Neighborhood street activity and greenspace usage uniquely contribute to predicting crime

Urban crime is a significant societal problem, characterized by considerable heterogeneity in rates across and within cities. While economic (education, employment, poverty) and sociological (social control, collective efficacy) factors have been extensively studied, the role of the physical environment, specifically neighborhood greenspace and its usage, remains less understood. Environmental criminology suggests that how individuals engage with their physical environment influences crime. This study uses cell phone mobility data to analyze neighborhood networks and test hypotheses regarding the relationship between physical environment variables (e.g., tree canopy), sociological variables (e.g., local street activity), and crime rates. Existing research on greenspace and crime yields mixed results, with some studies showing negative associations between crime levels and various types of greenspace (tree canopy, vegetation, greened lots), others finding no relationship, and some even reporting a positive association. A key limitation in previous research is the reliance on static measures of greenspace (physical presence), neglecting the crucial aspect of resident interaction. This study aims to address this gap by quantifying how individuals interact with greenspace using cell phone mobility data, which allows for the monitoring of how individuals engage with different physical environments in cities. Several mechanisms could explain the potential link between greenspace and crime. Sociologically, greenspaces might increase street activity, creating pleasant public spaces where neighbors interact and spend more time outdoors. This increased "eyes on the street" effect could deter crime, aligning with crime prevention through environmental design theory and routine activity theory. Psychologically, greenspace could improve cognitive functioning, potentially leading to increased self-control and reduced aggression. Alternatively, the causal direction could be reversed: less crime might lead to increased greenspace usage and street activity, or high-crime neighborhoods might experience neglect of greenspace maintenance. This study utilizes unique cell phone mobility datasets to measure street activity and park visits, enabling the examination of these factors' independent associations with crime in Chicago and New York City. The study aims to determine if street activity and greenspace usage add unique predictive information to crime models and to investigate the unique associations of intentional (park visits) and incidental (tree canopy) greenspace contact with crime.

The existing literature on the relationship between urban greenspace and crime presents a mixed picture. Some studies have found a negative correlation between the amount of greenspace and crime rates, suggesting that more green spaces lead to lower crime rates. Other studies have found no relationship between the two, while a few have even found a positive correlation, indicating that crime might be concentrated in or around green spaces. The inconsistencies may stem from methodological differences, particularly the reliance on static measures of greenspace, such as tree canopy density, rather than accounting for how residents actually use these spaces. This oversight potentially obscures the true relationship between greenspace use and crime. The study’s authors highlight the need for more nuanced measures that reflect the dynamic interaction between residents and their environment, rather than simply the physical presence of green spaces.

This study employed a multi-faceted approach to investigate the relationship between greenspace, street activity, and crime. The research utilized several datasets, including: 1. **Cell Phone Trace Data:** This data, obtained from Carto, provided location information for tens of thousands of residents in Chicago and New York City. The data points (or "pings") contained latitude/longitude coordinates, timestamps, estimated precision, and unique device identifiers. Data was anonymized and processed to respect privacy laws. The raw data were used to generate metrics for park visits, museum visits, and local street activity. 2. **Land Cover Data:** LiDAR data from the University of Vermont's Spatial Analysis Lab were used to measure percent tree canopy and percent grass coverage at the census tract level. This provided a measure of the physical presence of greenspace. 3. **Crime Data:** Open-source crime data from the cities’ open data portals for 2017 were used. Crimes were categorized as violent or non-violent and aggregated to the census tract level. Any crime without location data was removed. 4. **Demographic Data:** Demographic data were obtained from the U.S. Census Bureau's American Community Survey (2012-2017). This included data on population, working population, median household income, unemployment, poverty, housing conditions, educational attainment, racial composition, and residential stability. **Data Analysis:** The analysis employed a two-step process. First, the influence of percent Black and percent Hispanic populations on crime was regressed out to control for confounding effects. Second, spatial error models were used at the census tract level to assess the relationships between the independent variables (tree canopy, grass coverage, park visits, distance to parks, museum visits, street activity, and demographic factors) and the dependent variables (violent and non-violent crime). The Akaike Information Criterion (AIC) was used to compare model fit. Finally, an exploratory directed acyclic graph (DAG) analysis using the fast causal inference (FCI) algorithm was employed to investigate direct and indirect relationships between the variables of interest and crime. **Specific Variable Construction:** Park and museum visits were calculated as the number of visits per device per day to specific locations, and were aggregated to the census tract level. Street activity was calculated using a novel method to measure the portion of a resident's recorded location data that is in close proximity to their home (1-k nearest neighbors), corrected for variations in census tract populations. This approach addressed potential biases caused by varying tract sizes and populations, ensuring consistency in the measurement of local street activity.

The key findings of the study are summarized below: **Chicago:** * Significant negative associations were found between tree canopy, park visits, and street activity, and both violent and non-violent crime rates. These relationships held true even after controlling for socioeconomic factors. * The model including all three variables (tree canopy, park visits, and street activity) showed the best fit (lowest AIC). * A 5% increase in street activity was associated with 6.9% less non-violent crime and 9% less violent crime. A 25% increase in park visits (relative to the average) was associated with 4.9% less non-violent crime and 6.8% less violent crime. * A 5% increase in tree canopy was associated with 3.3% less violent and non-violent crime. * Museum visits showed no significant relationship with crime. **New York City:** * Significant negative associations were found between park visits and street activity, and both violent and non-violent crime. Tree canopy was not significant in these models. * Again, the model with all three variables had the best fit. * A 5% increase in street activity was associated with 5% less non-violent crime and 2.7% less violent crime. A 25% increase in park visits (relative to the average) was associated with 4.8% less non-violent crime and 5.7% less violent crime. * Museum visits were not significantly associated with crime. **DAG Analysis:** * The DAG analysis revealed direct relationships between both park visits and street activity and crime in both cities. No direct relationship was found between tree canopy and crime. The relationships between crime and park visits, poverty, and foreign-born population were bidirectional, indicating possible confounding factors or unmeasured variables.

This study provides robust evidence for the significant and independent contributions of greenspace usage (park visits) and neighborhood street activity to lower crime rates in two large US cities. The consistency of results across Chicago and New York City, despite differences in crime rates and other characteristics, strengthens the findings' generalizability. The finding that both park visits and street activity account for unique variance in predicting crime points to the importance of multiple interacting mechanisms. This supports both the sociological hypothesis that increased street activity deters crime through the “eyes on the street” effect and the psychological hypothesis that exposure to greenspace improves cognitive functioning, leading to better self-control. The weaker and less consistent association of tree canopy with crime compared to greenspace use highlights the importance of considering actual engagement with, rather than merely the physical presence of, greenspace. The lack of a relationship between museum visits and crime further suggests the specificity of the effects to park use. These findings have significant implications for urban planning and crime prevention strategies.

This study demonstrates the significant and independent contributions of park visits and street activity in predicting lower crime rates in Chicago and New York City. The results suggest that multiple mechanisms link greenspace use and street activity to reduced crime. Future research should investigate causal relationships using interventions, explore the role of sociocultural factors in greenspace engagement, and expand to a broader range of cities to further validate the generalizability of these findings. Interventions focusing on increasing both equitable access to greenspace and neighborhood amenities that promote street activity may prove to be cost-effective crime prevention strategies.

The study has several limitations. The reliance on cell phone mobility data introduces potential biases related to smartphone ownership and usage patterns, which could affect the generalizability of the findings. The measure of street activity may also be improved, as it does not capture the quality or type of social interactions. The DAG analysis is exploratory, and causal inferences should be viewed cautiously. The study's focus on two cities might limit its generalizability. Additionally, unmeasured confounding variables could be influencing the observed relationships.