Despite directives to investigate ways to abolish or, at the very least, further reduce the use of pecuniary sanctions for young offenders, fines remain the most common form of justice disposal among young people. This study examines the effects of replacing fines with diversion through prosecution deferrals on recidivism. The study utilises administrative data and exploits a shift in prosecutorial decision-making in Sweden as a natural experiment, using a difference-in-differences approach. The reform substantially increased the use of prosecution deferrals at the expense of (low-amount) fines. Recidivism was operationalised as both rearrest and reconviction. Heterogeneity analyses were conducted across three broad categories: demography (sex), socioeconomic status (highest attained parental educational level), and criminal history (prior convictions). The findings show no average effect of the reform on recidivism, whether measured by rearrest or reconviction. Analyses stratified by sex, socioeconomic status, and criminal history show little evidence of effect heterogeneity. The increased use of prosecution deferrals at the expense of fines had no crime-reducing effect and did not increase recidivism rates. Further reductions in the use of pecuniary sanctions for young offenders committing minor offences therefore appear to pose little risk to crime-reduction objectives. Rather, such a shift may signal societal investment in young individuals and conserve resources for legal authorities.
The New York City Police Department’s Summer All Out (SAO) was a recurring, presence-based foot-patrol initiative non-randomly deployed to precincts between 2014 and 2019. Using both two-way fixed effects (TWFE) and a novel matching and weighting estimator (PanelMatch),we assess whether the SAO deployment influenced a variety of crime outcomes. Since the treatment pattern varies over time intermittently, we analyze a 10-year precinct-month-year panel using TWFE, a difference-in-differences (DiD) methodology. We then expand with Imai et al.’s (2023) PanelMatch, a causal inference estimator never before used in this field, which selects and reweights control units based on similarity in pretreatment crime trajectories and precinct level covariates, thereby addressing endogenous targeting of precincts and evolving group composition over time. TWFE estimates suggest SAO was associated with increases in violent crime, while property crime shows no detectable change. When applyingPanelMatch, which employs propensity-score matching and weighting, and used lagged outcomes and precinct covariates in the refinement process, these effects attenuate toward zero. We find no statistically significant impact of SAO foot deployments on violent or property crime. These results persist with robustness checks using different crime types, including alternative functional forms. The divergence between TWFE and PanelMatch results is consistent with a scenario where SAO officers were assigned to precincts experiencing short-term crime surges. The divergence between the results of a more traditional TWFE and the novel PanelMatchanalysis suggest that evaluations of episodic, repeatedly targeted interventions should account for endogenous assignment using designs that leverage pretreatment outcome histories. Substantively, SAO’s presence-based foot patrols did not measurably reduce violent or property crime on average—reallocating officers for fixed deployments within large jurisdictions may not yield general deterrent effects.
The current study examines the temporal exclusivity of crime across urban space. Specifically, we look at whether the temporal structure of crime differs between weekdays and weekends and map how these spatiotemporal synchronizations diverge between violent and property crime in New York City Crime incident hours were transformed into unit vectors on a circular plane using an aoristic allocation framework to handle temporal uncertainty. The Mean Resultant Length was calculated to measure the degree of temporal exclusivity at the block group level. To distinguish systematic diurnal rhythms from ambient stochastic noise while accounting for multiple comparison inflation, we implemented the Rayleigh test of uniformity. Violent crime exhibits stronger temporal concentration within individual locations but collapses to a relatively small number of statistically significant hotspots under stringent multiple-comparison corrections. This pattern likely reflects the combination of lower incident frequency and limited statistical power at the block group level. In contrast, property crime displays a more spatially robust pattern of temporal synchronization. Although its hourly concentration is generally less pronounced within individual locations, the substantially larger number of incidents produces stable and statistically reliable temporal rhythms across the city. The findings suggest that the temporal structure of crime is shaped by routine activity patterns that vary across the weekly cycle. By introducing a circular statistical framework to measure the temporal concentration of crime, this study highlights the importance of integrating temporal dynamics into research on crime and place.
This study examines whether state-level variation in gun interest and firearm transactions primarily reflects local conditions or broader national dynamics. Monthly Google searches containing the word “gun,” NICS background checks, and state and national gun homicide counts were analyzed for all 50 states (2004–2019, with extensions around 2020). We estimated thousands of pairwise regressions predicting gun interest in one state from activity in another while conditioning on contemporaneous gun homicides. Robustness checks included principal components analysis, lagged models, placebo searches, and Tennessee permit data. State-level gun interest and transactions are highly synchronized. Activity in one state predicted activity in another nearly as well as a state predicted itself (average cross-state coefficients = .86 for Google searches and .73 for NICS checks), even after conditioning on contemporaneous state-level gun homicides. In contrast, the average effect of state-level gun homicides on gun interest was small and centered near zero prior to 2020 and changed only moderately using data centered around 2020. Over the last 20 years, national events and sentiment appear to account for a large share of the within-state variation in gun interest and transactions, while state-level conditions mattered primarily during periods of extreme national stress. These findings suggest that assumptions of widespread local reverse causality between crime and gun demand may require greater nuance in medium-run analyses. Limitations include the use of proxy measures of gun interest.
Our objective was to examine the association between anti-cartel military interventions and homicide rates in Mexico. Nearly half a million people became homicide victims in Mexico in the sixteen years following President Calderón’s declaration of war against the cartels in 2007. Mexico possesses persistently high but geographically varying homicide rates, and the military is central to these anti-cartel interventions. This context presents a unique opportunity to analyze the heterogeneous violence patterns associated with intervention exposure in a quasi-experimental framework. We examined the association between anti-cartel military interventions and total and gun (as a proxy for cartel-related) homicide rates in a sample of 2,429 Mexican municipalities from 2000 to 2022. We tested for multiple types of associations, specifying interventions as both “permanent” and “alternating.” We used Two-Way Fixed Effect Difference-in-Difference models to estimate the average treatment on the treated. We found no evidence that anti-cartel military interventions were associated with a reduction in lethal violence as intended. Instead, on average treated municipalities had higher total and cartel-related homicide rates than comparable municipalities that did not experience intervention. The results were consistent across a wide array of sensitivity tests. We interpret our findings in the context of illegal enterprise theory, which suggests that military intervention may be associated with destabilizing criminal organizations and thus can lead to heightened competition, fragmentation, and increased violence.
We estimate the causal effect of school holidays on crime by exploiting the staggered timing of the Swedish winter sports break (sportlov) across municipalities. We use a difference-in-differences design with municipality and year-by-week fixed effects on panel data covering 290 municipalities over 208 weeks (2021–2024). Poisson pseudo-maximum-likelihood (PPML) is the primary specification, with county-level cluster inference and Romano-Wolf step-down adjusted p-values for family-wise correction. The break reduces recorded assault by about 13 percent (PPML), concentrated among children of compulsory-school age (7–14: -46% ; 15–17: -32% ), with no effect on placebo groups (ages 0–6 or adults). Residential burglary increases by about 14 percent, consistent with empty homes during family travel, while crime rises at ski-destination municipalities during the sportlov season, consistent with tourism-driven inflows. Event-study estimates show no pre-trends and no intertemporal displacement. Aggregate total-crime and property-crime estimates point in the same direction but are smaller and not robust to family-wise correction. The school environment concentrates youth in ways that produce interpersonal-conflict opportunities; the break disperses that concentration for one week and assault drops. Aggregate effects on total and property crime are suggestive but sensitive to specification.
This paper analyzes the factors associated with a decline of 1.7 percentage points, from 5.7
This study assesses how physical and social distances influence locational choice decision-making across six crime types—burglary, larceny, vehicle theft, assault offense, robbery, and drug violations. It aims to identify how physical distance and social distance affect locational choices in offending and the exposure patterns underlying victimization, while comparing offender and victim mobility across different crime types. Employing a discrete choice modeling (DCM) framework, this analysis uses 341,804 police incident entries and 40,228 police arrest records from Dallas, covering 2014-06-01 to 2020-03-23. Data integrated from the 2010 Census and American Community Survey 5-year estimates are analyzed at the census block group level, controlling for features of target block groups. Both offender and victim mobility exhibit clear distance decay patterns, with higher physical distances significantly reducing the likelihood of crime involvement in a block groups. Racial dissimilarity suppresses both offender and victim mobility across all crime types. Victim mobility is uniformly constrained by income difference. However, offender mobility responds to income differences in a crime-type-specific manner. The findings demonstrate that offenders’ broader mobility pattern reflects intentional target selection and risk-seeking behavior, whereas victim mobility remains anchored in familiar social and spatial environments. These results emphasize the importance of disaggregating crime types and incorporating both physical and social distance in studying crime mobility.
We examine whether an officer’s likelihood of using physical force during a 911 call response is associated with exposure to peers who have previously used force. We use data on joint responses to 911 calls by Dallas police officers to reconstruct the social network of on-duty patrol interaction within the Dallas Police Department. Merging these data with Response to Resistance reports on use of force, we use a matched case-control design implemented with conditional logistic regression and permutation tests to estimate whether lagged and contemporaneous exposure to peers with a history of use of force is associated with an officer’s own likelihood of using physical force, holding constant incident-level situational context. Greater lagged exposure to peers’ prior use of force is associated with a higher likelihood that a focal officer subsequently uses physical force during a 911 response. In contrast, contemporaneous exposure is associated with a reduced likelihood of force in the same incident. These findings are consistent with social learning perspectives suggesting that peer influence may operate through longer-term processes of learning and socialization while also shaping how officers coordinate behavior during specific encounters. More broadly, the results highlight the importance of considering how routine on-duty interactions shape police behavior and suggest that social networks may both reinforce and constrain the use of force.
This study aims to identify key factors that shape the global network of illicit financial flows (IFFs) related to money laundering and other financial crimes. Specifically, it examines the factors determining both i) the selection of destination countries and ii) the volume of illicit funds laundered. We developed a Heckman-adjusted gravity model of illicit financial flows, utilizing data from Suspicious Activity Reports lodged between 2007 to 2017. The first stage of the model analyses the selection of destination countries, while the second stage estimates the volume of laundered funds. Key variables include GDP, financial service quality, Egmont membership, corruption levels, and geographic distance. The Heckman correction is applied to address selection bias. Our findings indicate that wealthier countries attract higher levels of illicit financial flows. However, high quality financial services deter both the selection of a country for laundering and the volume of funds laundered. Reported IFFs are more likely from countries with high corruption and conflict levels. Trade, culture and geographic proximity are also found to be correlated with the likelihood and magnitude of reported IFFs. Results show evidence of displacement and provide evidence of the link between illicit financial flows and the international flow of trade, people and remittances. Limitations include potential biases in the data and the exclusion of non-USD transactions.
The peer context is crucial in victimization research, yet its role in repeat victimization remains understudied. This study examines the mediating pathways through which youth who are victimized increase delinquent peer affiliations, thereby elevating the risk of victimization later in life. Further, this study explores how this mediating mechanism changes over the life course as well as how the pathway varies by gender. The analysis uses six waves of data from a longitudinal survey of South Korean youth aged 13–18. A Random-Intercept Cross-Lagged Panel Model (RI-CLPM) examines within-person changes over time to identify mediating mechanisms. In addition, to assess the gender-specific pathway of repeat victimization, a multi-group RI-CLPM is applied. Findings reveal that adolescents who are victimized increase delinquent peer affiliations, which in turn increase the risk of repeat victimization in later life. The mediating pathway operates during adolescence, but not at the beginning of emerging adulthood (age 18). In addition, for girls, this mechanism is more prominent earlier in development, while for boys, it is more evident later during adolescence. These findings highlight the need for programs aimed at improving peer relationships after victimization, as well as targeted interventions, especially in the early stages of adolescence for girls and later stages for boys. The current study takes an important step toward a developmental and life-course perspective on victimization. Future research should continue to explore how peer contexts influence the cycle of victimization over time.
Assess the crime prevention through environmental design (CPTED) framework as a neighborhood theory and propose a new computational approach for measuring place visuals. Inspired by lessons from cognitive psychology, this approach augments object-based measures of environments (e.g., how many trees are present) with measures of visual ‘gists’ capturing broad image assessments. 4,800 respondents were surveyed to provide human ratings of five CPTED-inspired gist metrics (preference, complexity, memorability, transparency, and enclosure) for 8,249 Chicago Google Street View images. Gist metrics were evaluated and interpreted using monte carlo-based reliability simulations and hierarchical linear models. Using residual neural networks, we trained a series of computer vision models that could predict human-rated gist scores on new images. After validating out-of-sample performance, these models were applied to estimate CPTED gist scores on a larger set of 187,048 Chicago street-view images. Multi-level variance decomposition analysis was used to probe the nested geographic structure of gist metrics. XGBoost were used to evaluate whether gist features were correlated with crime and prosociality (as measured via voting participation rates). Human-rated gist assessments were highly reliable across participants and were correlated with both the object composition of images and demographic features of the pictured neighborhood. AI-assigned gist labels correlated highly with human-ratings, suggesting very strong out-of-sample accuracy for all gist models. Variance decomposition results suggested CPTED gists substantially vary across census tracts even when accounting for micro-spatial visual differences. XGBoost results suggested violent crime is best predicted using object-based measures, non-violent crime is best predicted using gist measures, and voting participation is best explained using both sets of image features. Aggregate neighborhood visual features are important for understanding why some places experience more crime. Future research should seek to understand how broad visual gists and the presence of specific objects interact to shape behavior and decision-making among guardians, offenders, and targets.
This study evaluates the impact of acoustic gunshot detection technology (ShotSpotter) on crime rates, emergency call volume, police response times, and case clearance rates in Detroit, Michigan. We employ a difference-in-differences design leveraging the staggered rollout of ShotSpotter across Detroit neighborhoods. The treatment group consists of several areas receiving coverage in March 2021, while the control group is comprised of areas covered by ShotSpotter starting in October 2022. Using Detroit Police Department data from January 2017 to October 2022, we estimate Poisson regression models for crime and call volume outcomes and log-linear regression models for police travel times. All models include area and time fixed effects, with robust standard errors. ShotSpotter installation is associated with a 6.5
This study examines the relationship between sporting events and crime near sports venues based on a sample of Major League Baseball (MLB) and National Football League (NFL) regular season games. Open crime data drawn from 21 U.S. cities were analyzed using a quasi-experimental design that compared differences in crime around venues hosting home games to venues whose teams play an away game, at the same time, from 2015 to 2023. Within one kilometer of sports venues, larceny and assault rose when stadiums hosted games. Larceny peaked before and after games, while assaults tended to occur during and after games, suggesting different opportunity structures across crime types. The relationship between sporting events and crime varies by where games are hosted and is sensitive to some analytic considerations, which in turn have important implications for criminological theory and strategies to mitigate the criminogenic effects of these events.
This research examines the spatio-temporal associations between a curfew imposed in Quebec from January 9, 2021, to May 28, 2021, and crime patterns in Montreal. Specifically, it assesses (1) whether changes in crime levels are observed during the curfew period and how these patterns vary across crime types, and (2) whether the temporal organization and spatial distribution of crime differ during this period. The study employs a quantitative design based on official crime data from the Montreal Police Department. Temporal and spatio-temporal analyses were conducted using 3D space-time cubes, a GIS-based method that enables the joint analysis of spatial and temporal dimensions of crime. In addition, linear mixed-effects models were estimated to examine weekly crime trends while accounting for seasonality and temporal dependence. The findings indicate a temporary decline in total crime following the implementation of the curfew. A shift in the temporal distribution of crime is observed, with a greater concentration of offences occurring during curfew-free hours. At the weekly level, violent crime shows a modest reduction during the curfew period, while property crime exhibits a weaker and less consistent association that does not reach conventional levels of statistical significance. Disaggregated analyses by offender-victim relationship do not reveal clear differences in violent crime patterns between the curfew and broader COVID-19 periods. No evidence of substantial spatial displacement of crime associated with the curfew is observed. By integrating spatial and temporal dimensions within a single analytical framework, the space-time cube approach provides a descriptive and nuanced account of crime patterns during the curfew period. Rather than identifying definitive effects, the analyses highlight short-term and heterogeneous variations in the timing and distribution of crime under conditions of restricted mobility. These findings underscore the value of spatio-temporal and descriptive approaches for documenting how crime patterns evolve during exceptional social and policy contexts, and they provide a foundation for future research using longer time frames or comparative designs.
We develop and apply a quasi-experimental framework that adopts a causal inference perspective to examine spatial and temporal patterns of crime across multiple spatial and temporal scales. Using granular spatiotemporal police-recorded crime data from São Paulo state, Brazil, in 2016–2021, we employ difference-in-differences and event study methods to estimate the impacts of the COVID-19 mobility restrictions on crimes and crime hotspots and the heterogeneity in the impacts across space and time in an integrated way. The mobility restrictions temporarily decreased property crimes, not homicide, but increased crime hotspots and altered hourly temporal patterns of property crimes potentially in a persistent way. Patterns were distinct at the municipality and census block levels. Unique patterns were found in urban slums (favelas) and for domestic violence. Our framework effectively uncovered novel evidence regarding spatial and temporal responses to the COVID-19 pandemic for different types of crime in the Global South.
We evaluate the predictive accuracy of machine learning algorithms that forecast individual-level risk of behavioral health-involved encounters with police (BHIP) among community members with prior contact with first responders. We linked arrest and ambulance data from one city (May 2016-October 2017) for 199,679 unique individuals. The first 12 months provided predictive features; the subsequent 6 months served as the outcome window. We compared machine learning models utilizing 283 features to a baseline Prior High-User (PHU) approach based solely on prior BHIP counts. Machine learning achieved positive predictive values (PPV) of 50.8
