Interpretations of two bodies of crime-place research conflict. Land use and crime studies claim particular facilities increase crime. Risky facilities studies show most places of a single type have little or no crime, but a few of that type have a great deal of crime. How can a facility be generally criminogenic and mostly safe? To resolve this conflict, we make use of the fact that a single owner can own multiple facilities and each owner may have consistent management practices in their facilities. We first replicate findings of earlier land use studies with crime and land parcel data from Cincinnati. Second, we cluster land parcels by property owners and re-estimate the land use and crime relationship. The links between land use and crime in the replication decline or disappear after clustering. Findings suggest owners’ place management influences crime at their places, thus resolving the conflict between the two bodies of research.
Correctional authorities require accurate, unbiased, and interpretable tools to predict individuals’ chances of recidivating if released into the community. However, existing prediction models have serious limitations meeting these requirements. We overcome these limitations by applying an established medical diagnostic approach: a relaxed naïve Bayes classifier. Using logistic regression in the form of a naïve Bayes classifier, we estimate the weights of observed features of offenders on recidivism. We apply these weights in a relaxed naïve Bayes classifier to predict the probability of recidivism. Results show that acquired features are stronger predictors of recidivism than innate features. Relaxed naïve Bayes classifier produces far less racial disparity than most alternatives. Critically, it is easier for users to interpret than its alternatives.
This study explores the effectiveness of machine learning algorithms in predicting recidivism, focusing on the impact of race and geographic location variables. Leveraging a dataset from the prisons in Georgia, we assess six algorithms’ forecasting performance, both with and without these key variables. Our findings indicate that geographic location generally enhances predictive accuracy more consistently than race across models. This research highlights the importance of methodological diversity and the complex, model-dependent impact of demographic factors on recidivism prediction. It underscores the potential of machine learning in criminology and criminal justice to provide nuanced insights into criminal behavior, challenging traditional assumptions, and informing evidence-based policy development.
This study examines how crime propensity characteristics relate to cyber delinquency directly and indirectly, through crime opportunity. In doing so, it considers propensity by way of different aspects of personality and crime opportunity in the form of a comprehensive set of online activities. Both the direct and indirect relationships are modeled using path analysis in MPlus with the data from 2,015 students from schools in South Korea. Findings suggest several personality traits and online computer activities are significantly related to cyber delinquency. Moreover, these activities often mediate the effects of propensity as measured by personality factors. Overall, findings support the idea that individual enduring propensities and online activities work in tandem to predict adolescent cyber offending.
This study examines deviant identity in relation to youth offending by combining items tapping both self-appraisal and reflected appraisal. In particular, using survey data from 3,446 Korean youth across five waves of the Korea Youth Panel Survey (KYPS), findings from group-based trajectory modeling (GBTM) present four distinct offending groups—a high-rate chronic group, stable non-offending group, adolescence-limited group, and declining group. Then, findings from the multinomial logit model reveal that deviant identity is a robust predictor of offending for subgroups of adolescents involved in offending at any level in comparison to stable non-offenders. Accordingly, this study supports the idea that deviant identity should be considered as a prominent predictor of a variety of types of youth offending.
This study examines the effects of adherence to street codes on school-based violent victimization. In doing so, it separates street values into two distinct orientations: (1) retaliatory norms, and (2) general toughness norms. We analyze survey data from students across four waves of the Rural Substance Abuse and Violence Project using two-level mixed effects Poisson regression. The model specifies random intercepts across individuals to address possible endogeneity of the repeated measurements per respondent. Further, consistent with a fixed-effects approach, we control for differences across the schools in which students were nested. Findings reveal that adherence to retaliatory norms is negatively related to the frequency of serious violent victimization. In contrast, general toughness norms are positively related to serious violent victimization.