The Analysis of Bounded Count Data in Criminology

Background

Criminological research utilizes several types of delinquency scales, including frequency counts and, increasingly, variety scores. The latter counts the number of distinct types of crimes an individual has committed. Often, variety scores are modeled via count regression techniques (e.g., Poisson, negative binomial), which are best suited to the analysis of unbounded count data. Variety scores, however, are inherently bounded.

Methods

We review common regression approaches for count data and then advocate for a different, more suitable approach for variety scores—binomial regression, and zero-inflated binomial regression, which allow one to consider variety scores as a series of binomial trials, thus accounting for bounding. We provide a demonstration with two simulations and data from the Fayetteville Youth Study.

Conclusions

Binomial regression generally performs better than traditional regression models when modeling variety scores. Importantly, the interpretation of binomial regression models is straightforward and related to the more familiar logistic regression. We recommend researchers use binomial regression models when faced with variety delinquency scores.

     

A quasi-experimental evaluation of the impact of bike-sharing stations on micro-level robbery occurrence

Objectives

To examine if the implementation of bike-sharing stations is linked to robbery occurrence in micro-level street corner units in Cincinnati, OH, USA.

Methods

Propensity score matching was used to select comparison street corner units. The effect of bike-sharing station implementation on robbery occurrence across weekly, biweekly, and monthly observations was estimated using repeated measures multi-level logistic regression models.

Results

Bike-sharing stations did not statistically significantly link to robbery occurrence in immediate or nearby street corner units after implementation.

Conclusions

Numerous explanations consistent with Crime Pattern Theory may explain the null effect of bike-sharing stations on robbery occurrence. Future research should continue to examine how changes in the urban backcloth, such as bike-sharing stations, impact geographic crime patterns.

     

Working with Misspecified Regression Models

Objectives

Conventional statistical modeling in criminology assumes proper model specification. Very strong and unrebutted criticisms have existed for decades. Some respond that although the criticisms are correct, there is for observational data no alternative. In this paper, we provide an alternative.

Methods

We draw on work in econometrics and statistics from several decades ago, updated with the most recent thinking to provide a way to properly work with misspecified models.

Results

We show how asymptotically, unbiased regression estimates can be obtained along with valid standard errors. Conventional statistical inference can follow.

Conclusions

If one is prepared to work with explicit approximations of a “true” model, defensible analyses can be obtained. The alternative is working with models about which all of the usual criticisms hold.

     

The Recurring Victimization of Individuals with Mental Illness: A Comparison of Trajectories for Two Racial Groups

Objective

This study explores whether the trajectories of recurring victimization of Black persons diagnosed with major mental illnesses vary from the trajectories of their White counterparts. Further, the study examines whether the risk factors for recurring victimization among persons with major mental illness vary by race.

Methods

Using data from the MacArthur Violence Risk Assessment Study (MacRisk), two separate hierarchical binomial regression models were estimated to compare the recurring victimization trajectories of Black and White MacRisk participants. Cross-level interaction terms were also estimated to determine if the coefficients for each of the time-varying covariates included in the analyses were significantly different across race.

Results

The findings indicate that the trajectories of recurring victimization for Black persons with serious mental illness are significantly different from those of White persons with serious mental illness. Specifically, Black persons’ trajectories remain relatively stable over time, while the risk of recurring victimization declines for Whites as time since release from the hospital increases. Further, the effects of alcohol abuse on revictimization risk vary by race.

Conclusion

The findings suggest that the life experiences of Black persons with mental illness may be different from their White counterparts, which is likely to contribute to distinct patterns of recurring victimization over time. Future research should continue to explore recurring victimization among diverse samples to identify potential sources of the variation in revictimization trajectories across race.

     

Dissecting the Prevalence and Incidence of Offending During the Crime Drop of the 1990s

Objectives

To evaluate whether the 1990s crime drop reflects a decrease in offending prevalence (the fraction of the population engaged in crime), offending incidence (the frequency of offending among active criminals), or some combination of the two.

Methods

We use individual-level longitudinal data on adolescent offending patterns from the Pittsburgh Youth study (PYS), integrating information from the youngest and oldest cohorts to compare offending among 17–18 year old males at the beginning and end of the 1990s. Logistic and negative binomial regression models are estimated to assess whether there are significant differences in offending prevalence and incidence during the 1990s.

Results

The reduction in property crime rates in the PYS sample during the 1990s can be attributed to declines in both offending prevalence and incidence. The overall decline in serious violence during the 1990s for the full sample was primarily the result of a falloff in prevalence. However, for black youth our results indicate significant reductions in both the prevalence and incidence of serious violence. We did not detect a significant difference in illegal drug sales during the period.

Conclusions

Using longitudinal data on individuals to decompose aggregate crime trends into changes in the prevalence and incidence of offending offers insights into the nature of the 1990s crime drop that cannot be discerned from aggregate crime data. Future research should build on the current study by examining the specific mechanisms that influence change over time in crime prevalence, incidence, or both.

     

A long-term follow-up evaluation of the Minnesota High Risk Revocation Reduction reentry program

Objectives

This study examines the effectiveness of the High Risk Revocation Reduction (HRRR) program, a reentry program designed to reduce recidivism among offenders released from Minnesota state prisons.

Methods

Adult male release violators were randomly assigned to a treatment group that received supplemental case planning and access to community service and programs, or to a control group that received standard case management. Survival analysis was used to examine rearrest, reconviction, reincarceration for a new offense, and supervised release revocation.

Results

The results of Cox regression models showed that participation in HRRR significantly reduced the risk of rearrest but had no effect on the other measures of recidivism.

Conclusion

The results provide limited support for the program, although its effectiveness appeared to decline during the second phase of implementation. HRRR also reduced costs; however, the estimated benefits were not robust across all sensitivity analyses.

     

Punishing the Wicked: Examining the Correlates of Sentence Severity for Convicted Sex Offenders

Objectives

To examine the correlates of sentence severity for convicted sex offenders under sentencing guidelines, contrasted with individuals convicted of non-sexual, violent offenses.

Methods

Drawing on 7 years of data from the Pennsylvania Commission on Sentencing, we utilize a logit-negative binomial hurdle model to examine the predictors of incarceration and sentence length, and an accompanying Oaxaca–Blinder decomposition of the gap in sentencing outcomes between the groups. We then implement a quantile regression framework to examine variation in effects across the distribution of sentence lengths. All analyses are contrasted with a matched sample of violent offenders to consider the extent to which estimated associations are unique to sex offenders.

Results

The analyses suggest several predictors of sentence severity for sex offenders, and that these predictors vary between the incarceration and sentence length decisions. In comparing effects for sex and matched violent offenders, divergent effects were observed for both case and offender characteristics. An Oaxaca–Blinder decomposition suggests that differences in the coefficient estimates account for less than one-fifth of the gap in average sentencing outcomes between sex and violent offenders. Subsequent quantile regressions indicate that these effects vary considerably over the sentence length distribution in ways that are not captured or obscured by the hurdle models.

Conclusions

The predictors of sentence severity for sex offenders, and points of divergence from violent offenders, are congruent with the notion that judges utilize crime-specific stereotypes in arriving at sentencing decisions. Further, the application of quantile regression following point-based estimation can reveal meaningful patterns in sentencing disparities.

     

Human trafficking for labour exploitation: the results of a two-phase systematic review mapping the European evidence base and synthesising key scientific research evidence

Objectives

Our objectives were (1) to systematically map the contours of the European evidence base on labour trafficking, identifying its key characteristics, coverage, gaps, strengths and weaknesses and (2) to synthesise key scientific research.

Methods

We took a two-phase approach: a systematic map followed by a detailed synthesis of key scientific research evidence. Our search strategy included 15 databases, hand searches of additional journals, backwards searches, snowball searches and expert recommendations. We identified and screened 6106 records, mapped 152 and synthesised eight.

Results

Overall, the literature was limited and fragmented. Reports produced by official agencies dominated; academic authorship and peer-reviewed outputs were comparatively rare. Few publications met minimum scientific standards. Qualitative designs outweighed quantitative ones. Publications typically described trafficking’s problem profile and/or discussed interventions; they rarely assessed trafficking’s impacts or evaluated interventions. Even among the key scientific research, the quality of evidence was variable and often low. Particular weaknesses included poor methods reporting, unclear or imprecise results and conclusions not properly grounded in the data. The synthesised studies were all exploratory, also sharing other design features. Common themes identified included: poor treatment of victims; diversity of sectors affected and commonalities among victims; inadequacies of current responses; and barriers to interventions.

Conclusions

There is a lack of high-quality studies into European labour trafficking. Methodological opacity, insufficient rigour and publication in non-indexed locations impede the identification, assessment and synthesis of evidence. Adherence to higher reporting standards would further the field’s development and particular research gaps should be addressed.

     

Offender-focused police intervention in residential burglary and theft from vehicle hot spots: a partially blocked randomized control trial

Objectives

To test an offender-focused police intervention in residential burglary and residential theft from vehicle hot spots and its effect on crime, arrests, and offender recidivism. The intervention was prevention-focused, in which detectives contacted offenders and their families at their homes to discourage criminal activity.

Methods

The study was a partially blocked, randomized controlled field experiment in 24 treatment and 24 control hot spots in one suburban city with average crime levels. Negative binomial and ordinary least squares (OLS) regression were used to test the effect of the presence of intervention and its dosage on crime and offender recidivism, and examination of average and standardized treatment effects were conducted.

Results

The analyses of the hot spot impact measures did not reveal significant results to indicate that the treatment had an effect on crime or arrest counts, or on repeat arrests of the targeted or non-targeted offenders living in the hot spots. However, the relationships, while not significant, were in a promising direction.

Conclusions

The collective findings from all four impact measures suggest that the intervention may have had some influence on the targeted offenders, as well as in the treatment hot spots. So, while the experimental results did not show an impact, they are promising. Limitations include large hot spots, the low case number, low base rates, and inadequate impact measures. Suggestions are provided for police agencies and researchers for implementing preventive offender-focused strategies and conducting studies in suburban cities.

     

Crimes by Visitors Versus Crimes by Residents: The Influence of Visitor Inflows

Objectives

To disaggregate the crime impact of visitor inflows. There is increasing evidence that visitors can make a major contribution to levels of crime in a given neighbourhood: crimes by visiting offenders may add to those committed by local offenders, while visitors (and their property) may provide local offenders with additional opportunities for crime.

Methods

Using police-recorded crime data for a large Eastern Canadian city we determined whether individuals charged or chargeable for property and violent crimes were visitors or residents of census tracts (CT) where crimes had been committed. This information was combined with data from a large transportation survey, allowing us to estimate daily population flows into each CT for four purposes (work, shopping, recreation, and education). Negative binomial regression models including spatial lags were used.

Results

An increase in visitor inflow not only increases the number of visitors charged with crimes but also the number of local residents charged. These effects vary significantly by visit purpose: more infractions are committed in tracts where visits are for recreation and, to a lesser extent, for shopping. Findings for work and education are mixed.

Conclusions

One important implication of our results is that, because most studies of aggregate crime counts or rates fail to account for whether crimes have been committed by visitors or residents, previous research may have provided hasty, partial, or even erroneous explanations for crime concentrations.

     

Retrieving administrative data to assess long-term outcomes: a case study of the 23-year follow-up of the Milwaukee domestic violence experiment

Objectives

To describe how social scientists, criminal justice practitioners, and administrative agencies collected administrative data to follow-up a criminological experiment after two decades. To make recommendations that will guide similar long-term follow-ups.

Methods

A case study approach describes the processes of and sociological benefits to collecting administrative data to assess criminal justice and life-course outcomes.

Results

While maintaining experimental integrity, we developed, executed, and verified processes to retrieve arrest, mortality, and residential data for the experimental subjects, which enabled us to complete the longest ever follow-up of a criminal justice experiment.

Conclusions

When experiments have policy implications, administrative data may be preferable to survey data for assessing primary effects. Successful social science research can be conducted in conjunction with multiple administrative agencies.

     

Criminals and the Price System: Evidence from Czech Metal Thieves

Objectives

This paper tests the economic theory of criminal behavior. Specifically, it looks at “the carrot” side of the theory, studying how thieves react to changes in monetary gains from crime.

Methods

Using a unique crime-level dataset on metal theft in the Czech Republic, we study thieves’ behavior in a simple regression framework. We argue that variation in metal prices represents a quasi-experimental variation in gains from crime. It is because (1) people steal copper and other nonferrous metals only to sell them to scrapyard and (2) prices at scrapyards are set by the world market. This facilitates causal interpretation of our regression estimates.

Results

We find that a 1% increase (decrease) in the re-sale price causes metal thefts to increase (decrease) by 1–1.5%. We show that the relationship between prices and thefts is very robust. Moreover, we find that thieves’ responses to price shocks are rapid and consistent.

Conclusion

Our results are in line with the economic model of crime, wherein criminal behavior is modeled as a rational agent’s decision driven by the costs and benefits of undertaking criminal activities. Our estimates are also consistent with recent results from the United Kingdom, suggesting these patterns are more general.

     

Addressing Unobserved Heterogeneity in the Relationship Between Crime and Consumer Confidence

Objectives

This study revisits the relationship between property crime and economic conditions with the latter being represented by a collective economic perception variable in the form of the Index of Consumer Confidence (ICC). The present work takes this application to assess the severity of cross-sectional dependence and nonstationarity, two issues that are deemed pervasive in macro panels but have not been given sufficient consideration in previous research.

Methods

The dataset comprises information for five Canadian regions over a time period of 24 years from 1982 to 2005. The study compares the parameter estimates and residual properties of the commonly used two-way fixed effects (2FE) model and the augmented mean group (AMG) estimator where the latter can accommodate nonstationarity and cross-sectional dependence that potentially arise from unobserved common factors.

Results

In contrast to the 2FE approach, when using the AMG estimator one can reject the null hypothesis that the current ICC has no impact on crime. Some of the effects still hold when an alternative economic indicator, the unemployment rate of young males, is added to the model. Diagnostic tests confirm that the commonly used 2FE estimator yields nonstationary and cross-sectional dependent residuals, whereas the heterogeneous parameter model produces more favorable diagnostic results.

Conclusions

The findings provide evidence supporting the hypothesis that subjective measures of economic conditions are linked to financially-motivated crime rates. Through this application, the study demonstrates the importance of examining underlying data properties and regression residuals in empirical work to ensure the validity of estimates.

     

Conditional Race Disparities in Criminal Sentencing: A Test of the Liberation Hypothesis From a Non-Guidelines State

Objectives

To test the liberation hypothesis in a judicial context unconstrained by sentencing guidelines.

Methods

We examined cross-sectional sentencing data (n = 17,671) using a hurdle count model, which combines a binary (logistic regression) model to predict zero counts and a zero-truncated negative binomial model to predict positive counts. We also conducted a series of Monte Carlo simulations to demonstrate that the hurdle count model provides unbiased estimates of our sentencing data and outperforms alternative approaches.

Results

For the liberation hypothesis, results of the interaction terms for race x offense severity and race x criminal history varied by decision type. For the in/out decision, criminal history moderated the effects of race: among offenders with less extensive criminal histories blacks were more likely to be incarcerated; among offenders with higher criminal histories this race effect disappeared. The race x offense severity interaction was not significant for the in/out decision. For the sentence length decision, offense severity moderated the effects of race: among offenders convicted of less serious crimes blacks received longer sentences than whites; among offenders convicted of crimes falling in the most serious offense categories the race effect became non-significant for Felony D offenses and transitioned to a relative reduction for blacks for the most serious Felony A, B, and C categories. The race x criminal history interaction was not significant for the length decision.

Conclusions

There is some support for the liberation hypothesis in this test from a non-guidelines jurisdiction. The findings suggest, however, that the decision to incarcerate and the sentence length decision may employ different processes in which the interactions between race and seriousness measures vary.

     

Examining the Relationship Between the Structural Characteristics of Place and Crime by Imputing Census Block Data in Street Segments: Is the Pain Worth the Gain?

Objective

The current study proposes unique methods for apportioning existing census data in blocks to street segments and examines the effects of structural characteristics of street segments on crime. Also, this study tests if the effects of structural characteristics of street segments are similar with or distinct from those of blocks.

Methods

This study compiled a unique dataset in which block-level structural characteristics are apportioned to street segments utilizing the 2010 U.S. Census data of the cities of Anaheim, Santa Ana, and Huntington Beach in Orange County, California. Negative binomial regression models predicting crime that include measures of social disorganization and criminal opportunities in street segments and blocks were estimated.

Results

The results show that whereas some of the coefficients tested at the street segment level are similar to those aggregated to blocks, a few were quite different (most notably, racial/ethnic heterogeneity). Additional analyses confirm that the imputation methods are generally valid compared to data actually collected at the street segment level.

Conclusions

The results from the street segment models suggest that the structural characteristics from social disorganization and criminal opportunities theories at street segments may operate as crucial settings for crime. Also the results indicate that structural characteristics have generally similar effects on crime in street segments and blocks, yet have some distinct effects at the street segment level that may not be observable when looking at the block level. Such differences underscore the necessity of serious consideration of the issues of level of aggregation and unit of analysis when examining the structural characteristics-crime nexus.

     

What Drives the Relationship Between Early Criminal Involvement and School Dropout?

Objective

We examine the relationship between early criminal involvement and school dropout, and analyze which factors underlie this relationship, making use of administrative data from the Netherlands.

Methods

We start by determining the unconditional correlation between early criminal involvement and school dropout, using a basic ordinary least squares model. As this association is likely to be driven by different factors, we proceed by including an extensive set of observable family and individual characteristics into the estimation model. We further proceed to models that account for the influence of unobservable heterogeneity by estimating school, class, sibling and twin fixed effects.

Results

Criminal involvement is associated with an 11 percentage point higher probability of school dropout. The magnitude of this relationship decreases gradually when we account for larger shares of observed and unobserved heterogeneity. The coefficient in the same-gender twin fixed effects model indicates a 3 percentage point higher probability of school dropout, which is statistically significant at a 10 % level. We also find that the association between criminal involvement and school dropout is stronger if juveniles are involved in severe criminal activities.

Conclusions

We conclude that the observable and unobservable factors for which we account explain around 73 % of the unconditional correlation between criminal involvement and school dropout. The remaining variation likely reflects individual-specific characteristics that are different between same-gender twins. A true treatment effect, if existing, is likely to be relatively small. At the same time, serious criminal behavior appears to causally affect school dropout.

     

Does providing inmates with education improve postrelease outcomes? A meta-analysis of correctional education programs in the United States

Objectives

Our study addresses the question: Does providing inmates with education while incarcerated reduce their chances of recidivism and improve their postrelease employment prospects?

Methods

We aggregated 37 years of research (1980–2017) on correctional education and applied meta-analytic techniques. As the basis for our meta-analysis, we identified a total of 57 studies that used recidivism as an outcome and 21 studies that used employment as an outcome. We then applied random-effects regression across the effect sizes abstracted from each of these studies.

Findings

When focusing on studies with the highest caliber research designs, we found that inmates participating in correctional education programs were 28% less likely to recidivate when compared with inmates who did not participate in correctional education programs. However, we found that inmates receiving correctional education were as likely to obtain postrelease employment as inmates not receiving correctional education.

Conclusion

Our meta-analysis demonstrates the value in providing inmates with educational opportunities while they serve their sentences if the goal of the program is to reduce recidivism.

     

Measuring the Spatial and Temporal Patterns of Police Proactivity

Objectives

To measure where officers engage in proactive, self-initiated activities, how much time they spend being proactive, and whether their proactive activities coincide with crime patterns.

Methods

This study uses Andresen’s Spatial Point Pattern Test to compare the spatial similarity between police proactivity and crime, as well as regression modeling to explore the relationship between proactivity and crime and the time spent on proactivity and crime.

Results

In the jurisdiction examined, high levels of proactivity are noted. This proactive activity is more likely to occur in places where crime is most concentrated. Additionally, the number of proactive calls and the proactive time spent per crime-and-disorder call remain high and stable across spatial scales. For each crime call received at a street block, police initiated 0.7 proactive activities and spent approximately 28 min carrying out proactive works.

Conclusions

This study develops a way of measuring proactive activity by patrol officers using calls for service data. We find that not only do officers in this jurisdiction exhibit higher levels of proactivity to prevent crime (compared to reacting to crime), but they also do so in targeted, micro-place ways. Agencies may consider using similar techniques to gauge the levels of proactivity in their agencies if proactive activity is a goal.

     

Acquisitive Crime and Inflation in the United States: 1960–2012

Objectives

Inflation is conspicuous by its absence from recent research on crime and the economy. We argue that price inflation increases the rate of crimes committed for monetary gain by fueling demand for cheap stolen goods.

Methods

The study includes inflation along with indicators of unemployment, GDP, income, consumer sentiment, and controls in error correction models of acquisitive crime covering the period from 1960 to 2012. Both short- and long-run effects of the predictors are estimated.

Results

Among the economic indicators, only inflation has consistent and robust short- and long-run effects on year-over-year change in the offense types under consideration. Low inflation helps to explain why acquisitive crime did not increase during the 2008–2009 recession. Imprisonment rates also have robust long-run effects on change in acquisitive crime rates.

Conclusions

Incorporating inflation into studies of crime and the economy can help to reduce the theoretical and empirical uncertainty that has long characterized this important research area in criminology.

     

A meta-analysis update on the effects of early family/parent training programs on antisocial behavior and delinquency

Objective

To update Piquero et al.’s meta-analysis on early family/parent training programs.

Methods

Screening of eligible studies was carried out for the period between January 2008 and August 2015. An additional 23 studies were identified, which were added to the original database of 55 studies, totaling an overall sample of 78 eligible studies. A random-effects model was used to obtain an overall mean effect size estimate. Additional analyses were performed to assess publication bias and moderation.

Results

An overall average, positive, and significant effect size of 0.37 was calculated, which corresponds to 32 out of 100 in a treated group versus 50 out of 100 in a control group who offended. There was some evidence of publication bias and moderation.

Conclusions

Early family/parent training programs are an effective evidence-based strategy for preventing antisocial behavior and delinquency.