Attachment_1.A_-_What_Works.pdf

“What Works” in Reducing Community Violence:

A Meta-Review and Field Study for the Northern Triangle

Thomas Abt, Christopher Winship

Submitted November 24, 2015

Abstract

Introduction Evidence-Informed Policy Violence Continuum

Methodology Systematic Meta-Review

Systematic Reviews and Meta-Analyses Eligibility Criteria Search Strategy Coding Strategy Potential Threats to Validity Results

Supplemental Review Field Study Analysis

Findings from the Meta-Review Place-Based Approaches People-Based Approaches Behavior-Based Approaches

Findings from the Field Study Discussion

What Works Focused Deterrence Cognitive Behavioral Therapy (CBT) Policing Rehabilitation Gangs Firearms Illegal Drugs Elements of Effectiveness

What is Missing The Concentration Principle The Coordination Principle The Implementation Imperative The Evaluation Imperative

Recommendations Conclusion

References Appendices

Abstract1 This reported was prepared for the Central American Regional Security Initiative (CARSI), a United States government initiative executed by the U.S. Agency for International Development (USAID) and the Bureau of International Narcotics and Law Enforcement Affairs (INL) at the U.S. Department of State. In preparation for this report, we performed a systematic meta-review of 43 reviews, including over 2,200 studies, to identify “what works” in reducing serious community violence. In addition, we supplemented our findings with fieldwork in El Salvador, Guatemala, Honduras, and the United States, visiting over 20 sites and conducting over 50 semi-structured interviews.

We found that a few interventions, such as focused deterrence and cognitive behavioral therapy, exhibited moderate to strong effects on crime and violence and were supported by substantial evidence. A few others, such as scared straight and gun buyback programs, clearly demonstrated no or negative effects. The vast majority of programmatic interventions, however, exhibited weak or modest effects. We identified six “elements of effectiveness” shared by the most impactful interventions, including maintaining a specific focus on those most at risk for violence;

proactive efforts to prevent violence before it occurs whenever possible; increasing the perceived and actual legitimacy of strategies and institutions; careful attention program implementation and fidelity; a well-defined and understood theory of change; and active engagement and partnership with critical stakeholders.

Given the modest effects of most interventions, that violence generally clusters around a small number of places, people, and behaviors, and that violence is not displaced from those clusters when they are targeted, we reach the simple yet powerful conclusion that it is advisable to concentrate and coordinate anti-violence efforts where they matter most. We further conclude that increased attention to program implementation and evaluation is necessary. We close with ten recommendations to assist CARSI in addressing serious community violence in the Northern Triangle.

Introduction Latin America has the unfortunate distinction of being the most violent, murderous region in the world, accounting for 9% of the world’s population but 33% of its homicides (Jaitman et al., 2015). The countries of the Northern Triangle – El Salvador, Guatemala, and Honduras – are among the most violent in the region. Violence, particularly homicide, imposes enormous social and economic burdens on the region. In the Northern Triangle, the costs are truly staggering, with one studying estimating the economic costs of violence for El Salvador, Guatemala, and Honduras at 7.5%, 5.4%, and 7.2% of gross domestic product respectively (Jaitman et al., 2015).

With an average homicide rate of 51 per 100,000 inhabits per year (Igarape, 2015), the Northern Triangle is in the midst of a violence epidemic according to international standards.

Despite the overwhelming urgency of the issue, the phenomenon of violence remains poorly understood in the Northern Triangle. The problem is not simply a lack of knowledge – although major gaps remain – it is that current knowledge, particularly evidence derived from rigorous research and evaluations– is not accessible to policymakers in the region in a readily usable and understandable format.

1 The authors thank Roberto Patino and Jason Wilks for their invaluable contributions to this report. Without them, this effort would not have been possible.

In this systematic meta-review, we summarize and analyze evidence concerning a vast array of programmatic interventions in order to better inform policies to reduce violence in the region.

Specifically, this report is based on three research components:

1. A systematic meta-review, or review of reviews, of systematic reviews and meta-analyses considering causal evidence relating to violence reduction;2

2. A supplemental review considering materials beyond the scope of the meta-review in order to enhance and provide further context for its findings; and

3. A field study employing semi-structured interviews and site visits to offer additional guidance regarding implementation and adaptation.3

This report may be the first of its kind – “To date, no meta-reviews have included the full range of programs that are intended to prevent youth violence; additionally, no meta-reviews have used both quantitative and qualitative approaches” (Matjasko et al., 2012).4 Our goal is provide policy recommendations informed by rigorous evidence but grounded in the practical realities of implementation in real-world settings.

Evidence-Informed Policy Evidence-informed policy, meaning policy informed by the best evidence and data currently available, has many advantages. The effective use of evidence and data enhances the accuracy, reliability, objectivity, consistency, and transparency of public decision-making. We avoid the term “evidence-based” deliberately, however, as a reminder that public policy can never be based exclusively on science. Evidence and data should be used to improve policy, not replace it entirely (Robinson & Abt, 2016).

In the U.S., both the supply and demand for evidence and data related to crime and violence has increased dramatically in recent years. In Latin America, however, and particularly in the Northern Triangle, the evidence-informed movement is still in its infancy. This report examines causal evidence, little of which exists outside high-income nations. This means that in many instances the recommendations of this report rely on evidence produced in settings quite different from those of the Northern Triangle. It is important to be mindful that in order “to provide effective policy, causal effects must be understood within a larger organizational, political, and social structure” (Sampson, Winship, & Knight, 2013). In this report we identify evidence-informed interventions and strategies, primarily from the U.S., that are worth exploring in the very different contexts of the Northern Triangle. This process of exploration should involve a careful and thoughtful process of adaptation and experimentation.

Our focus on causal evidence also creates a strong bias towards the programmatic interventions that are capable of generating such evidence. Evaluating institutions or systems is a much more complicated exercise where establishing causality may be difficult or even impossible. For this reason, only programmatic evaluations and evidence are discussed here, but it should be remembered that this is one of several important policy areas concerning crime and violence

2 Causal evidence is evidence that identifies a causal relationship between an intervention and its intended effect, and is typically generated by evaluations featuring an experimental or quasi-experimental research design.

3 Protocols for the systematic meta-review and field study are provided in Appendices A and C respectively.

4 Internal citations and quotations are omitted throughout this report.

reduction. For instance, as Manual Eisner (2015) recently argued, “An effective rule of law, based on legitimate law enforcement, victim protection, swift and fair adjudication, moderate punishment, and humane prisons is critical to sustainable reductions in lethal violence.”

Our focus on causal evidence is intended to provide us with the most reliable evidence of “what works” to reduce violence, but this does not mean that other evidence and information is not valuable. We simply believe that a better understanding of causal evidence, albeit mostly from high-income counties and programmatic evaluations, can significantly enhance decision-making with regard to violence reduction in the Northern Triangle.

Violence Continuum For the purposes of this report, we adopt a narrower version of the World Health Organization (WHO) definition violence as the intentional use of physical force or power, threatened or actual, against another person or group that results in physical injury or death (Krug et al., 2002).

We further limit our examination to what we term “serious community violence.” To understand what we mean by this, we first delineate six different dimensions of violent behavior. We then suggest that different forms of such behavior can be described along a rough continuum according to the associations between these dimensions.5 Violence comes in many forms. First, it can vary in its severity – a single punch versus a fatal shooting. Second, it can occur in different settings – in a home or on the street. Third, the number of individuals involved may be few, as with a dispute between neighbors, or many, as with conflicts among gangs. Fourth, it may be spontaneous as in a bar brawl or it may be planned as with an assassination. Fifth, it may be expressive of an emotion like anger or instrumental in pursuit of illegal economic activity. Sixth and finally, it may be common as domestic violence or uncommon as warfare.

Although these six dimensions are not perfectly associated with each other, they are strongly associated. For example, violence between family members is typically spontaneous whereas violence between nations is generally planned. To capture these associations we collapse them into a single dimension along a continuum as represented by Figure 1 on the next page.

Obviously, this continuum is neither entirely complete nor perfectly accurate, but we believe that viewing violence along a continuum is a helpful means of understanding difference kinds of violence while avoiding overly simplistic categories.

5 The World Health Organization, United Nations, and others divide forms of violence into discrete categories (Krug et al., 2002; UNODC, 2013). These schemes are often difficult to use in practice, as violence in the real world generally fails to comply with rigid classification.

Figure 1: Continuum of Violence

Figure 1 includes six forms of violence: violence among children due to bullying; violence between family members and/or intimate partners; violence among and between community members; violence committed by gang members; violence committed by organized criminal groups; and violence between nation states, i.e. war. These categories are intended to be illustrative, not exhaustive. Community violence is emphasized because it is the focus of this report and because it likely accounts for the largest number of homicides (see discussion below).

At one end of the continuum, violence is interpersonal, i.e. generally occurring between individuals known to one other. It is common but limited in severity, rarely fatal, and unlikely to cause permanent physical injury. It is unplanned, disorganized, emotional, and impulsive in nature. This violence is traditionally a private matter, occurring between friends, schoolmates, or family members. If addressed by public institutions, it will likely involve a wide array of public health stakeholders with limited law enforcement participation, if any. Bullying exemplifies the violence at this end of the continuum.

On the other end of the continuum, violence occurs between groups, often large in size, where individuals are generally unknown to one another. Unlike bullying, this violence is uncommon but severe, often resulting in significant numbers of casualties. It is highly planned, organized, and instrumental by nature. This violence is a generally state matter and traditionally the province of law enforcement and military institutions. Formally declared conflicts between states exemplify the violence at this end of the continuum.

In the middle of this continuum lies serious community violence, the focus of this report. Serious community violence, particularly homicide, occurs primarily in public settings. It is interpersonal, taking place between individuals and small groups that may or may not know one another. It is generally unplanned and impulsive in nature but its impact is nevertheless severe, often resulting in death or disabling injury. Its perpetrators and victims are generally, but not exclusively, young men from disadvantaged backgrounds and communities. It may result from disputes or from conventional forms of street crime, e.g. robberies. Serious community violence violence implicates both the public health and public safety fields and multi-disciplinary, multi-sector responses.

Other forms of violent behavior exist on the continuum alongside serious community violence.

Categorizations are fluid and allowed to overlap. Gang violence, for instance, if it occurs for impulsive reasons involving small, disorganized groups, would also be considered community violence. If the gangs involved are larger and more organized, with the reasons for violence more instrumental in nature, this violence would likely overlap with violence caused by organized crime.

It is generally difficult to disaggregate overall rates of violence into particular categories, and this is particularly so for the countries of the Northern Triangle. The Geneva Declaration Secretariat (2015) estimates that formal conflicts account for just 15% of all global homicides.

The United Nations Office on Drugs and Crime (UNODC, 2014) estimates that 30% of all homicides in Latin America are linked to organized crime or gangs, with high and low estimates varying wildly between approximately 43% and 7% respectively. The same report estimates that across the Americas, family and intimate partner violence accounts for 8.6% of all homicides.

Based on these estimates and others, we believe that serious community violence as defined in this report constitutes the largest single category of lethal violence, but no study has measured violence in the region in precisely these terms.

Further complicating matters, all forms of violence are interconnected. For example, a drug cartel “sicario” or assassin might shoot a public official during the day, stab a stranger during a dispute at a party later that evening, and then abuse his wife and children once he returns home, thus implicating organized criminal, community, and family violence respectively. The contagion between different forms of violence is a subject worthy of serious exploration but beyond the scope of this report.

While violence is clearly interrelated, it is equally clear that different forms of violence are best addressed by different strategies. A key component of any strategy is the number and type of partners to be mobilized. In this regard, responses to violence will differ greatly – a response to bullying may involve coalitions of educators and parents, while organized criminal violence typically demands the coordination of law enforcement groups.

In the Northern Triangle, at least three forms of violence bear special mention alongside serious community violence: family/intimate partner, gang, and organized criminal violence. Countries in Latin America, but especially the Northern Triangle, suffer from unusually high rates of all three. El Salvador in particular is currently facing crisis levels of lethal violence, driven primarily by violence between and among its largest and most organized gangs – MS 13 and 18th Street – and the government. As discussed later, strategies addressing these forms of violence should be aligned with, but distinct from, the strategies discussed in this report.

Methodology

Systematic Meta-Review

Systematic Reviews and Meta-Analyses

Policymakers and practitioners need tools to make use of evidence that is voluminous, diverse, and fragmented across disciplines. For research to be relevant to policy, it must be captured and consolidated in a reliable and accessible manner.

A systematic review uses predetermined and explicit methods to identify, select, appraise, and combine the results from individual studies in a clear, unbiased, and systematic manner. A meta-analysis combines quantitative data from individual studies using established statistical techniques. Systematic reviews may or may not include meta-analyses, and meta-analyses may or may not be systematic in nature (Petrosino and Lavenberg, 2007). This report uses a particular form of systematic review called a meta-review, which is a review of other reviews.

Systematic reviews and meta-analyses are designed to overcome limits to the external validity6 of individual studies (Killias & Villetaz, 2008). These techniques have grown popular as scholars recognize their usefulness for identifying themes and patterns across large numbers of empirical studies (Makarios & Pratt, 2012), and are being employed with increasing frequency in medicine, education, criminal justice, and many other fields (Aos et al., 2006). As Mark Lipsey noted in 1999, “Practice and policy, therefore, are best guided by a cumulation of research evidence sufficient to balance the idiosyncrasies of individual studies and support more robust conclusions than any single study can provide” (Lipsey, 1999). Ten years later, Lipsey (2009) elaborated further, “The most useful guidance for practitioners, and the most informative perspective for program developers and researchers, will not come from lists of the names of programs shown by research to have positive effects. Rather, they will come from identification of the factors that characterize the most effective programs and the general principles that characterize what works.”

Wherever possible, our meta-review follows the protocols and conventions of the Campbell Collaboration, the exemplar for systematic and meta-analytical reviews in the social sciences.

“Campbell reviews are intended to provide the most rigorous assessment possible of the effectiveness of studied interventions” (Koper & Mayo-Wilson 2006). The most important distinction between the methodology here and that of Campbell is the unit of analysis – the Campbell methodology compiles individual studies into a systematic review while our methodology compiles individual systematic and meta-analytic reviews into a meta-review, or review of reviews.

Eligibility Criteria In order to identify eligible reviews and analyses, we applied following criteria:

1. Reviews. Eligible reviews were systematic reviews and/or meta-analyses that synthesized the results of multiple studies. Individual studies were excluded.

6 Internal validity concerns the extent to which a causal conclusion based on a study is warranted. External validity concerns the extent to which the results of one study, even one with high internal validity, can be generalized to circumstances beyond the study’s original setting.

2. Outcomes. Eligible reviews reported on at least one or more of the following outcomes:

crime, violence, victimization, recidivism, and disorder. Reviews concerning specialized types of offenders or forms of violence not directly related to serious community violence (e.g. mentally ill offenders, sex offenders, family/intimate partner violence) were excluded.

3. Quality. Eligible reviews met minimum standards for methodological quality, scoring a six or higher on the AMSTAR7 eleven-point scale. Additionally, eligible reviews reported only on studies featuring causal evidence generated by experimental or quasi-experimental designs. Reviews with non-causal evidence from non-experimental studies were excluded.

4. Time period. Eligible reviews were published between January 1990 and July 2015.

5. Geography. Eligible publication locations were the Americas, the Caribbean, and Europe.

6. Languages. Eligible reviews were written in English and Spanish.

7. Sources. Both published and unpublished reviews were eligible for consideration.

Search Strategy We used multiple strategies to search exhaustively for reviews meeting the above-described criteria. First, keyword searches were performed on thirteen reputable English- and Spanish-language online criminal justice and public health databases. Second, hand searches of leading criminal justice and public health journals from both the United States and Latin America were performed. Third, a preliminary list of eligible studies was emailed to leading scholars knowledgeable in areas of crime and violence reduction in an effort to identify additional relevant studies. Fourth, we reviewed the “Practices” section of CrimeSolutions.gov for eligible reviews. Finally, we consulted an information specialist in order to ensure that appropriate search strategies were used to identify eligible reviews, including performing a grey literature search.8 Initial searches were performed in July 2015, with final searches performed in November 2015.

Coding Strategy A graduate research assistant in frequent consultation with Mr. Abt and Dr. Winship coded all eligible reviews on a variety of criteria. Mr. Abt reviewed each eligible study and determined final coding decisions in consultation with Dr. Winship.

Potential Threats to Validity Like all research methods, systematic reviews and meta-analysis are not without their limitations.

Potential criticisms include the “apples and oranges” threat of mixing dissimilar studies into a single analysis; the “file drawer” problem of selecting only published studies for analysis; and the “garbage in garbage out” challenge, where poor quality studies can compromise analysis (Hollin, 1999).

In this report, we take steps to anticipate and address all three criticisms. First, with regard to “apples and oranges,” we group similar interventions into categories, analyzing each separately to ensure some comparability. In addition, we forego a quantitative or meta-analysis of eligible reviews in favor of a more flexible narrative analysis. Second, we address the “file drawer”

7 AMSTAR (A Measurement Tool to Assess Systematic Reviews) is a widely accepted instrument used to assess the methodological quality of systematic reviews.

8 We are indebted to Phyllis Schultze for her assistance.

problem by exhaustively searching the grey literature9 for unpublished studies. Third, to address the “garbage in garbage out” challenge, we screen each review for methodological quality and accept only reviews with studies that rely on causal evidence generated by experimental or quasi-experimental designs.

A final threat concerns overlap in the reviews considered. More specifically, certain studies may appear in more than one review, limiting the independence of reviews and potentially confounding analysis. We believe the risk here is modest, given the variability of the eligible reviews and the flexibility of our narrative analysis, and where we believe it may impact our conclusions or those of the reader we discuss the issue explicitly.

Results

Our search strategies yielded 5,465 distinct abstracts for review. Of those, 188 were selected for full-text examination. After examination, 43 reviews were selected according to the eligibility criteria.10 These reviews included a total of 2,214 studies, with a mean of 51 studies per review, ranging from 3 to 700, examining 30 separate crime and violence control strategies. The mean AMSTAR rating was 8.8 on a scale from zero to eleven, indicating that most reviews were of good quality, with all reviews receiving a score of six or higher. 26% of the eligible reviews were unpublished, indicating that the search strategy was effective in addressing the threat of publication bias. Interestingly, only 47% of eligible reviews were published in the U.S.

Disappointingly, none were published in Latin American countries. Finally, only 12% of all interventions were classified as either intervention or secondary prevention strategies. The table below further describes the characteristics of the selected studies:

Table 1: Characteristics of Eligible Reviews

N %

Publication type

Journal article 33 77

Government/technical report 5 12

Dissertation/thesis 1 2

Campbell Collaboration review 4 9

Conference paper 1 2

Book chapter 1 2

Total published 34 79

Total unpublished 11 26

Year of publication

1990-1994 0 0

1995-1999 1 2

2000-2004 4 9

9 Grey literature is information or research published outside of traditional commercial or academic publishing and distribution channels.

10 A list of the eligible reviews is provided in Appendix B.

2005-2009 21 49

2010-2014 15 35

2015 onwards 4 9

Place of Publication

United States 20 47

United Kingdom 6 14

Netherlands 6 14

Germany 5 12

Norway 4 9

Canada 2 5

Spain 2 5

Language

English 43 100

Intervention focus

Place 9 21

People 22 51

Behavior 10 23

Intervention level11

Primary Prevention 10 21

Secondary Prevention 3 6

Tertiary Prevention 3 6

Suppression 12 25

Rehabilitation 20 42

Total Reviews 43 100

Supplemental Review In addition to eligible reviews, our searches identified a number of resources that, while not eligible for inclusion in our meta-review, were too useful to exclude entirely, including other relevant meta-reviews and important individual studies. When cited in the report, care is taken not to conflate these materials with eligible reviews, which were determinative in terms of our conclusions and recommendations. In legal terminology, the authority of supplemental materials was merely persuasive, not mandatory.

Field Study In order to provide additional guidance regarding implementation and adaptation, we selected and visited programs in the U.S. that were representative of leading interventions identified by

11 For the sake of clarity, we separate tertiary prevention from rehabilitation strategies, as we observe that combining them frequently causes confusion among practitioners in the field.

the meta-review. We conducted semi-structured interviews with intervention leaders, staff, and participants, as such formats are recommended when individuals are interviewed only once (Bernard, 1988). In each interview, we asked subjects a standard set of questions in order to identify the conceptual, operational, and contextual components of their intervention’s effectiveness. In total, we visited 11 sites in 5 states, conducting a total of 27 semi-structured interviews.

In addition, we travelled to El Salvador, Guatemala, and Honduras in order to better appreciate the context in which our findings, conclusions, and recommendations might be applied. In each country, we conducted semi-structured interviews with representatives from local government, law enforcement, non-governmental organizations, the faith-based community, and community representatives, along with USAID and INL leadership, management, and staff, among others.

We asked subjects a standard set of questions concerning the nature and concentration of local violence and the capacity of local institutions to address such violence. In total, we visited 11 sites across three countries, conducting a total of 24 semi-structured interviews.

Analysis This report summarizes the causal evidence produced by a systematic meta-review, then supplements and enhances those findings with additional research and fieldwork that may be non-causal and/or descriptive in nature. The benefit of this approach is that it enables us to provide guidance in the areas most lacking in the field of evidence-informed policy: synthesis, emphasis, practice, and adaptation. The risk associated with this approach is that if components are not carefully separated and remain independent from one another, it may not be clear to the reader how the authors reach their conclusions, and what evidence – causal, non-causal, or descriptive – they are relying on. In order to eliminate (or at least mitigate) these risks, we discuss our meta-review and fieldwork findings in separate sections.

It should be noted that a formal meta-analysis of the reviews produced by the systematic review was considered but ultimately rejected. In some cases it is possible to perform a meta-analysis aggregating individual meta-analyses, but here the wide scope of the review, which was a significant advantage in terms of comprehensiveness, produced a large number of reviews that were highly variable in nature. This made a quantitative synthesis of such reviews at best unreliable, at worst impossible. Differences among reviews included units of analysis, intervention delivery, comparison groups, and statistical techniques, among other dissimilarities.

Additionally, where we reach conclusions, the evidence is quite clear, minimizing the need for technical discussions concerning statistical power or to reconcile conflicting results (Koper and Mayo-Wilson 2006). In short, we agreed with others (Koper and Mayo-Wilson 2006, Gravel et al., 2012) who have decided against pooling together incommensurate studies.

Findings from the Meta-Review A key conclusion of this report is that the “available empirical and theoretical evidence suggests that crime is concentrated at a small number of high-risk places during high-risk times and generated by a small number of very risky people” (Braga, 2015). Further, “Crime should be analyzed in multiple units, or categories, including offenders, criminal networks, victims, micro-places (hot spots), communities, times, days of the week, and other categories” (Sherman, 2012).

Balancing accuracy and simplicity, we categorize and discuss interventions as either place, people, or behavior-based.

Place-based interventions target the geographic locations where violence occurs, people-based interventions focus on the individuals and groups who perpetrate violence, and behavior-based interventions focus on behavior that may be likely to trigger violence, such as carrying firearms, selling illegal drugs, excessive consumption of alcohol, and belonging to a gang, among others.

We believe a place/people/behavior-based framework is helpful to policymakers, who can quickly grasp that in order to reduce violence they must focus on where violence happens, who is involved, and how those involved are behaving.

In the tables below, we describe interventions by the strength of the evidence supporting them, the size of their impacts, relevant outcomes, intervention level, and sources. Descriptions of evidence strength are made relative to the other evidence identified in the meta-review.

Characterizations of intervention impact are drawn primarily from the conclusions of the review authors.

Outcomes for crime and violence were place-based in that they measured change in these indicators in a specific geographic region. Outcomes for recidivism measured individual changes in reoffending. Because place-based measures capture the changes in the offending of any individuals within a given area, they should be accorded more weight than individual measures in terms of assessing impact. For example, 15% reduction in crime in a given neighborhood is more substantial than a 15% reduction in recidivism of a specific individual living in the neighborhood.

With regard to program level, primary prevention interventions reduce risk behaviors associated with violence in the general population. Secondary and tertiary prevention reduce violent behavior in a population at risk for or already engaging in violence. Suppression interventions reduce violence via deterrence, generally but not exclusively using threats of arrest and incarceration. Rehabilitation interventions assist former offenders to reenter society after they have been imprisoned. These descriptions and labels are imperfect but have the benefit of being easily recognized by practitioners in the field.

Place-Based Approaches Our searches returned 9 eligible reviews discussing 7 discrete strategies for place-based crime and violence reduction.

Table 2: Place-Based Approaches Intervention Evidence Impact Outcomes Level Source(s)

Hot spots policing

Strong Modest Crime; violence

Suppression Braga et al., Disorder (broken windows)

Strong Modest Crime; violence

Suppression Braga et al., 2015; Distler, Community-oriented

Moderate None Crime; violence

Suppression Gill et al., 2015

Urban renewal Modest Modest Crime; violence

Primary prevention

Cassidy et al., CPTED Moderate Modest Crime; violence

Cassidy et al., 2014;

Farrington et al., 2007; Welsh and Farrington, 2009;

Farrington and Welsh, 2002

Neighborhood watch

Modest Modest Crime; violence

Bennett et al., Poverty deconcentration

Modest Moderate Crime, violence

Cassidy et al., Hot spots policing strategies focus on small geographic areas where crime is concentrated. Braga et al.’s Campbell review (2012), with 19 evaluations, 10 of which were randomized controlled trials, provided the strongest evidence of effectiveness and clearly established two points that are important for policy. First, focusing police efforts on micro-locations yields positive benefits.

Second, when hot spots are targeted, crime is not displaced. In fact, crime and violence are more likely to decrease for those living near or adjacent to hot spots enforcement. That said, the impact of these interventions was modest, with results ranging from no change to a 33% reduction in violence. Hot spots policing performed better when combined with problem-oriented policing.

Disorder or broken windows policing addresses physical and social disorder in neighborhoods in order to prevent crime and violence. With 9 of 28 studies using randomized controlled trials, Braga et al.’s Campbell review (2015) clearly establishes that broken windows is modestly effective. That said, certain approaches work better than others, with community and problem-oriented approaches preferred over aggressive “zero tolerance” strategies, which can create community tension and undermine collective efficacy.12 Distler’s (2012) review largely reinforced these findings, albeit with a smaller effect size.

Community policing leverages partnerships with residents and the community, often with a problem-solving approach, in order to reduce crime and disorder. Gill et al. (2015) analyzed 25 studies (only one of which was a randomized controlled trial) and found that community-oriented policing had no discernable impact on crime and violence, although it did have positive effects on citizen satisfaction, perceptions of disorder, and police legitimacy. This is consistent with previous findings by the prestigious National Research Council (Skogan and Frydl, 2004).

Urban renewal involves the improvement of various elements of the physical environment, including but not limited to transportation, housing, lighting, and vegetation. Cassidy et al.

(2014) examined five studies, only one of which featured randomized assignment. They found that urban renewal was associated with reduced crime and violence, especially in Medellín, Columbia, as well as improvements in police legitimacy and collective efficacy, but the number and quality of studies supporting this finding was limited.

12 Collective efficacy concerns the link between community cohesion and shared expectations for action, and is associated with reduced community violence (Sampson, 2004).

CPTED, or Crime Prevention Through Environmental Design, seeks to deter or prevent crime through the manipulation of the physical environment. Three reviews examined the effects of CPTED (Cassidy et al., 2014; Farrington et al., 2007; Welsh and Farrington, 2009; Farrington and Welsh, 2002) and found only modest, if any, impacts on crime and especially violence.

Additionally, certain CPTED elements can create a “fortress mentality” whereby residents withdraw behind walls, fences, and fortified homes, thereby undermining collective efficacy.

Neighborhood watch programs engage community residents to increase surveillance and deter or prevent crime. Bennett et al. (2006) reviewed the effects of neighborhood watch but found only modest impacts on crime and violence. Cassidy et al. (2014) found stronger effects for poverty deconcentration, either by encouraging the poor to move to wealthier areas or vice versa, but their conclusions were drawn from a sample of only four studies.

People-Based Approaches Our searches returned 23 eligible reviews identifying 14 discrete people-based strategies to reduce crime and violence.

Table 3: People-Based Approaches

Intervention Evidence Impact Outcomes Level Source(s)

Problem-oriented

Moderate Modest Crime; violence

Suppression Weisburd et al., Procedural justice

Modest Modest Recidivism Suppression Mazerolle et al., Focused deterrence

Moderate Strong Crime; violence

Suppression;

tertiary prevention

Braga & Weisburd, 2012

Cognitive behavioral therapy

Strong Strong Recidivism Secondary prevention;

rehabilitation

Lipsey et al., Vocational training

Mixed Modest Recidivism Secondary prevention;

primary prevention;

rehabilitation

Visher et al., 2005; Aos et al., Mentoring Modest Modest Recidivism Secondary prevention

Joliffe & Farrington, Juvenile curfews

Moderate None Crime; violence Primary prevention

Adams, 2003

School-based programs

Moderate None Violence Primary prevention

Alford & Derzon, 2012

RNR framework Strong Strong Recidivism Rehabilitation Lipsey, 1999;

Dowden et al., 2003; Dowden & Andrews, 2000; Garrido & Quinto, 2007;

Jolliffe & Farrington, 2007; Killias & Villetaz, 2008;

Smith et al.

Restorative justice

Moderate Modest Recidivism Rehabilitation Sherman et al., 2015; Lattimer et al., 2005

Aftercare programs

Moderate Modest Recidivism Rehabilitation James et al., 2013; Weaver & Campbell, 2015

Electronic monitoring

Strong None Recidivism Rehabilitation Renzema and Mayo-Wilson, 2005; Aos et al., Boot camp Strong None Recidivism Rehabilitation Mitchell et al., Scared straight Strong Negative Recidivism Rehabilitation Petrosino et al., Problem-oriented policing uses analysis to tailor police responses to specific public safety problems using the SARA (scanning, analysis, response, assessment) method. Despite widespread adoption, Weisburd et al. (2010) identified only 10 eligible studies, none of which were randomized controlled trials, showing a modest average impact on crime and violence.

Interestingly, problem-oriented policing appears to improve the performance of other policing strategies, such as hot spots, disorder, and community-oriented policing, and may have greater impact in a supporting rather than leading role in police strategies to reduce crime and violence.

Procedural justice seeks to improve police legitimacy and increase voluntary compliance with the law by providing greater transparency, demonstrating neutrality and trustworthiness, and showing respect for citizens, thereby increasing public confidence in the process by which policing is conducted. Mazerolle et al. (2013) reviewed 30 studies, 4 of which were randomized controlled trials, and demonstrated that such techniques enhance cooperation and satisfaction with police. Importantly, they also examined 15 studies that addressed whether procedural justice impact recidivism and revictimization. Here the results were more equivocal, with a very small but statistically significant reduction in reoffending.

Focused deterrence, as demonstrated by Braga and Weisburd (2012), has the largest direct impact on crime and violence, by far, of any intervention in this report. Focused deterrence involves the identification of specific offenders and offending groups, the mobilization of a diverse group of law enforcement, social services, and community stakeholders, the framing of a response using both sanctions and rewards, and direct, repeated communication with the individuals and groups in order to stop their violent behavior. While the authors urged caution in interpreting their results due to the absence of randomized controlled trials, they reported that 9 of 10 interventions substantially reduced crime and violence, with homicide reductions ranging from 34% to 63%. Since publication of the review, more studies have documented additional examples of focused deterrence success (Corsaro and Engel, 2015).

Cognitive behavioral therapy (CBT) uses clinical psychological techniques to alter the distorted thinking and behavior of criminal and juvenile offenders. Lipsey et al.’s Campbell review (2007) strongly reinforced, with 58 studies, 19 of which were randomized controlled trials, what numerous reviews before had established: CBT works. CBT has been effective in reducing recidivism of juvenile and adult offenders, in institutional or community settings, as part of a broader program or as a stand-alone intervention. No other intervention in this report can match its reliability and versatility. CBT was associated with a relatively large 25% average decrease in recidivism, but when the most effective types of CBT were used, recidivism declined 52%. The most positive results were not an extreme outlier – the authors reported that approximately 1 in 5 interventions studied produced such an effect or better.

Visher et al. (2005) and Aos et al. (2006) disagree as to whether vocational training has modest impacts on recidivism, reflecting a broader uncertainty in the field as to the effectiveness of stand-alone employment and training programs.

Jolliffe and Farrington (2007) conducted a “rapid evidence assessment” of 18 studies examining the impact mentoring on “at risk” youth. Overall, mentoring reduced subsequent offending by 4 to 11%, but this result was primarily driven by studies of lower methodological quality. Only 7 of the 18 studies showed statistically significant reductions, however, and mentoring was successful only when it was one of several interventions given. Given these results, the authors concluded, "Mentoring is a promising, but not proven intervention."

Alford and Derzon (2012), using an approach that admittedly may have been “overly conservative,” found only one school-based anti-violence intervention out of 24 eligible interventions that showed significant positive effects on more than one violence or anti-social behavioral outcome.

Recidivism reduction interventions that employ a risk/needs/responsivity framework (CBT is one such program) generate much stronger results than those that do not. This is well established by five reviews (Lipsey, 1999; Dowden et al., 2003; Dowden & Andrews, 2000; Smith et al. 2009) and others not included in this meta-review. An important additional finding is that these programs can also work as effectively in community settings as correctional ones (Killias & Villetaz, 2008). These treatments are effective (although less so) even with violent offenders, both adult and juvenile, particularly when cognitive or cognitive behavioral therapy is used (Garrido & Quinto, 2007; Jolliffe & Farrington, 2007).

Sherman et al. (2015) and Lattimer et al. (2005) both find moderate support for modest impacts with regard to restorative justice, whereby offenders and their victims are brought together to collectively address the aftermath of the offence. A significant limitation in the application of many restorative justice interventions is requiring the affirmative consent of both the victim and offender.

Weaver and Campbell (2015) find no evidence of impact for juvenile aftercare programs, which employ reintegrative services to prepare youth for reentry into their communities. However, the intervention appears to be more effective for older youth with more violent histories. James et al.

(2013) found a small impact of such programs when well implemented.

Renzema and Mayo-Wilson (2005) and Aos et al. (2006) agree that electronic monitoring in the U.S. has no impact on recidivism when compared to traditional community supervision, but Di

Tella and Schargrodsky (2013) have found that electronic monitoring in Buenos Aires reduced crime at less cost as compared with prison.

The evidence is clear that surveillance, deterrence, and discipline strategies are ineffective at best in reducing recidivism among youth (Mitchell et al., 2007; Petrosino et al., 2013; Adams, 2003).

While boots camps and juvenile curfews have no impacts, Scared Straight programs actually cause harm in that they are associated with modest increases in juvenile recidivism.

Behavior-Based Approaches Our searches returned 10 eligible reviews concerning 9 discrete strategies for behavior-based crime and violence reduction.

Table 4: Behavior-Based Approaches Approach Evidence Impact Outcome Level Studies

Drug enforcement

Moderate Negative Violence Suppression Mazerolle et al., 2006; Werb et al., 2011

Gun enforcement

Moderate Moderate Crime; violence

Suppression Koper & Mayo- Wilson, 2006;

Makarios & Pratt, 2012

Comprehensive gun violence reduction

Moderate Strong Crime; violence

Suppression;

tertiary prevention

Petrosino et al., Gang behavior regulation

Modest Moderate Crime; violence Tertiary prevention

Gravel et al., Drug courts and treatment

Strong Strong Recidivism Rehabilitation Holloway et al., 2006; Mitchell et al., 2012;

Mitchell et al., 2007; Shaffer, Gun buybacks Strong None Crime; violence Primary prevention

Makarios & Pratt, 2012

Gun laws Mixed Modest Crime; violence Primary prevention

Makarios & Pratt, 2012

Gang prevention Modest Modest Crime; violence Primary prevention

Gravel et al., Comprehensive gang reduction

Modest None Crime; violence Not applicable Gravel et al., 2012;

Hodgkinson et al., 2009

Two reviews (Koper and Mayo-Wilson, 2006; Makarios and Pratt, 2012) find that targeted gun enforcement is moderately effective in reducing gun crime and violence, while gun buybacks are ineffective and that the evidence supporting the effectiveness of gun laws is mixed at best.

Petrosino et. al (2015), found that 10 of 11 cross-sector, multi-agency interventions achieved significant reductions in gun violence, but as discussed further below most of these were focused deterrence initiatives.

Somewhat conversely, four reviews here (Holloway et al., 2006; Mitchell et al., 2012; Mitchell et al., 2007; Shaffer, 2006) reinforce the well-established finding that drug treatment and drug courts can significantly impact recidivism. Drug enforcement, however, appears to have limited impact (Mazerolle et al., 2006) and may actually increase violence by destabilizing drug markets and increase violence among drug sellers (Werb et al., 2011).

Gravel et al. (2012) conducted a wide-ranging review of gang control strategies, although, much like our work here, its diversity ultimately precluded a quantitative analysis of results. Based on a set of relatively relaxed criteria, the authors identified a set of 38 studies for review. Dividing these into categories, they found that gang prevention strategies showed little signs of effectiveness, which was explained by an overly broad approach to gang prevention that includes many youth who would never consider joining gangs. The authors also found that gang activity regulation strategies showed some signs of effectiveness, especially for focused deterrence interventions. Programs strictly focused on providing prosocial alternatives to gang members were less effective than those focused on preventing specific gang behaviors. Finally, the authors found that comprehensive and holistic strategies were not effective at reducing gang activity.

This finding was reinforced by Hodgkinson et al. (2009), who found that comprehensive gang intervention had only a small, non-statistically significant effect as compared to non-comprehensive strategies.

Findings from the Field Study We performed field research in order to supplement the findings of the systematic meta-review with additional practical guidance concerning program implementation and adaptation. In the U.S., we sought to identify “elements of effectiveness” that leading evidence-informed interventions shared. In the Northern Triangle, we focused on better understanding the nature of the violence challenge confronting each country along with the local capacity to adapt and implement evidence-informed solutions.

In the U.S., we conducted site visits and semi-structured interviews in Boston, Chicago, Providence, Los Angeles, and Philadelphia, examining five leading interventions including cognitive behavioral therapy, focused deterrence, hot spots policing, streetworkers and violence interruption, and the comprehensive approach to violence reduction. While it is difficult to generalize across interventions, we observed a number of shared characteristics that may prove useful for future adaptation and implementation.

First, each intervention examined made great efforts to identify and engage where the risk of violence was greatest.13 Dr. Gary Slutkin, leader and developer of Cure Violence, approaches violence as an infectious disease. He noted Cure Violence focused on the most infected, contagious individuals with regard to violence and that “tremendous specificity was required” in order to be successful (G. Slutkin, September 16, 2015).

Second, each intervention made extensive efforts to meet high-risk individuals “where they were at,” both literally and figuratively. Intensive outreach and follow-up was a hallmark of these programs, as those at the highest risk for violence were often disconnected from mainstream

13 Secondary prevention programs like Becoming A Man focused on at-risk youth, but not the highest risk youth.

institutions and society. In addition, they approach this cohort in a non-judgmental, compassionate manner, while still demanding accountability. Christopher Mallette, Executive Director of the Chicago Violence Reduction Strategy, noted that programs need “folks who can honestly engage these people in a sincere,…