Neighborhood street activity and greenspace usage uniquely contribute to predicting crime

The study investigates how neighborhood physical environments—particularly greenspace—and social activity relate to urban crime. Motivated by environmental criminology and theories of routine activity and crime prevention through environmental design, the authors hypothesize that realized use of greenspace (e.g., park visits) and neighborhood street activity may be associated with reduced crime via sociological mechanisms (increased guardianship, social cohesion, “eyes on the street”) and psychological mechanisms (restored attention and self-control). Prior work on greenspace and crime has yielded mixed results, often relying on static measures of green presence (e.g., tree canopy) rather than usage. This study asks whether local street activity and exposure to urban greenspace add unique information to models predicting crime, and whether intentional greenspace contact (park visits) and incidental contact (tree canopy) have distinct associations with crime, while controlling for socioeconomic and demographic factors. Analyses are conducted in Chicago and independently replicated in New York City; museum visits are included as a comparison amenity to test whether effects are specific to parks.

Prior literature shows heterogeneous associations between environmental factors and crime. Studies have considered climate, vacant lots/buildings, lighting, and disorder. Regarding greenspace, several studies report negative associations between crime and tree canopy, vegetation, and greening interventions, though others find null or even positive associations in specific contexts (e.g., crime clustering around parks or at residential–industrial interfaces). A common limitation is reliance on static, coarse measures of physical greenery rather than how residents actually use green amenities. Sociological theories suggest greenspace can increase street activity and social cohesion, enhancing informal social control and guardianship. Psychological theories (attention restoration) link nature exposure to improved cognitive control and reduced aggression. Alternative causal hypotheses posit that higher crime suppresses greenspace use and investment (e.g., reduced tree planting due to safety concerns) and lowers street activity via fear. The paper situates its approach within this mixed evidence base by focusing on realized use (park visits) and observed street activity from mobility data.

Design: Observational, cross-sectional analyses at the census tract level in Chicago (pilot) and New York City (preregistered replication). Dependent variables: log-transformed counts of violent and non-violent crimes (violent counts offset by +1 where zero). Independent variables: percent tree canopy, percent grass coverage, park visits, distance traveled to parks, museum visits, local street activity, population (log), working population (log), median household income (log), percent unemployed, percent below poverty, percent living in crowded housing, percent foreign born, percent residential stability, percent with less than high school diploma, percent with bachelor’s degree or higher, percent Black, and percent Hispanic. All independent variables were z-scored (with log transforms where noted). Data sources: • Land cover (LiDAR, 2010, 2 ft resolution) classified into trees, grass, etc., aggregated to tract. • Cell phone mobility (Carto, May 2017): anonymized device pings with lat/long, timestamp, precision. Home tract defined as modal location between 12–6 am. Park and museum visits counted as at most one per resource per day, excluding visits within 100 m of home centroid; aggregated by residents’ home tracts as average visits per user. Street activity computed as the share of residents’ out-of-home pings occurring within a locally defined vicinity of the home tract comprising the nearest neighboring tracts containing up to 40,000 people (with correction for the boundary tract), per Saxon’s method. • Crime data (2017; also May 2017 in supplementary): city open data portals; crimes categorized as violent or non-violent; geocoded to tracts; removed records lacking precise locations. • Demographics: ACS 2012–2017 5-year estimates; working population from LEHD (2017). Units included: 792 tracts (Chicago) and 2,098 tracts (NYC). Statistical approach: Two-step modeling. (1) Adjust crime by regressing out percent Black and percent Hispanic (to statistically control for race/ethnicity as proxies of residential inequality without theoretical justification). Residuals used as dependent variables. (2) Fit hierarchical linear models with census tracts nested in neighborhoods (Community Areas in Chicago; Neighborhood Tabulation Areas in NYC). Test for spatial autocorrelation (Moran’s I); when present, diagnose via Lagrange multiplier tests using queen contiguity weights and fit spatial error models (errorsarlm) where appropriate. Four model specifications per city and crime type: (1) physical greenspace only (tree canopy, grass), (2) add park visits, (3) add street activity, (4) include both park visits and street activity; all control covariates included throughout. Model fit compared via AIC. Exploratory causal structure: Directed acyclic graphs using the FCI algorithm (pcalg in R) on total crime (log of violent + non-violent), allowing for latent confounding; report direct edges to/from crime. Software: R 3.6.3 with corrplot, lme4, pcalg, RColorBrewer, rgdal, spdep, spatialreg, tidycensus, tidyverse, tigris; Graphviz for supplementary network figures; Open Science Grid for data processing.

• In both Chicago and New York City, park visits and local street activity are each significantly and negatively associated with both non-violent and violent crime, controlling for socioeconomic and demographic covariates. Including both variables yields the best-fitting models (lowest AIC), indicating unique, non-overlapping explanatory contributions.
• Tree canopy shows a weaker association: significant negative associations in Chicago persist in full models; in New York City, tree canopy is not significant after controls.
• Quantified associations (full models): Chicago: a 5% increase in local street activity is associated with 6.9% less non-violent crime and 9.0% less violent crime; a 25% of average increase in park visits is associated with 4.9% less non-violent crime and 6.8% less violent crime; a 5% increase in tree canopy is associated with 3.3% less non-violent and violent crime. New York City: a 5% increase in local street activity is associated with 5.0% less non-violent crime and 2.7% less violent crime; a 25% of average increase in park visits is associated with 4.8% less non-violent crime and 5.7% less violent crime; tree canopy not significant.
• Museum visits do not have significant associations with crime in either city (with or without park visits in the model), suggesting the effect is specific to park use rather than generic amenity visitation.
• Exploratory FCI DAGs identify direct relationships between park visits and crime, and between street activity and crime, in both cities; no direct relationships were found between tree canopy and crime or museum visits and crime. Population shows direct relationships with crime in both cities; other direct links differ by city (e.g., poverty and foreign born in Chicago; unemployment and working population in NYC). Many edges are bidirectional, indicating potential reciprocal influence or latent confounding.

Findings support the hypothesis that both realized greenspace use (park visits) and local street activity relate to lower crime through distinct pathways. The unique contributions of park visits and street activity, together with DAG-identified direct paths, suggest multiple mechanisms: sociological (greater guardianship, social cohesion, informal control) and psychological (attention restoration improving self-control, reduced aggression). Physical green presence (tree canopy) shows weaker and less consistent associations than usage, aligning with arguments that realized access and engagement better capture greenspace’s effects. Cross-city replication strengthens generalizability of the park and street activity associations, though effect sizes are larger in Chicago, potentially reflecting differences in baseline crime, park characteristics, or urban form. Policy implications include supporting green infrastructure and programs that increase park use and foster safe, active streets as potentially cost-effective strategies with co-benefits for health and equity. The absence of associations for museum visits indicates parks have specific relevance for neighborhood safety beyond general cultural amenity usage.

Using large-scale cell phone mobility data integrated with land cover, crime, and demographic data, the study demonstrates that neighborhood street activity and park use uniquely and negatively associate with both violent and non-violent crime across two major U.S. cities. Tree canopy exhibits weaker, city-dependent effects. Exploratory causal graphs indicate direct connections from park use and street activity to crime. These results highlight multiple pathways—social and cognitive—through which greenspace and neighborhood activity may influence safety. The work underscores the value of measuring realized greenspace engagement and leveraging mobility data to study urban behavior. Future research should: (1) conduct longitudinal and intervention studies to establish causality and directionality; (2) expand to more diverse global cities to test generalizability; (3) investigate which features of parks and neighborhoods most effectively increase beneficial street activity; and (4) develop richer measures of the quality and type of social interactions underlying street activity.

• Observational, cross-sectional design limits causal inference; FCI DAG analyses identify possible direct relationships but cannot recover the true causal DAG and do not account for spatial structure. Bidirectional edges suggest potential reciprocal effects or latent confounding.
• Mobility data reflect a sample of smartphone users (underrepresentation of children and elderly; modest socioeconomic biases), though tract-level aggregates are reported to be reasonably representative.
• Street activity metric captures quantity but not quality/type of social interactions, which may differentially relate to crime.
• Potential measurement limitations: static LiDAR-based canopy (2010) predates 2017 crime/mobility data; museum and park visit counts do not capture duration or within-day repeat visits; exclusion rules (e.g., within 100 m of home) may imperfectly filter spurious visits.
• Differences across cities (urban form, park characteristics, baseline crime) may influence effect magnitudes; generalizability beyond Chicago/NYC requires further study.