This section provides an analysis of the state of the art in crime economics research, not only to identify critical knowledge gaps in the literature but also to contribute to the dissemination of evidence-based crime reduction and prevention strategies. The aim is to provide a solid starting point to take stock of the lessons learned that can be relevant for Latin America and the Caribbean, and to determine what the priorities should be considered for future research. To limit the scope of the review, a critical inclusion criterion was that an evaluation must have a control or comparison group for causal inference of a crime-and-violence-related outcome. Random assignment studies have been highlighted, but solid quasi-experimental analyses are also incorporated. Also, it is important to note that the crimes studied are related to interpersonal violence, and we exclude other types of crime that are important in the region, such as corruption.
Under the setup developed in Sect. 3, high crime rates in Latin America and the Caribbean are a consequence of higher expected net benefits from illegal activities for prospective criminals. This is due to a low expected return to education or legal jobs (\(U^{\text{nc}}\)), or to a low expected cost of committing crimes (\(U^{\text{c}}\) and \(U^{\text{cp}}\)), or most likely to a combination of both factors.
Most of the empirical literature on crime economics has focused on interventions that may increase the expected cost of committing crimes, and thus have a deterrent effect on prospective criminals. There are three main concepts of the deterrence theory: the certainty, severity, and (less studied) celerity of punishment which prove to be important in a dynamic theoretical crime economics model. Certainty refers to the probability of legal sanctions given the commission of a crime (\(p\)); severity refers to the onerousness of the sanction imposed (\(f\)) and the sentence length; and celerity refers to the lapse in time between the commission of a crime and its punishment (Nagin 2013).
The section of this paper on the police focuses on the certainty and celerity of punishment, while the imprisonment section focuses on the severity of punishment. The imprisonment section also analyzes re-insertion and rehabilitation programs that aim to improve the situation of inmates when they are released and position them to have better opportunities in the legal job market, thus reducing recidivism.
There are also interventions that may increase the expected net benefit of legal activities. These interventions that aim at the social prevention of crime and violence are supposed to increase the payoff of legal activities, thus making crime less attractive. In the case of social prevention targeting youth, a training or employability program would increase the possibilities of labor market attachment and, thus, potential earnings in the legal sector, ultimately reducing crime.
Increasing the expected cost of committing crimes
Police and crime
Under the setup of Sect. 3, the amount of crime will respond the probability that an individual is apprehended after having committed a particular offense (\(p\)). This probability may depend on the society’s investment in a police presence, and whether that presence is operationalized through increased manpower or increased productivity. To the extent that potential offenders are able to observe an increase in police resources and perceive a correspondingly higher risk to criminal participation (\(p\)), crime is expected to decline through the deterrence channel. This section analyzes the responsiveness of crime to both police size and police strategies. For the literature on each of those two topics, the challenges will be discussed with respect to both econometric identification and the interpretation of the results as evidence in favor of deterrence.
With regard to the effect of police size on crime, the identification problem arises from the fact that areas with higher crime rates usually by definition have more police presence. Therefore, when regressing the crime rates on police force size variables using a cross-sectional approach, there is an omitted variable bias. There has been a growing body of literature using quasi-experiments or natural experiments to overcome the endogeneity problem of causal inference in assessing the effect of police force size on crime.
The seminal paper of Levitt (1997) is the first quasi-experimental paper to use instrumental variables to identify the causal effect of police size on crime. Levitt (1997) uses mayoral elections as an instrument, while Levitt (2002) uses the number of firefighters. Evans and Owens (2007) exploit the variation in the timing and size of grants provided by the Community Oriented Policing Services Program in the United States to examine the relationship between police force size and crime. Each of the studies mentioned in this paragraph finds different magnitudes of the police size effect on crimes such as robbery, burglary, auto theft, and aggravated assault. Moreover, McCrary (2002) re-estimates the instrumental variables coefficients in Levitt (1997), which were not significant with a weighting correction.
The other strand of research on the relationship of police force sizes to crime is the exploitation of natural experiments relating to exogenous shocks to police presence in very particular settings, such as a response to terrorist attacks. Di Tella and Schargrodsky (2004) study the deployment of police officers in Jewish institutions after a terrorist attack on the main Jewish administrative building in Buenos Aires, Argentina. They find that car thefts fell by 75% in the blocks in which the protected institutions are situated. However, the effect is local, with no appreciable impact outside the narrow area where the police are deployed. A similar identification strategy was used by Draca et al. (2011) to analyze the deployment of police officers in London after the terror attacks of July 2005. They find that the extra presence of police reduces what they refer to as susceptible crimes (i.e., those that are more likely to be prevented by police visibility). Specifically, a 10% increase in police activity reduces susceptible crimes by 3 to 4%. Klick and Tabarrok (2005) use terror alert levels in Washington, DC to make inferences about the crime–police relationship. They show that an increase in police presence by about 50% leads to a statistically significant decrease in the level of crime on the order of 15%. Cohen and Ludwig (2003) exploit short-term variation in the intensity of police patrols by day of the week in several different patrol areas in Pittsburgh, Pennsylvania. They find that shootings were considerably lower in areas and on days that received more intensive police patrols.
In relation to the long-term consequences of patterns of police deployments, MacDonald et al. (2016) use a spatial regression discontinuity design to study the impact of especially intensive policing around the University of Pennsylvania. While areas adjacent to the university received police patrols from both the university and municipal police, areas slightly further away received only municipal police patrols. Results show that street crimes were substantially higher in the blocks just outside the area patrolled by the university police relative to the blocks just inside the university patrol area.
In sum, the review by Chalfin and McCrary (2017) highlights that the magnitude of the impact of the elasticity of police on crime varies between studies, and cross-crime elasticities generally favor a larger effect of police on violent crimes than on property crimes, with especially large effects of police presence on murder, robbery, and motor vehicle theft.
There is also literature on the effect of different policing strategies on crime. The literature on police deployments and tactics has focused predominantly on three types of interventions: hot spot policing, problem-oriented policing, and proactive policing.
Hot spot policing
Hot spot policing describes a strategy in which police are disproportionately deployed to areas in a city where there are disproportionate levels of crime. The first-order question that the hot spot policing literature seeks to address is the degree to which highly localized crime is responsive to a change in the intensity of policing. Various evaluations of such interventions suggest a statistically significant decrease in crime, non-significant spatial displacement, and some evidence of diffusion of benefits (Bowers et al. 2011). Sherman and Weisburd (1995), Telep et al. (2012) and Ratcliffe et al. (2011) exploit randomized control design where treated hot spots are exposed to more police patrols. Each study finds that the treated areas with more police presence experienced less calls for services relative to calls in control hot spots. Indeed, a review of the literature by Braga (2001) identified nine experiments or quasi-experiments involving hot spot policing and noted that seven of the nine studies, including a majority of the randomized experiments, found evidence of significant and large reductions in crime.
Most of the literature finds no evidence of displacement of crime to adjacent neighborhoods as a result of hot spot policing, and a number of studies find that the opposite is true—that there tends to be a diffusion of benefits to non-treated adjacent locations (Sherman and Rogan 1995; Braga 2007; Braga et al. 1999; Caeti et al. 1999).
Hot spot policing might be an even more important and effective strategy in Latin America, as it relies on deterrence rather than incapacitation, given the very low clearance rates for many crimes. Alves and Arias (2012) use a time series design to evaluate the Fico Vivo hot spot program that originally targeted a poor neighborhood in Belo Horizonte, Brazil with very high homicide rates. Results show that the program had significant success in reducing homicides.
Problem-oriented policing
A promising approach for dealing with crime hot spots is having officers who incorporate principles of problem-oriented policing, which involves organizing residents and property owners to help the police identify the sources of violent and property crime, and then targeting these problems with focused deterrence-based warnings to repeat offenders, more police, citizen, and technological monitoring, and better control of physical and social disorders (Nagin 2013). These activities usually involve joint action with police, prosecutors, and even the community.
One of the most well-known evaluations of a problem-oriented policing approach is that of Boston’s Operation Ceasefire by Kennedy et al. (2001). The stated purpose of Operation Ceasefire was to reduce youth gun violence in Boston. The intervention involved a multifaceted approach and included efforts to disrupt the supply of illegal weapons in Massachusetts. It also included messages communicated by police directly to gang members that authorities would use every available “lever” to punish gangs collectively for violent acts committed by individual gang members. In particular, the police indicated that the stringency of drug enforcement would hinge on the degree to which gangs used violence to settle business disputes. The result of the intervention was that youth violence fell considerably in Boston relative to other U.S. cities included in the study.
Since Operation Ceasefire, the strategy of “pulling every lever” has been the centerpiece of field interventions in many U.S. cities, including Richmond, Virginia, Chicago, Illinois, Stockton, California, High Point, North Carolina, and Pittsburgh, Pennsylvania (see Kennedy 2006). The focus of the approach, based on deterrence theory, is that the responses be certain, severe, and swift. A 2012 review of the literature by Braga and Weisburd suggests that the strategy of pulling every lever has been effective in reducing serious violent crime, with all reviewed studies finding negative point estimates, most of which were significant (Braga and Weisburd 2012).
With respect to individual evaluations, reductions in crime have been found across U.S. cities. For instance, Braga and Bond (2008) conducted a randomized controlled trial to examine the effects of problem-oriented policing strategies in reducing crime and disorder problems in hot spots in Lowell, Massachusetts. Only the treatment group received problem-oriented policing. The authors found that the treated hot spots experienced statistically significant reductions in total calls for service, as well as varying reductions in all subcategories of crimes, relative to controls. In turn, Taylor et al. (2011) conducted the first randomized experimental study that compared different treatments to hot spots in Jacksonville, Florida. Results showed there was no significant crime decline in the problem-oriented policing hot spots during the intervention period, although in the 90 days after the experiment, street violence declined by a statistically significant 33%.
In sum, evaluations of pulling all lever strategies show promising results. Nevertheless, inference is invariably complicated by a lack of randomized experiments and the inherent difficulty of identifying appropriate comparison cities and by the small sample sizes, among other problems. Therefore, caution is needed in assessing this literature and it would be important to explore these policies further in other regions such as Latin America and the Caribbean.
Proactive policing
Finally, proactive policing refers to the police tactics literature in criminology that investigates the responsiveness of crime to the intensity of policing, holding resources constant.
One strand of such police tactics is predictive policing, which investigates the use of information and advanced analysis to inform forward-thinking crime prevention. Predictive methods do not predict where and when the next crime will be committed. Rather, they predict the relative level of risk that a crime will be associated with a particular time and place (Perry et al. 2013).
The potential of predictive policing to reduce crime relies on the fact that crime does not randomly disperse across a geographic area. Instead, crime is clustered in particular areas that usually can be explained as a function of certain factors that create vulnerabilities for victims at certain places and times (Weisburd 2015; Chainey et al. 2008). According to previous studies in the developed world, crime tends to cluster geographically (Eck and Weisburd 1995; Evans and Herbert 1989; Felson 1987; Gil et al. 2017; Pierce et al. 1988; Weisburd and Green 1994), and crime indeed tends to be repeated in time and space (Johnson et al. 2007). A main finding on crime and place is what Weisburd (2015) posits as a general “law of crime concentration at place”.
Different theories have been posited for why concentration of criminal activity occurs. The recognition that the risk of crime increases because of identifiable factors has clear implications for predictive policing. If police can identify the location of potential crimes and understand the relevant factors for those predicted crimes, then they can focus their resources on those locations.
Although there has been a surge of interest in predictive policing in recent years, there is limited rigorous empirical evidence to date on whether it has more of an impact on crime than other policing strategies (Bennett et al. 2016). Current empirical evidence refers mainly to experiences with property crimes in the United States. Moreover, only a few studies have used rigorous methods to assess the impact of predictive policing on crime that can establish a causal link between the use of predictive policing technologies and changes in crime.
Regarding this line of study, Mohler et al. (2015) conducted two controlled randomized experiments using an epidemic-type aftershock sequence model (ETAS) that estimates the risk associated with both long-term hot spots and short-term models of near-repeat risk involving the Los Angeles Police Department and the Kent Police Department (United Kingdom). In the experiments, the ETAS algorithm (treatment) was compared with hot spot maps produced each day and each shift by dedicated crime analysts (control). Because hot spot locations dynamically changed each day, an experimental design was used in which days were randomly assigned to treatment or control. The experiment focused on burglary, car theft, burglary theft from a vehicle, criminal damage, violence against a person (including sexual offenses), and robbery. Police patrols using ETAS forecasts led to an average 7.4% reduction in crime volume as a function of patrol time, whereas patrols based on analyst predictions showed no significant effect.
On the other hand, Hunt, Saunders, and Hollywood (2014), using a blocked randomized approach, found no statistical evidence that crime was reduced more in experimental districts that used predictive policing models than in control districts that relied on traditional crime analysis. This experiment, conducted in the Shreveport Police Department in Louisiana, aimed to evaluate the policing strategy known as Pilot that developed a prediction and prevention model to reduce residential, auto-related, and business property crimes. Specifically, district pairs were randomly assigned to treatment and control groups. Treated districts were given maps that highlighted blocks predicted to be at higher risk of property crime. To generate predictions of future property crimes, the Shreveport Police Department performed multivariate logistic regression analysis for grid cells (400 × 400 feet) covering each treatment district. These districts were also provided with overtime resources to conduct special operations. Control districts also conducted property-crime-related special operations using overtime resources, but just targeting areas that had recently seen property crimes. Hence, the principal difference for control districts was that special operations targeted small areas where clusters of property crimes had already occurred—that is, hot spots were derived from conventional crime mapping techniques rather than from a predictive model. Results showed no statistical evidence that crime was reduced more in experimental districts than in control districts. However, several factors that might explain the overall null effect were identified, including low statistical power, program implementation failure, and program theory failure.
Recently, the shift from predicting and ranking “hot spots” to “hot people” has become a new focus for predictive policing (Ferguson 2016). For example, Saunders et al. (2016) evaluated the impact of a predictive policing program on gun violence based on a list of people estimated to be at the highest risk of gun violence, who were then referred for a preventive intervention. In addition, Mastrobuoni (2017) analyzed the effect of an offender-based software in Italy. Both find reductions on crime.
Using a quasi-experimental study, Kennedy et al. (2011) examined the use of risk terrain modeling, a predictive crime analytic approach. This study compares street segments and intersections that received police proactivity using results of risk terrain modeling with control segments derived from propensity score matching that did not receive extra police effort. The study found positive effects on crime rates. However, since control segments did not receive targeted patrols, it is still unanswered whether it was the technology or the targeted patrols that caused the reduction in crime.
The “law of crime concentration at place” also holds for a sample of Latin American cities. For example, Galiani and Jaitman (2016) analyze the performance of predictive policing in Uruguay. They conducted a randomized control trial to compare the predictive software with the status quo predictions developed by the police department and found no statistically significant differences for robberies between the police precincts randomly assigned to the crime analysts’ predictions and those assigned to the predictive policing software.
As can be noted, there are insufficient robust empirical studies to draw conclusions about the efficacy of crime prediction software. This has been recognized in the last report of the National Academies of Sciences, Engineering, and Medicine (2018). In addition, relatively little evidence-based knowledge exists about whether and to what extent predictive policing techniques will have crime prevention benefits at larger jurisdictional levels or across all offenders, and whether those benefits will continue over the long term (National Academies of Sciences, Engineering, and Medicine 2018). Thus, there is a need to more deeply understand how technology affects police agencies and, in turn, reduce crime. This is especially true for Latin America and the Caribbean, where, for instance, information software, such as predictive policing, is scarce but has a high penetration mainly from foreigner companies. Some countries in the region such as Chile are using predictive software developed in the country by researchers joint with the government (Baloian et al. 2017). It will be important to do research in their effectiveness to reduce crime in the region.
A second focus of the literature has been on the advent of what is called “broken-windows policing,” also known as “order maintenance” or “disorder” policing (Wilson and Kelling 1982). This strategy is based on strict enforcement by the police of laws governing relatively minor infractions, such as vandalism and turnstile jumping. Following the framework of Sect. 3, broken-windows policing operates primarily through perceptual deterrence. If offenders observe that police are especially vigilant, they may update their perceived probability of apprehension for a more serious crime, and accordingly they will decrease their participation in crime (\(p\)).
The literature that assesses the impact of broken-windows policing on crime has focused mainly on the case of New York City, which experienced the largest decline in crime among major U.S. cities. Most of the research presents identification problems, and although these problems can be set aside, it is unclear that this literature can isolate the impact of broken-windows policing from other changes that drove crime down in New York City (Chalfin and McCrary 2017).
Three studies that used especially strong research designs are worth mentioning. First, Braga et al. (1999) provide the first experimental evidence of a strategy aimed at addressing disorder. In Jersey City, New Jersey, 12 of 24 crime hot spots were randomly assigned to receive an intervention that involved broken-windows policing as well as other place-specific treatments that were intended to reduce crime. Results show significant crime reduction in treated locations. Second, Braga and Bond (2008) randomly assigned a general broken-windows policing strategy to 17 hot spots. They found strong reductions in crime in treated areas, but evidence of an effect of misdemeanor arrests was far more limited. Third, Caetano and Maheshri (2014) use an identification strategy that leverages highly detailed micro-data of all reported crimes and police response to these crimes in Dallas, Texas from 2000 to 2008. The identification strategy isolates the causal behavioral effect of prior crimes on future crimes and is robust to a variety of sources of endogeneity. They find no evidence of an effect of “zero tolerance” law enforcement policies on crime using micro-data from police precincts.
Finally, a systematic review conducted by Weisburd (2015) of studies that use experimental or quasi-experimental design or a before–after assessment of outcomes finds a slightly negative, albeit statistically not significant, impact.
In all, the evidence suggests mixed results. More research is needed, preferably based on experimental or quasi-experimental evidence, to support or refute any clear impact.
In sum, this section has reviewed a large literature on the responsiveness of crime to police size and tactics. Some key conclusions are worth noting. First, there is robust evidence that crime responds to increases in police manpower and to many varieties of police redeployments. However, it is necessary to further study the effect on long-term outcomes. It is necessary to better understand which interventions are cost-effective and how strategies can be maximized to improve the relationships between the police and the public, especially in Latin America and the Caribbean.
Incarceration and the severity of punishment
The prison rate in Latin America and the Caribbean is on the rise. This is clear in a simple comparison. In the United States, imprisonment increased 19% between 1995 and 2012, from 595 to 709 inmates per 100,000 population. In the same period, crime was significantly reduced: the homicide rate fell from 8 per 100,000 population to 5 per 100,000 population. In Latin America and the Caribbean, the opposite occurred. While the prison population also exponentially increased between 1995 and 2012 from 101.2 to 218.5 inmates per 100,000 population, an increase of 116%, crime also increased further during this period, with regional homicide rates doubling from 13 to 26 homicides per 100,000 inhabitants.
This section is focused on deterrence (i.e., emphasis on the negative consequences of delinquent behavior), incapacitation (i.e., emphasis on preventing at-risk individuals from becoming involved in delinquent activities), discipline (i.e., imposition of a rigorous regime to avoid re-offending), and rehabilitation (i.e., counseling, therapy, skill-building and re-entry programs to change the attitudes and behavior of prisoners).
According to the conceptual framework in Sect. 3, these interventions can be effective because they (1) raise the expected cost of lawless behavior by increasing the probability of authorities detecting such behavior (\(p\)), as well as the severity of the punishment once caught (\(f\)) that also enters through lengthier sanctions in the dynamic model; (2) change cognitive processes, as well as the capacity for self-control and empathy, reducing the expected gains of delinquency; and (3) make individuals more educated and employable, thus increasing the quantity and quality of legitimate opportunities and reducing the opportunity cost of deviant behavior (\(U^{\text{nc}}\)).
Deterrence
The effectiveness of deterrence has been repeatedly validated empirically. Under the setup of the model in Sect. 2, a dynamic perspective of deterrence is essential: if offenders have short time horizons, then it is hard to imagine punishment acting as an important deterrent. This is highly relevant because serious crimes are usually punished by long prison sentences, measured in years or even decades.
Evidence suggests that longer sentences seem to be effective in preventing crime. For instance, Philippe (2013) studies a shift in the instantaneous probability of re-offense when offenders are threatened with longer sentences and finds a decline in the hazard rate of re-offense of approximately 5%. Drago et al. (2009) assess the impact of an increase in sentence time in the context of the large Italian amnesty of 2006 and estimate that when expected sentences are increased by 25%, the propensity to re-offend in seven months decreases by approximately 18%. Bell et al. (2014) show that significant increases in sentencing severity, induced by the London riots of August 2011, led to a 13% decline in riot crimes in non-riot sub-wards. Lee and McCrary (2009), however, exploit the fact that young offenders are legally treated as adults (and face longer lengths of incarceration) the day they turn 18 and estimate an elasticity of crime with respect to sentence lengths of only about − 0.05.
The possible anti-crime effect of incarceration acts in two concrete ways: it can deter crime by making it less attractive because of the harsher sanctions (\(U^{\text{cp}}\)), and it can reduce crime through the “incapacitation” effect, since (in theory) incarcerated criminals are isolated from the illegal labor market.
Incapacitation
The effect of incapacitation on incarceration and recidivism is well documented in the literature. Owens (2009) exploits a 2001 change in Maryland’s sentencing guidelines that reduced the sentences of 23-, 24-, and 25-year olds with juvenile delinquent records by a mean of 222 days to estimate the effect of sentence length on recidivism. She shows that former delinquents between the ages of 23 and 25 would have been involved in 1.4–2.9 index crimes per person each year had they not been incarcerated. Vollaard (2013) exploits a legislative change that extends a criminal’s prison term tenfold in the Netherlands, and concludes that even when only 5% of the prison population was sentenced under the law 6 years after its introduction, the rate of theft declined on average by 25%, and by as much as 40%. Buonanno and Raphael (2013) in a different setting in Italy finds a sizable incapacitation effects mainly for the crimes of theft and robbery ranging from 17 to 21 crimes per prison year served. They exploit a natural experiment that affected incarceration. In August 2006, the Italian government released more than one-third of the nation’s prison inmates in an attempt to relieve prison overcrowding. The collective pardon did not change sentencing for future offenders while it enhanced sentences for pardoned offenders who reoffend. These changes likely induced a modest deterrent effect on criminal activity. Thus, the effects they find on crime associated with the pardon reflect a lower-bound incapacitation effect estimate.
Nevertheless, certain factors may favor the criminogenic effects of prisons: high overcrowding rates (occupancy on average is almost double the availability of places), deficiencies in rehabilitation services and re-insertion of prisoners (including the inability to analyze the level of risk posed by prisoners and treat them accordingly), and high rates of prisoners without conviction. For instance, Drago et al. (2011), based on variation in prison assignment, suggest that the prison environment is criminogenic; that is, harsh prison conditions actually increase post-release criminal activity in Italy. By exploiting a discontinuity in the assignment of federal prisoners to security levels in the United States, Chen and Shapiro (2007) arrive at the same conclusion. Gaes and Camp (2009) show that when offenders are placed into higher-than-necessary levels of security, they are more likely to have higher rates of recidivism than if they were placed at the appropriate security level. Finally, in a study of framing effects, by exploiting a legal change in Maryland that altered recommended but not actual sentences for a subset of offenders, Bushway and Owens (2013) find that criminals who receive a large reduction in their sentence may internalize the notion that the criminal justice system is lenient and be less deterred in the future.
Discipline
Programs based mainly on fear, shock, incarceration, punishment, or military discipline has surprisingly been the subject of negligible empirical research. What scarce literature does exist on this approach reveals no appreciable impact on recidivism.
The archetypical model for juvenile awareness is the Scared Straight intervention in the United States, which attempts to deter youths from delinquent behavior by terrifying them about prison life. In a review of studies that randomly assigned delinquents into control or intervention (Scare Straight) groups, Petrosino et al. (2003) show that in no case was there a decline in recidivism in the treated group. Furthermore, treated delinquents were in fact up to 28% more likely to recidivate than those in the control groups. However, the authors suggest taking these results with a degree of caution given potential methodological limitations. Aizer and Doyle (2015) find that imprisonment as a punishment for youths leads to a higher probability of adult incarceration.
Experimental impact evaluations of three boot camp programs in Cleveland, Ohio, Denver, Colorado, and Mobile, Alabama by the U.S. Department of Justice’s Office of Juvenile Justice and Delinquency Prevention found little support for this sort of intervention. While the recidivism rate in the control group was 50%, 72% of treated inmates in Cleveland’s boot camp recidivated. In Denver and Mobile, rates of recidivism were found to be comparable in the experimental and control groups (39% vs. 36%, and 28% vs. 31%, in the respective cities) (Peters 1996a, b; Thomas and Peters 1996). Based on a quasi-experimental design, Zhang (2000) evaluated a juvenile boot camp in Los Angeles that, unlike most military discipline interventions, includes an aftercare component combined with intensive supervision and counseling. The author shows that boot camp graduates exhibited almost identical recidivism outcomes to inmates in the comparison group. Furthermore, boot camp participants were more likely to have probation revocations. Wells et al. (2006) compared the recidivism of juveniles who completed a shock incarceration program that included a systematic aftercare phase with a matched comparison group of youths released from traditional residential placements. They found no differences in reconvictions or the seriousness of re-offenses at 8- or 12-month follow-ups.
Some governments in developed and developing countries have considered re-implementing military conscription as a policy mechanism to reduce violence among youth. However, exploiting the random assignment of young men to conscription in Argentina through a draft lottery, Galiani et al. (2011) show that compulsory enlistment into the armed forces actually increases the likelihood of developing a criminal record. The authors do not necessarily suggest this is the result of deleterious behavioral changes induced by extremely regimented activities, but rather they attribute this effect to reduced entry costs into crime caused by firearm training received during military service, as well as lost employment opportunities resulting from delayed insertion into the labor market.
Rehabilitation
Finally, rehabilitation for re-entry has become an essential mechanism for effective re-integration into society. Rigorous evidence shows promise for interventions offering support for offenders released to the community. For instance, Braga et al. (2009) evaluate the Boston Reentry Initiative, which offers comprehensive interventions (social, health, mentoring, and counseling services) to high-risk, violent criminals. They find that the program reduced recidivism rates by 30%, relative to a matched comparison group. Bloom et al. (2007) conduct a randomized evaluation of a large prisoner re-entry program and show that there is a modest, but statistically significant, decline in felony convictions and incarceration for new crimes during the first year of follow-up.
In line with this, one promising approach to improving prisoner re-entry outcomes is cognitive behavioral therapy. There are several well-identified studies measuring the effect of cognitive behavioral therapy on recidivism. In general terms, results are somewhat mixed, and may depend on both the intensity of treatment as well as the quality of the instructors. For instance, Pearson et al. (2016) conduct a randomized control trial to test the effectiveness of a program (known as “Citizenship”) in the United Kingdom that has a cognitive behavioral basis and focuses on education, promoting motivation to change, and community integration. They find that the program reduces convictions for high-risk probationers. Similarly, Barnes et al. (2017) use a randomized control trial to assess the impact of a cognitive behavioral therapy program on the recidivism of high-risk offenders. The analysis indicates that the overall cognitive behavioral therapy group was significantly less likely to reoffend, although this effect is concentrated in nonviolent offending. Heller et al. (2017) conduct three large-scale randomized control trial interventions to reduce crime and dropout rates by changing the decision-making of economically disadvantaged youth. They find support for the hypothesis that the programs work by helping youth slow down and reflect on thoughts and behaviors. Nevertheless, Bahr et al. (2016), who use a randomized control trial to evaluate a cognitive behavioral program that combines cognitive training, goal setting, and a phone-coach follow-up, find no effects on arrests.
Doleac (2018) reviewed the literature on re-entry programs in the U.S. and found that the programs that seem most promising are court-issued rehabilitation certificates, cognitive behavioral therapy, diversion from short incarceration spells, reducing intensity of community supervision, and expanding DNA databases. She found mixed evidence regarding the effectiveness of multisystemic therapy, and negative results on the effectiveness of transitional jobs programs, Ban the Box, and wrap-around services.
In summary, research has shown that deterrence and incapacitation are elements that help to reduce recidivism and overall crime. However, the fact that both homicide and incarceration rates in Latin America and the Caribbean have increased raises serious doubts about the proper functioning of these elements. Moreover, there are factors that favor the criminogenic effects of prisons in the region: high overcrowding rates, deficiencies in rehabilitation services and re-insertion of prisoners into society, and large numbers of prisoners without convictions. Empirical evidence supports the notion that re-entry programs and cognitive-behavioral therapy may be effective tools to reduce recidivism. Conversely, fear and punishment mechanisms are insignificantly effective. Doleac (2018) reviews re-entry programs in the U.S. and highlights the need for more evaluations to assess the effectiveness of the programs. This is even more pressing in countries in Latin America and the Caribbean where the prison population is growing exponentially and there is scant evidence on effective ways to reduce recidivism and achieve crime deterrence.
Social prevention of violence
Interventions for the social prevention of violence have been undertaken across multiple sectors. Evidence suggests that social and institutional interventions, including school attendance, self-control and conflict resolution, job training, and social interaction programs, can help mitigate risk factors. Under the framework of Sect. 3, the aim of these initiatives is to make legal activities more attractive among prospective criminals (\(U^{\text{nc}}\)) by increasing the cost of deviant behavior and reducing the gains from it, and hence reducing the overall return to deviant behavior.
Educational attainment and quality
There is quite strong evidence of a negative causal effect of educational attainment and quality on crime and incarceration rates. In a seminal paper, Lochner and Moretti (2004) find that a 1-year increase in the average education level reduces state arrest rates by 15% in the United States. Deming (2011) shows that attending a first-choice school leads to a 50% reduction in crime among high-risk youth, with defects lasting 7 years after random assignment, echoing previous evidence by Cullen et al. (2006) of reduced criminal activity. For the United Kingdom, Machin et al. (2011) find that a 1-year increase in average male educational attainment reduces incarceration rates by 20%. Bell et al. (2016) offer a reassessment of the relationship between crime, compulsory schooling laws, and education. They report evidence of a significant reduced-form relationship between crime and compulsory schooling laws. The same causal mechanism was studied by Hjalmarrson et al. (2015), who find that an additional year of education reduces the incarceration rate for men by almost 16% in Sweden. Through an incapacitation effect, Jacob and Lefgren (2003) show that the level of property crime committed by juveniles decreases by 14% on days when school is in session.
In Colombia, Klevens et al. (2009) indicate that a teacher-delivered intervention was an effective mechanism to reduce aggressive behavior. Berthelon and Kruger (2011) analyze the effect of a school reform that lengthened the school day from half- to full-day shifts in Chile and find that the reform reduced youth crime. Obach et al. (2011) assess the effect of educational workshops held for young men in public schools in Chile and show a significant decline in the acceptance of violence as a conflict-resolution mechanism. In Brazil, Pulerwitz et al. (2006) study the effect of community campaigns that target young men about violence against women and find significant positive changes in 10 of 17 gender-attitude items for the treatment group 6 months after implementation, with no changes exhibited among program non-participants.
Improvements in child development
Related to the education findings, improvement of child development—especially in social-cognitive dimensions—is found to be promising for reducing crime. For instance, after 2 years in the Perry Program, a randomized trial that targeted disadvantaged, low IQ African American children aged 3–4, all participants left the program and entered the same public school. Data were collected for treatment and control groups through age 40 show that the Perry Program significantly enhanced adult outcomes, including education, employment, earnings, marriage, and participation in healthy behaviors, and reduced participation in crime (Belfield et al. 2006; Heckman et al. 2010a, b). Finally, Olds (2008) summarizes a 30-year research program that has attempted to improve the health and development of mothers and infants and their future life prospects through the use of prenatal and infancy home visits by nurses. The program was tested in a randomized trial. Results show that nurse-visited children reported fewer sexual partners, fewer criminal convictions, and fewer violations of probation among those who had been involved in criminal activity.
Programs to help youth at risk
Progress has also been made with regard to youth at risk, which generally includes youths who do not study or work. Along these lines, Heller et al. (2013) report results from a large randomized control trial of a cognitive-behavioral therapy intervention for disadvantaged male youth from high-crime Chicago neighborhoods. They find that program participation reduced violent crime arrests during the program year by 44%, although the impact fades over time. Woolfenden, Williams, and Peat (2004) show that family and parenting interventions lead to a lower risk of being re-arrested. Piquero et al. (2009) provide evidence in support of parenting programs, indicating that treated young children who later got involved in criminal activity exhibited a 33% recidivism rate, while 50% of the children in the control group for that category relapsed into criminal behavior. Van der Stouwe et al. (2014) report more modest but significant treatment effects on delinquency, particularly among juveniles under the age of 15 who had started out in particularly difficult circumstances. In contrast, Littell et al. (2005) analyze eight randomized control trials in the United States, Canada, and Norway and find no statistical effect of multi-systemic therapy on arrests and convictions.
Skill programs and job-related interventions
Programs to remedy youth unemployment may also help to reduce delinquent behavior. Under the setup of Sect. 2, the mechanism through which this occurs increases in the expected utility of legal activities (\(U^{\text{nc}}\)). However, these types of interventions face the challenge of integrating unskilled workers into the job market, which is especially difficult considering the role of technological change. It has been argued that the adoption of new technologies is positively correlated with the relative demand for skilled workers, and it has been a major determinant of the rise in wage inequality. In view of that, the results of the adoption of new technologies for workers are mixed. For example, Mocan and Unel (2011) investigate the impact of unskilled workers’ earnings on crime using panel data and instrumental variables. Following the literature on wage inequality and skill-biased technological change, they find that technology-induced variations in unskilled workers’ earnings affect property crime with an elasticity of − 1, but that wages have no impact on violent crime.
Regarding the effectiveness of these types of programs, Schaeffer et al. (2014) use a randomized control trial to evaluate the capacity of a promising vocational and employment training program in the construction sector. The program was focused on juvenile offenders with substance use problems. However, the study does not report favorable effects on measures of criminal activity. Schochet et al. (2008) analyze a large vocationally focused education and training program for disadvantaged youths in the United States, and show that approximately 33% of controls were arrested during the 48-month follow-up period, compared to 29% of treatments. However, the authors conclude that, considering all measurable positive impacts, program costs are so high that they exceed program benefits for the full sample under most scenarios. On the other hand, there is convincing, albeit limited, evidence of the effectiveness of short-term and low-cost programs, such as summer jobs. Heller (2014) finds that assignment to a summer job program decreases violence by 43% over 16 months (3.95 fewer violent crime arrests per 100 youths). Gelber et al. (2014) study randomized lotteries for access to a summer youth employment program in New York City and find that participation in the program leads to a 10% reduction in the incarceration rate, relative to the baseline.
Prevention of violence against women
Violence prevention and care centers for women have been observed to reduce the likelihood of domestic violence. Randomized control trials in Washington, D.C. (Kiely et al. 2010), and Hong Kong (Tiwari et al. 2005) show significantly lower rates of violence revictimization among pregnant women who received psychosocial support, compared with women in control groups. Agüero (2013) examines the role of women’s care centers in Peru and shows that, in general, the likelihood of violence decreases with age, although the gradient is steeper for districts with women’s centers compared to those without them, providing empirical support for the theoretical model by Farmer and Tiefenthaler (1996).
In summary, educational attainment and quality are very effective crime-reducing mechanisms, and their effects are sustained over time. Cognitive behavioral programs appear to show the most favorable results. Improvement of child development—especially in social-cognitive dimensions—is promising for reducing crime. On the other hand, there is mixed evidence on the effectiveness of job-related interventions in reducing crime, as the challenge remains to introduce the role of skill-biased technological change into the design of these types of programs. Finally, violence prevention and care centers for women seem to be effective in reducing domestic violence.