs (Table 1; random-effects): - Assault: IRR 0.65 (95% CI 0.56–0.76), ≈35% reduction; high heterogeneity (I² ≈ 98.4%). - Burglary: IRR 0.72 (0.61–0.85), ≈28% reduction; heterogeneity high (I² ≈ 96.6%). Range from -84% (Lima) to +38% (San Francisco). - Robbery: IRR 0.54 (0.45–0.64), ≈46% reduction; no city showed a significant increase; high heterogeneity (I² ≈ 97.9%). - Theft: IRR 0.53 (0.42–0.66), ≈47% reduction; all cities with theft data showed significant declines; I² ≈ 99.2%. - Vehicle theft: IRR 0.61 (0.49–0.75), ≈39% reduction; 8/18 cities showed no significant change; I² ≈ 97.9%. - Homicide: IRR 0.86 (0.74–0.99), ≈14% reduction overall; only Lima, Cali, and Rio de Janeiro showed significant declines; heterogeneity comparatively lower (P statistic 54.6%). - Descriptive example: In Barcelona, thefts dropped from an average of 385.2/day pre-restriction to 38.1/day during restrictions. - Time dynamics: Visual trends showed sharp declines peaking about 2–5 weeks after implementation, with gradual reversion toward pre-treatment levels as restrictions eased. - Stringency matters (meta-regression, Table 2): Higher stay-at-home stringency predicted larger declines for burglary (b = -0.37, P = 0.01; exp(b)=0.69; adj. R² 34.4%), robbery (b = -0.40, P = 0.002; exp(b)=0.67; adj. R² 35.9%), theft (b = -0.33, P = 0.03; exp(b)=0.72; adj. R² 24.3%), and vehicle theft (b = -0.39, P = 0.01; exp(b)=0.67; adj. R² 28.7%). Assault showed a negative but marginal association (b = -0.25, P = 0.06), which became significant when excluding an outlier (Barcelona). Homicide was not significantly related to stringency (b = -0.26, P = 0.12). - Other policies and mobility: Reductions/closures in public transport were associated with larger declines in robbery and vehicle theft; other individual policy components and economic support were generally not significant. Greater declines in use of public spaces and increased residential stay (Google mobility) correlated with larger crime reductions (except homicide).

The analysis shows that city-wide crime levels respond quickly to abrupt changes in opportunity structures induced by stay-at-home policies. Large declines in crimes dependent on public-space convergence of offenders and targets (robbery, theft, many assaults) are consistent with opportunity and routine activity theories rather than broad shifts in offender motivation. Declines were substantial but short-lived, reversing as mobility resumed. Homicide decreased modestly overall, likely reflecting its higher share of domestic and organized-crime contexts, which are less sensitive to general mobility. Nonetheless, some high-gang cities (Cali, Lima, Rio de Janeiro) saw significant homicide drops, possibly due to non-state enforcement (for example, criminal groups imposing curfews). Burglary reductions likely reflect increased occupancy and informal guardianship of dwellings; where distinguishable, residential burglary dropped more than commercial. Variation across cities was partly explained by the strictness of stay-at-home rules; other concurrent restrictions added little explanatory power. Mobility analyses reinforced the central role of reduced public-space activity in driving declines. The study did not find evidence within these categories for displacement between measured crime types but notes the potential shift to online offending or domestic settings not fully captured by police data.

This global comparative study demonstrates that COVID-19 stay-at-home policies were associated with significant short-term reductions in urban crime, particularly for public-space offenses. Policy stringency on staying at home was a key predictor of effect size. The findings support opportunity- and routine-activity-based mechanisms over broad motivational changes in the short run. Future research should: (1) assess longer-term trajectories as policies evolve, (2) examine micro-place and neighborhood heterogeneity and enforcement variation, (3) disentangle policy components and their enforcement intensity, and (4) evaluate potential displacement to cybercrime and domestic violence using multiple data sources.

- City sample is non-random and weighted toward Europe and the Americas, limiting generalizability. - Reliance on police-recorded data introduces under-reporting and definitional/recording heterogeneity across jurisdictions, potentially exacerbated during the pandemic (changes in victim reporting and police priorities). - Inability to consistently separate residential vs commercial burglary and domestic vs non-domestic assault in many cities. - Sparse daily counts for some outcomes (for example, homicide) in certain cities necessitated exclusions and may affect stability. - Concurrent policy changes occurred broadly at similar times, making it challenging to identify specific causal components beyond stay-at-home stringency. - Potential displacement to online offenses or private/domestic contexts could not be measured comprehensively.