R E S E A RC H A RT I C L E
F O C U S E D

D E T E R R E N C E
O R L E A N S

I N

N E W

Most Challenging of Contexts
Assessing the Impact of Focused Deterrence on Serious
Violence in New Orleans
Nicholas Corsaro
University of Cincinnati, Police Foundation

Robin S. Engel
University of Cincinnati

Research Summary
The use of focused deterrence to reduce lethal violence driven by gangs and groups of
chronic offenders has continued to expand since the initial Boston Ceasefire intervention
in the 1990s, where prior evaluations have shown relatively consistent promise in
terms of violence reduction. This study focuses on the capacity of focused deterrence
to impact lethal violence in a chronic and high-trajectory homicide setting: New
Orleans, Louisiana. Using a two-phase analytical design, our evaluation of the Group
Violence Reduction Strategy (GVRS) observed the following findings: (a) GVRS team
members in the City of New Orleans closely followed model implementation; (b)
homicides in New Orleans experienced a statistically significant reduction above and
beyond changes observed in comparable lethally violent cities; (c) the greatest changes
in targeted outcomes were observed in gang homicides, young Black male homicides,
and firearms violence; and (d) the decline in targeted violence corresponded with the

This project was funded by the New Orleans Community Support Foundation. Points of view or opinions in
this document are those of the authors and do not represent the official position or policies of the City of New
Orleans. The authors would like to thank Anthony Braga and the anonymous reviewers for their helpful
feedback and comments. We are also grateful to the New Orleans Police Department and the City of New
Orleans officials for their efforts and commitment to the project. Additionally, this project benefited from
considerable guidance from David Kennedy and the National Network for Safe Communities. We also thank
Murat Ozer, Jeffrey Clutter, Samantha Henderson, and Sam Peterson from the University of Cincinnati Institute
of Crime Science for their assistance with data collection. Direct correspondence to Nicholas Corsaro, School
of Criminal Justice, University of Cincinnati, P.O. Box 210389, Cincinnati, OH 45221 (e-mail:
nicholas.corsaro@uc.edu).

DOI:10.1111/1745-9133.12142

C 2015 American Society of Criminology
Criminology & Public Policy r Volume 14 r Issue 3

1

 Research Article

Focused Deterrence in New Orleans

implementation of the pulling levers notification meetings. Moreover, the observed
reduction in crime outcomes was not empirically associated with a complementary
violence-reduction strategy that was simultaneously implemented in a small geographic
area within the city.
Policy Implications
The findings presented in this article demonstrate that focused deterrence holds considerable promise as a violence prevention approach in urban contexts with persistent
histories of lethal violence, heightened disadvantage, and undermined police (and institutional) legitimacy. The development of a multiagency task force, combined with
unwavering political support from the highest levels of government within the city,
were likely linked to high programmatic fidelity. Organizationally, the development of
a program manager and intelligence analyst, along with the use of detailed problem
analyses and the integration of research, assisted the New Orleans working group in
identifying the highest risk groups of violent offenders to target for the GVRS notification sessions. The impacts on targeted violence were robust and consistent with the
timing of the intervention.

“(In 2004), researchers conducted an experiment in which police fired 700
blank rounds in a New Orleans neighborhood in a single afternoon. No one
reported the gunfire.” (Associated Press, 2005, para. 1)

T

he City of New Orleans has a rich history with unique cultural traditions (Hirsch
and Logsdon, 1992). Although perhaps best known for its blend of Creole, French,
and African cultures that influences its music, cuisine, activities, and celebrations,
it is also well known as a city of misfortune, ravished by Hurricane Katrina in 2005, enduring
decades of political corruption, police misconduct, poverty, and persistently high levels of
lethal violence (Moore, 2010). The city demonstrates both the best and the worst of urban
America, and it is where the most innovative criminal justice reform efforts are most needed
but are least likely to succeed. Within this challenging context, city and public officials
adopted a violence-reduction strategy to combat persistent patterns of lethal gun violence.
Since the early 1990s, the national urban homicide rate in the United States has hovered
between 13.0 and 20.0 per 100,000 in large U.S. cities with a population of 250,000 and
greater, and approximately between 8.0 and 10.0 in all U.S. cities with 100,000 or more
residents (Uniform Crime Reports, 2013). Based on Uniform Crime Reports homicide
data for the years 2008–2012, New Orleans averaged 55 homicides per 100,000 residents.
Additionally, since the mid-1970s, New Orleans has maintained a citywide homicide rate
trajectory that is significantly higher than 90% of all U.S. cities (McCall, Land, and Parker,
2011). Certainly, the underlying factors that contribute to lethal violence within New
2

Criminology & Public Policy

 Corsaro and Engel
Orleans have been lasting—and require an innovative approach to disrupt the stability in
the city’s persistently high homicide levels. As such, city officials implemented the Group
Violence Reduction Strategy (GVRS) to address the conditions and dynamics that generate
the ongoing criminal homicide problem.
The New Orleans GVRS drew on the focused deterrence framework (Kennedy, 1997,
2009a) and operated through interagency partnerships to impact persistent citywide patterns
in violence by using data-driven approaches (i.e., homicide incident reviews and gang audits)
to identify key offenders (specifically gangs and criminally active groups) responsible for a
disproportionate share of the city’s serious violence. Officials in New Orleans attempted to
increase the perceived risk of apprehension as well as the consequences leveraged against
gangs and criminally active groups through offender notification strategies backed up by
credible enforcement actions. The approach was specifically designed to halt ongoing violent
behavior by the most violent gangs and groups in the city. In the current study, we conduct
an empirical evaluation of the New Orleans GVRS first implemented by the City of New
Orleans and the New Orleans Police Department (NOPD) in 2012.
This article proceeds as follows: First, we review briefly the principles that guide
focused deterrence violence-reduction strategies and the body of evidence that demonstrates
their potential effectiveness. Second, we discuss the distinctive context of New Orleans by
outlining its unique challenges and describe the problem analyses used by the local criminal
justice working group to identify which gangs and groups drive lethal violence within the
city. We likewise document the specific tactics that were used to enhance the perceived
risk of future sanctions against such gangs and groups. Third, we present a two-phase
quasi-experimental design to examine the potential impact of the New Orleans GVRS.
Despite the challenging context presented in the City of New Orleans, the findings of
this study demonstrate that the focused deterrence initiative corresponded with statistically
significant reductions in targeted violence, which likewise corresponded with the timing of
the intervention. Finally, we conclude with a discussion of the importance of these findings
for both research and practice. This study adds to the growing body of literature that shows
focused deterrence has the potential to impact persistent violence in some of the most
challenging urban contexts.
Literature Review
A deterrence-based theoretical framework posits that three primary components deter highrisk individuals from engaging in future patterns of offending: the certainty, severity, and
celerity of punishment (Cook, 1980; Gibbs, 1975; Nagin, 1998; Paternoster, 1987; Zimring
and Hawkins, 1973). Based on an extensive review of both classic and contemporary studies
that assessed the various dimensions of deterrence, Nagin (2013) noted that there is little
evidence that severity-based deterrence approaches (e.g., life without parole or “three-strikes”
laws) are effective (see also Durlauf and Nagin, 2011); however, the evidence in support of
deterrent effects related to the certainty of punishment is far more consistent. Moreover,
Volume 14 r Issue 3

3

 Research Article

Focused Deterrence in New Orleans

Nagin contended that most compelling deterrent effects are seemingly linked to the police
generating an increased certainty of apprehension, which can be accomplished in one of two
ways:
1. By apprehending the suspect
2. By creating the perception that apprehension risk is sufficiently high
A promising deterrence-based crime prevention initiative, pioneered in Boston during
the 1990s, was the “pulling levers” violence-reduction strategy. The well-known Operation
Ceasefire initiative was developed in an effort to increase the risks of criminal justice
sanctions faced by active, chronic offenders (Braga, Kennedy, Waring, and Piehl, 2001;
Kennedy, 1997, 2009a). The cornerstone of the pulling levers focused deterrence strategy is
to communicate incentives and disincentives directly to targeted chronic violent offenders
to curb illicit and violent behavior and to obtain positive crime control gains. To change the
perception of apprehension risk, highly active and violent street gangs (and other criminally
active groups) are summoned by the police to “call-in” sessions to make them aware of the
specific penalties that will be leveraged against each individual associated with the group if
any member continues to engage in serious violence after the notification session.
Deterrence in these types of interventions is most likely achieved by identifying and
delivering a message of swiftness and certainty of apprehension (and punishment) to groups
of chronically violent offenders who are responsible for most of the city’s crime problems.
Ultimately, practitioners across various criminal justice agencies work together to convey
that violence will no longer be tolerated and that further violations will be followed with
legal, but harsh, sanctions available when future violence occurs (Kennedy, 1997, 2009a).
High-risk groups of offenders are susceptible to coordinated criminal justice responses such
as implementing strict probation and parole enforcement, shutting down otherwise nonviolent sources of income (e.g., gambling houses), paying attention to low-level street crimes
such as public intoxication, and ensuring direct and nonlenient prosecutorial attention.
Focused deterrence is rooted in the problem-oriented policing framework. This facilitates the customization of the focused deterrence approach to local conditions. Problemoriented policing is a highly focused law enforcement approach that is designed to assess,
recognize, and disrupt the underlying causes behind chronic crime problems (Goldstein,
1979). In a review of the most effective police approaches to crime prevention, the National
Academy of Sciences (2004) described the problem-oriented policing model as follows:
The heart of problem-oriented policing is that this concept calls on police to
analyze problems, which can include learning more about victims as well as
offenders, and to consider carefully why they came together where they did.
The interconnectedness of person, place, and seemingly unrelated events needs
to be examined and documented. Then police are to craft responses that may
go beyond traditional police practices. (National Research Council, 2004: 91)
4

Criminology & Public Policy

 Corsaro and Engel
What is perhaps most significant, from an organizational standpoint, is that the
problem-oriented policing model (including focused deterrence) is achieved by (a) diagnosing local problems, (b) using research to inform the problem analysis, (c) developing
interagency partnerships to address local crime problems, and (d) customizing the operational strategy locally to build long-term capacity. Indeed, the problem-oriented policing
model has shown considerable impact on targeted crime problems (National Research
Council, 2004; Weisburd, Telep, Hinkle, and Eck, 2010).
It is also worth noting that the notification sessions are held publicly in crime-stricken
communities (e.g., neighborhood churches) to illustrate a collective public response to the
violence (see Kennedy, 2009a). An emerging body of research has framed the use of offender
notification meetings as a way to enhance the perceived legitimacy of the criminal justice
system by providing an unbiased and procedurally just response to violence by involving
multiple members from the criminal justice agency, citizens, and clergy from high-risk
communities in the “deterrence-based” (i.e., threat of enhanced sanctions) notification
processes (see Brunson, Braga, Hureau, and Pegram, 2013; Papachristos, Meares, and
Fagan, 2007; Wallace, Papachristos, Meares, and Fagan, forthcoming). Thus, various sound
theoretical perspectives and organizational approaches guide focused deterrence in realworld settings.
Prior Evaluations of Focused Deterrence Strategies
In field settings that have focused on group and gang violence, reductions in the levels of
citywide and/or community violent crime, homicide, and gang-related (or group-related) violence have been observed in Boston (Braga et al., 2001; Piehl, Cooper, Braga, and Kennedy,
2003), Chicago (Papachristos et al., 2007), Cincinnati (Engel, Tillyer, and Corsaro, 2013),
Lowell (Braga, Pierce, McDevitt, Bond, and Cronin, 2008), Indianapolis (McGarrell, Chermak, Wilson, and Corsaro, 2006), and Los Angeles (Tita et al., 2004), and Stockton (Braga,
2008). Similar crime-prevention benefits have been documented in urban settings attempting to decrease drug-market–related neighborhood crime when drawing on the focused
deterrence model (Corsaro, 2013; Corsaro, Brunson, and McGarrell, 2010; Corsaro, Hunt,
Hipple, and McGarrell, 2012; Saunders, Lundberg, Braga, Ridgeway, and Miles, 2014).
The timing of the broader reduction in violence in each of these high-risk urban settings
typically corresponds with the onset of programmatic implementation. In a recently completed, Campbell systematic review and meta-analysis of focused deterrence programs, Braga
and Weisburd (2012: 341) reported that these interventions were associated with an overall
effect size on crime outcomes that was generally between .47 and .61, which is consistent
with a medium (or moderate) standardized effect size (Cohen, 1988).1

1.

The studies described here employed at least a quasi-experimental (sometimes with case-control
matching) design to rule out (where possible) extraneous influences on targeted crime outcomes
(Braga and Weisburd, 2012). Sherman et al. (1998) documented a process by which evaluation rigor can

Volume 14 r Issue 3

5

 Research Article

Focused Deterrence in New Orleans

More recent advancements in the literature have begun to illustrate the mechanisms by
which focused deterrence approaches potentially impact aggregate crime rates. Indeed if the
strategy is true to form, then groups and gangs who receive a threat of enhanced sanctions
combined with social service provisions should experience appreciable declines in offending
post implementation (i.e., post call-in). When the Boston Ceasefire strategy was reinstated in
the late 2000s after a period of discontinuation from the initial intervention (in the 1990s),
research found that gangs called into notification sessions were significantly less likely to
generate shootings and to be victimized by firearms when compared with highly comparable
(matched) control gangs (Braga, Hureau, and Papachristos, 2014). Additionally, gangs in
Boston that were associated with treated gangs (i.e., socially networked), but were not direct
recipients of the call-in sessions, also experienced significantly fewer shootings than matchcontrol gangs (Braga, Apel, and Welsh, 2013). A similar study in Chicago showed that gang
factions who attended a call-in experienced a 23% reduction in overall shooting and a 32%
reduction in firearm victimization in the year that followed the treatment (Papachristos and
Kirk, 2015, this issue).
In terms of individual-level studies, a cross-national implementation of the strategy
in Glasgow likewise showed evidence that gang-involved youths from treatment neighborhoods were significantly less violent than nontreated gang-involved youths from comparably
economically distressed areas (Williams, Currie, Linden, and Donnelly, 2014). Within the
United States, research (again from Chicago) showed that individuals who attended focused deterrence notification sessions had a lower recidivism rate, across multiple re-offense
categories, when compared with offenders from the same neighborhoods who were not
identified and selected to attend call-ins (Wallace et al., forthcoming). The literature has
thus illustrated that where focused deterrence strategies are implemented with a high degree
of fidelity, aggregate rates of violence often experience significant and sizeable declines; more
recent studies have highlighted that the crime reduction benefits are driven by reductions in
offending by the very groups and gangs as well as the individuals who receive the treatment
(both direct and indirect recipients of the notification sessions). Based on this research
foundation, problem analyses within New Orleans suggested that focused deterrence would
be a viable intervention strategy because detailed problem analyses illustrated that a discrete
number of groups and gangs was responsible for driving disproportionally high rates of
lethal violence within the city.

be measured, which is referred to as the “scientific methods scale” (or SMS) that ranges from 1 to 5 (low
to high). Studies with multiple units that receive or do not receive the program and attempt to control
for other factors (i.e., a classic quasi-experimental design) or that use time-series methods are coded
with a score of “3” in the SMS scale. More rigorous studies that use multiple cases and controls and
include matching are coded as a “4” on the SMS scale, whereas a score of “5” is reserved for randomized
controlled trials. Only evaluations that are graded with a score of “3” or higher are typically included in
rigorous systematic reviews (see also Braga and Weisburd, 2014).

6

Criminology & Public Policy

 Corsaro and Engel
GVRS in New Orleans
Practitioners often insist that the unique local context of their city, political environment,
violence problem, level of resources, neighborhoods, citizens, and so on would prevent the
successful implementation of violence-reduction initiatives that have demonstrated success
in other jurisdictions. New Orleans certainly presented such a challenge. The legitimacy of
the city government of New Orleans, and the NOPD specifically, has previously been called
into question (Moore, 2010). Unfortunately, prior allegations of corruption, misconduct,
and abuse of force have continued to plague the city. For example, in July 2014, former New
Orleans Mayor Ray Nagin was sentenced to 10 years in prison for various acts of bribery and
corruption while in office. Likewise, based on a 10-month investigation by the Department
of Justice (DOJ) and a subsequent scathing written report documenting unconstitutional
conduct by the NOPD, in 2012 the City of New Orleans, the NOPD, and the DOJ
entered into the “nation’s most expansive Consent Decree” in an effort to force sweeping
department-wide reform.2 Despite recent reform efforts, the perceived legitimacy of the
NOPD remains a challenge as scandals of police misconduct continue to be exposed (e.g.,
a damaging report released in November 2014 documented extensive failures in reporting
and investigations conducted by the NOPD Special Victims Section). These scandals led
to a continual questioning of the legitimacy of the NOPD (Tyler, 1990).
Despite such difficulties, between 2010 and 2012 specifically, government officials in
New Orleans including the Mayor’s Office, the NOPD, federal and local prosecutors, and
federal law enforcement drew on promising strategies such as the GVRS and the Project
Safe Neighborhoods (comprehensive gun violence-reduction approach—see McGarrell,
Corsaro, Hipple, and Bynum, 2010) to build the interorganizational capacity necessary for
strategic implementation. The city also supplemented the GVRS with the CURE Violence
model (formerly CeaseFire Chicago) as part of the broader NOLA for Life murder reduction
strategy (Skogan, Hartnett, Bump, and Dubois, 2009), which relied on violence interrupters
and outreach workers to mediate conflicts between conflicting groups within the Central
City area (City of New Orleans, 2013). Thus, the city relied on both a multiagency and
a comprehensive problem-solving framework to address the persistent citywide patterns in
violence.
As part of the GVRS problem identification phase (with a specific focus on the causes
and correlates of lethal violence), law enforcement officials in New Orleans partnered with
researchers to conduct a series of homicide incident reviews (the first beginning in June
2012) as well as gang audits to identify potential groups most prone to violence across
the different police districts within the city. The incident reviews included NOPD officers;
Bureau of Alcohol, Tobacco, Firearms, and Explosives (ATF) special agents; researchers
from the Institute of Crime Science at the University of Cincinnati; and members of the

2.

For more details, see nola.gov.

Volume 14 r Issue 3

7

 Research Article

Focused Deterrence in New Orleans

Cincinnati Police Department who had extensive experience with gang audits.3 Information
about violent street gangs was converted into actionable intelligence by organizing it along
the following dimensions: (a) individual gang members, (b) geography, (c) social networks,
and (d) participation in violence.
The working group identified 59 potential street gangs in six of the seven police districts
within the city. Officials estimated that there were approximately 600–700 individual
members within these gangs. Social network analyses indicated that a handful of gangs were
diverse and at risk for violence via their social networks (such as active feuds or alliances with
other gangs). Social network and geographic analyses supported officers’ descriptions of the
changing nature of groups and gangs in New Orleans. Officers described these groups as less
likely to be hierarchical, intergenerational, structured gangs, and more likely to be loosely
knit with continual changes in membership and affiliations. In addition, officers suggested
that the structure and territorial nature of violent groups and gangs changed dramatically
after Hurricane Katrina because of the displacement of low-income residents (see also Kirk,
2009) and the subsequent rebuilding phase, which separated group and gang members and
disrupted traditional territorial boundaries (where applicable).
Loose gang network structures and corresponding definitional issues have previously
served as a challenge within the City of New Orleans. For example, in a DOJ-sponsored
review of homicides from 2009 to 2010, Wellford, Bond, and Goodison (2011) found that
only 1.0% to 2.5% of all homicides in the city were officially classified as gang related.
Wellford et al. (2011: 12) cited cases in which the gang units specified that small groups
(three or four individuals) of unorganized young men often identified with geographic
areas where they lived committed high levels of violence within the city; however, because
they were not in “structured” and formalized “gangs” the NOPD at that time did not
define such activity as gang related. In their conclusion, Wellford et al. suggested the use
of homicide incident reviews would likely better unravel the network of loosely structured
groups of offenders. In the homicide reviews conducted in preparation for the GVRS, the
research team and NOPD officials placed a greater emphasis on identifying the loosely
affiliated networks of offenders engaging in violence. Based on this more in-depth and
comprehensive review, the current investigation identified 54.3% of all lethal incidents
between January 1, 2010 through March 31, 2014 as group or gang member involved

3.

The Cincinnati Initiative to Reduce Violence was a replication of the Boston Ceasefire strategy that
began in 2007. Key officials from the enforcement team from the Cincinnati Police Department worked
with the University of Cincinnati research team and the NOPD to explain their previous organizational
experiences with the Cincinnati Initiative to Reduce Violence (see Engel, Baker, Tillyer, Eck, and Dunham,
2008), conduct several gang audits, and lead their homicide incident reviews. Additionally, David
Kennedy and members from the National Network for Safe Communities played a key role by providing
project oversight and guidance throughout the various stages of the strategy to the NOPD and City of
New Orleans officials.

8

Criminology & Public Policy

 Corsaro and Engel
(GMI), indicating that the incident involved a group or gang member as a victim, suspect,
or both.4
As part of the GVRS, officers within the NOPD suggested that the structure of
groups and gangs in New Orleans was more fluid and less geographically based than in
the past. Similar descriptions have been observed regarding the changing nature of group
and gang affiliations in other cities as well (Engel et al., 2013; Kennedy, 2009b). Law
enforcement officials most familiar with these groups and gangs provided detailed feedback
to the criminal justice working group about gang participation in violence leading to a
continually updated list of potential gangs to include in the call-ins. The identified (mostly
loosely structured) gangs became the focus of a multipronged approach that included law
enforcement, the threat of enhanced prosecution, and the use of social services. During
the course of the strategy evaluated in this study, the NOPD conducted five separate
offender notification sessions to deliver antiviolence messages to offenders associated with
problematic gangs that were incarcerated or were on probation or parole (between October
2012 and March 2014). During these combined sessions, 158 individuals (representing
54 high-violence gangs) directly received communication that enhanced sanctions would
follow any involvement in violence, and the notified offenders were asked to disseminate
the message to other members. More specifically, the notified group and gang members
were warned that the next murder or shooting committed by any individual associated with
the notified gang would result in immediate and enhanced law enforcement scrutiny of the
entire group for any criminal activity. Illustrations of previous gangs that were apprehended
and facing rigid federal and state prison terms were shared to underscore the seriousness of
the message.
In terms of organizational structure, a multiagency law enforcement task force (including local and federal partners) was created to track gang violence, review data sources and
intelligence, and build criminal cases on violent gang members. Two newly created positions (i.e., program manager and criminal intelligence analyst) helped provide direction. In
addition to the 158 individuals who attended the call-ins, six individuals were visited by
police and received a personalized antiviolence message, which was referred to as a “custom notification” session (see Kennedy and Friedrich, 2014). Social service provisions were

4.

GMI homicides were classified by the NOPD and researchers during the homicide incident reviews (see
Azrael, Braga, and O’Brien, 2013). The name of the victim and suspect(s) (if known) as well as the totality
of situational homicide characteristics were considered. Such characteristics included the following:
location of the offense, suspected involvement of the victim in illicit acts preceding the homicide,
manner and type of death, known characteristics of the victim, time of day, likely suspects, and other
relevant characteristics of the incident. If the totality of the circumstances suggests that group members
were involved in the incident, then it was coded as a GMI homicide unless evidence existed to the
contrary. Each case was reviewed retrospectively for proper GMI determination when additional suspect
information was gathered. The same team of NOPD officers and researchers was responsible for the
final GMI classification of all homicides examined in these analyses; therefore, there are no concerns
regarding coder inter-rater reliability.

Volume 14 r Issue 3

9

 Research Article

Focused Deterrence in New Orleans

presented to those individuals in attendance at the call-in sessions; 59 of the 158 individuals
(37.3%) signed up for some type of social services, although only 25 of the 59 individuals
actually participated in or received such services.5
To summarize, the process (or implementation) of the GVRS strategy in New Orleans
is consistent with the overall model of the focused deterrence framework adopted in other
cities (Braga and Weisburd, 2012; Kennedy, 2009a). Specifically, high-risk groups and
gangs were identified through problem analyses, notified of future sanctions in call-in
sessions, subjected to enhanced enforcement actions when antiviolence rules were broken,
and provided access to social service opportunities.
Assessment of Impact: Two-Phase Analytical Approach
Assessing the potential impact of the New Orleans focused deterrence strategy presents
a unique challenge because the intervention was implemented throughout all areas in
the city that experienced high homicide and persistent gun-violence problems. Given the
political and social climate that led to programmatic implementation, prioritization was
placed on identifying and addressing citywide violence rather than on optimizing the
evaluation design. We thus employ a two-phase methodological design. First, we compare the
relative homicide rate change in New Orleans with cities that have been classified as having
the most volatile and stable homicide rates (see McCall et al., 2011). Next, we examine
changes in targeted violence within New Orleans by employing a standard interrupted
time-series design that compares multiple targeted crime outcomes (i.e., homicides, gun
homicides, gang homicides, and firearm assaults) in the postintervention period relative to
preintervention trends after controlling consistent shocks and drifts in the longitudinal data
to better isolate potential programmatic effects within the city (Cook and Campbell, 1979).
We also include a comparative time-series analysis on nontargeted outcomes (i.e., overall
violent crimes, property crimes, and nongang homicides) to assess whether the potential
changes in targeted outcomes correspond with a more general trend in crime within the
city. Finally, we include a series of sensitivity and placebo tests in both analytical phases and
control for the potential influence of simultaneous strategies (i.e., CURE Violence) within
New Orleans to rule out, where possible, the impact of confounding influences on the
outcomes examined.

5.

10

We note that the current study does not attempt to disentangle the various potential intervention
mechanisms. Although the literature has framed the focused deterrence framework under the umbrella
of “deterrence” (Kennedy, 2009a) as well “legitimacy” (Papachristos et al., 2007), it is possible that the
increased use of social service provisions by high-risk offenders could increase levels of institutional
engagement. It is important to consider that structural criminological research would suggest that
enhanced institutional engagement could potentially lead to reductions in homicide (see McCall, Land,
Dollar, and Parker, 2013). The Cincinnati GVRS evaluation did not find any significant relationship
between service provisions and changes in city-level violence (Engel et al., 2013: 28). However, parsing
out such mechanisms is both an analytical challenge and a vital next step in future evaluation research
where interventions are guided by various theoretical frameworks.

Criminology & Public Policy

 Corsaro and Engel
Phase I: Homicide Rate Change in New Orleans Contrasted with Similar
High-Trajectory Cities
Phase I is designed to assess whether New Orleans experienced a change in homicide
above and beyond cities with highly comparable homicide rates. The City of New Orleans
has maintained a persistently high rate of lethal violence since the 1970s (McCall et al.,
2011). Prior research has demonstrated the benefits of using group-based trajectory analysis
when attempting to achieve balance between cases and controls in observational (i.e.,
nonexperimental) settings (see Haviland and Nagin, 2005, 2007). Previous studies that have
classified geographic units such as street segments, neighborhoods, and cities into different
trajectory groups consistently illustrate that places with the highest levels of violence are also
responsible for the largest peaks and valleys (i.e., variability) in violence, whereas moderate
and lower trajectory classifications typically experience far fewer ebbs and flows in crime
(Braga, Papachristos, and Hureau, 2010; Griffiths and Chavez, 2004; McCall et al., 2011;
Weisburd, Bushway, Lum, and Yang, 2004).
As an initial step, we drew from work by McCall et al. (2011) that was based on a
long-term trajectory analysis of homicide rates in U.S. cities from 1976 to 2005. McCall
et al. identified 15 cities (including New Orleans) that had the most persistently high
homicide rates, similar structural factors that predicted group classification, and vastly
similar changes in homicide rates during a sustained period of time.6 By restricting our
comparison of changes in homicides in New Orleans with these previously validated hightrajectory homicide cities, we restricted our comparison to cities that are also more likely
to experience similar shifts in their homicide rates over time. The homicide data examined
in this analysis are for the years 2008 through 2013. We conducted a series of differencein-difference Poisson regression models (with an offset exposure variable accounting for
the annual population for each city—thus transforming the outcome into a homicide rate)
based on Equation (1):
log(Homicides)it = α + I(NewOrleans)it B1 + I(Treatment)it B2 + I(NewOrleans)it
× I(Treatment)it B3 + log Population

it

+ εit

(1)

where log(Homicides)it denotes the homicide count for each city between 2008 and 2013
(which is transformed into a homicide rate via the natural logarithm with the inclusion of the
population exposure variable on the right-hand side of the equation), I(New Orleans)it is an
indicator variable that equals 1 if the city is New Orleans and 0 otherwise, I(Treatment)it is

6.

The 15 cities that were classified as high-trajectory homicide rate cities (and thus were used as
treatment and comparison cities in subsequent trend comparisons) were as follows: Atlanta, Georgia;
Baltimore, Maryland; Birmingham, Alabama; Cleveland, Ohio; Dallas, Texas; Detroit, Michigan; Flint,
Michigan; Gary, Indiana; Miami, Florida; New Orleans, Louisiana; Newark, New Jersey; Oakland, California;
Richmond, Virginia; St. Louis, Missouri; and Washington, DC. All annual homicide data from 2008 to 2013
were obtained from the Federal Bureau of Investigation’s Uniform Crime Reports (2013).

Volume 14 r Issue 3

11

 Research Article

Focused Deterrence in New Orleans
TA B L E

1

Difference-in-Difference Model Estimates (New Orleans Compared with 14
High-Trajectory Homicide Cities)
Model 1
Treatment = 2013
b

Estimate

Intercept
New Orleans
Treatment
Difference-in-difference
**

(SE)
**

–8.276
.765**
–.012
–.266**

(.011)
(.034)
(.027)
(.090)

Model 2
Treatment = 2012
b

(SE)
**

–8.290
.786**
.035
–.191**

(.012)
(.039)
(.021)
(.068)

Model 3
Treatment = 2011
b

(SE)
**

–8.287
.767**
.018
–.088

(.014)
(.045)
(.020)
(.064)

p < .01.

an indicator variable that equals 1 if the year is in the posttreatment period and 0 otherwise,
and where + I(New Orleans)it × I(Treatment)it is the difference-in-difference estimator to
examine the direct impact of the change in homicides in New Orleans compared with other
highly chronic lethal violent U.S. cities.7
Model 1 in Table 1 illustrates that if the regression model corresponds with an intervention date that equals 2013 (i.e., the first full year that is in the true postintervention
period), then the City of New Orleans experienced a statistically significant homicide rate
decline above the average homicide rate change for the 14 highly comparable cities identified by McCall et al (2011). Specifically, the incident rate ratio is written as e–0.266 or
0.766, which equates to –.23, or a 23% homicide rate decline that was specific and unique
to New Orleans. However, true implementation began in late October 2012, and thus, we
alter the intervention period to include both 2012 and 2013 as the postimplementation
period (Model 2). Similarly, the results show a statistically significant decline unique to New
Orleans that was 17.3% lower (b = –.191, standard error [SE] = .068) when compared with
the homicide rate change in the 14 comparison sites. Although the magnitude of the effect
via the point estimate is reduced (from –23% in Model 1 to –17% in Model 2), the analysis
still shows a robust reduction in homicides that was unique for New Orleans in 2012–
2013. Finally, as a sensitivity test, we model the postimplementation period in the analysis
as 2011–2013 (Model 3), which thus includes a statistical implementation period one year
longer than the true postimplementation period. The difference-in-difference estimate for

7.

12

A key assumption of the difference-in-difference framework is the parallel trend assumption. Research
by McCall et al. (2011) highlighted that New Orleans and the other 14 high-trajectory cities have a long
criminological history with similar levels as well as comparable rates of homicide rate change.
Additionally, graphical analyses illustrated two consistent trends. First, New Orleans typically had
homicide rates that were on the high end of the distribution, and second, no noticeable differences
were observed in the shifts in homicide rates between New Orleans, as most comparison sites during
the preintervention period were examined in this study.

Criminology & Public Policy

 Corsaro and Engel
Model 3 was no longer statistically significant (b = –.088, SE = .064), which indicates
that the relative homicide rate change in New Orleans was specific only for years 2012 and
2013 (i.e., the true postintervention period in New Orleans). In sum, the homicide rate in
New Orleans experienced a decline that was unique when compared with other chronic,
high-trajectory homicide cities.
Supplemental analyses to test potential regression to the mean. Although the previous
analysis provides evidence that the homicide rate change was unique to New Orleans,
the highly comparable control sites were based on a classification using city crime and
structural data that ranged from 1976 to 2005 (McCall et al., 2011). To generate the closest
possible city-level homicide rate trajectories, we next follow a comparative analysis procedure
outlined by Haviland, Nagin, and Rosenbaum (2007: 250–251). Specifically, we modeled
homicide rate data from 2009 to 2011 for all U.S. cities (>100,000 population as of the
2010 U.S. Census) since this 3-year period immediately preceded the GVRS implemented
in New Orleans in late 2012. All group-based trajectory analyses (GBTAs) relied on the
Proc Traj procedure in SAS version 9.1 (Jones, Nagin, and Roeder, 2001; SAS Institute
Inc., Cary, NC). Latent growth curves were operationalized as annual homicide rates for
the years 2009–2011 by a set number of trajectories. Following the model identification
procedures outlined by Nagin (2005), linear models were found to be the most appropriate,
which was anticipated because only three observational periods were used to estimate the
trajectory models. The Bayesian information criterion (BIC) was also evaluated to select
the appropriate number of trajectory groups. The BIC can be viewed as an approximate
standardized model fit indicator because it penalizes when an increase in the number of
trajectory groups (k) is estimated.8 As shown in Table 2, most large U.S. cities were in
the low homicide trajectory group (n = 219, 81.4%), which averaged four homicides per
100,000 from 2009 to 2011. A total of 43 cities (15.9%) were classified in the moderate
trajectory group that had an average homicide rate of 16 per year. A small number of cities
(n = 7, 2.7%) was classified in the high-trajectory group for the years 2009–2011 (i.e., the
period that immediately preceded the 2012 New Orleans intervention), which averaged
39 homicides per 100,000. The cities in this trajectory include New Orleans along with
Detroit, Baltimore, St. Louis, Newark, Flint, and Richmond (California).
We conducted a sensitivity difference-in-difference regression analysis to compare the
shift in homicides in New Orleans with these six (immediate) high-trajectory cities. These
8.

The BIC values clearly indicated the three-group solution was the most appropriate for city homicide
rate trends. The three-group solution was identified as follows: BIC = –2341.48, where N = 269 (269
cities). In terms of model diagnostics, Nagin (2005) suggested that average posterior probabilities above
.70 are acceptable, whereas values closest to 1.0 are ideal. The average posterior probability of group
membership for each group was higher than 0.95 for each of the three groups identified. Nagin (2005:
89) also specifically indicated that the odds of a correction classification for each group should have an
estimated value above 5.0 to achieve high assignment accuracy (i.e., low residual deviation). The lowest
odds of a correction classification for any group were 18.93, which exceeds the acceptable diagnostic
criteria.

Volume 14 r Issue 3

13

 Research Article

Focused Deterrence in New Orleans
TA B L E

2

GBTA Model Estimates—Urban Homicide Rates (2009–2011)
Parameter

Low Group
(k = 1)

Moderate Group
(k = 2)

High Group
(k = 3)

0.989
4.58
4.30
4.30

0.959
16.53
16.41
16.31

1.000
39.45
37.13
40.25

Group probability
Homicide rate 2009
Homicide rate 2010
Homicide rate 2011

TA B L E

3

Difference-in-Difference Model Estimates (New Orleans Compared with Six
High-Trajectory Homicide Cities)
Model 1
Treatment = 2013
b

Estimate

Intercept
New Orleans
Treatment
Difference-in-difference
**

(SE)
**

–7.888
.415**
.057
–.375**

(.041)
(.082)
(.076)
(.102)

Model 2
Treatment = 2012
b

(SE)
**

–7.915
.459**
.109
–.314**

(.036)
(.096)
(.080)
(.149)

Model 3
Treatment = 2011
b

(SE)
**

–7.929
.475**
.104
–.241

(.031)
(.127)
(.068)
(.162)

p < .01.

six control sites were most comparable with New Orleans in terms of homicide rate levels
for the 3-year period that immediately preceded the implementation of the intervention
in 2012. The difference-in-difference findings displayed in Table 3 mirror those presented
previously; specifically, New Orleans experienced a statistically significant 31.2% homicide
rate divergence (b = –.375, SE = .102) when the intervention date in the model is set to
year 2013 (Model 1). Likewise, Model 2 shows that New Orleans had a 26.9% statistically
significant reduction (b = –.314, SE = .149) in homicides in the years 2012–2013 when
compared with the control cities. Finally, the difference-in-difference estimate was not
statistically significant in Model 3 (b = –.241, SE = .162), which indicates that the
homicide rate departure in New Orleans did not occur in 2011—or before the intervention
was actually implemented.
The combined evidence from both sets of models shows the homicide rate decline in
New Orleans was unique in two important respects. First, the relative homicide rate change
between the preintervention and postintervention periods in New Orleans was distinctive
when compared with cities that followed similar homicide rate trajectories. Second, the
multiple models used in this analysis illustrate the relative homicide rate change was greatest
for New Orleans in 2013 (the first full year in the postintervention period), was marginally
14

Criminology & Public Policy

 Corsaro and Engel
lower for New Orleans in 2012–2013 (the intervention was implemented in late 2012),
and did not seem to exist in 2011 (the year prior to the New Orleans intervention). These
combined results suggest that New Orleans experienced a distinctive change in homicides
that corresponded with the implementation of its citywide GVRS strategy.
Phase II: Time-Series Analyses on Homicides and Firearm Assaults within New Orleans
To assess the programmatic impact within New Orleans more fully, we move to an interrupted time-series analysis that accounts for unique changes in additional types of targeted
crime outcomes. The interrupted time-series design is appropriate when over-time data are
analyzed to assess the degree to which a treatment shifts the trajectory of a single case over
time (McCleary and Hay, 1980). Although any analytical design has limitations, time-series
modeling can be enhanced in many ways. In the context of the current study, Morgan and
Winship’s (2007: 245–252) review of the original Boston Ceasefire DOJ report that relied
on time-series analysis (see Braga et al., 2001) found the Boston evaluation was of “high
quality” largely because of the use of supplemental analyses, which included the examination
of multiple outcomes that were hypothesized to be influenced by the citywide intervention.
Morgan and Winship (2007) specifically noted that the use of multiple within-city outcomes improved the rigor of the time-series design, which can bolster the case for causal
assertion.
Research has suggested that the underlying dynamics that generate homicides, gun
homicides, and gun assaults are powerfully influenced by gang and group conflicts (e.g.,
Braga et al., 2014), which are the same dynamics targeted by the GVRS. Therefore, we
examined four key citywide monthly outcome variables as a way to triangulate possible
program impacts:
(1)
(2)
(3)
(4)

Overall homicides9
Firearm-related homicides
Firearm assaults10
GMI homicides

The various violent crime incident data examined here were provided by the NOPD
Crime Analysis Unit.11

9.

Homicides excluded police intervention shootings and those classified as self-defense or justifiable.

10.

A change in the data reporting structure by the NOPD did not allow for an analysis of earlier shooting
data prior to 2010; likewise, given the detailed nature of the GMI incident reviews, the research team
was not confident that retrospective classifications of homicide incidents that occurred prior to 2010
would yield the same level of measurement validity as those examined between January 1, 2010 and
March 31, 2014.

11.

Officially reported crime incident data are not without limitations, such as citizen reporting biases
(Warner and Pierce, 1993) and officer decisions to alter crime classifications and actions taken on

Volume 14 r Issue 3

15

 Research Article

Focused Deterrence in New Orleans
F I G U R E

1

Mean Monthly Difference in Targeted Crime Outcomes Between
Preintervention and Postintervention Periods

We use November 2012 as the period of implementation onset in our monthly preintervention and postintervention time-series comparisons given that the first pulling levers
call-in meeting occurred on October, 25, 2012, when approximately 40 probationers and
other high-risk prior offenders affiliated with gang networks were called into a session in
an Orleans Parish courtroom. Prior evaluations of focused deterrence indicate that the
strategy can have a “light-switch” (i.e., immediate and sustained) impact on homicide and
gang-related offenses (Kennedy, 2006). Therefore, the analyses presented in this article are
modeled to examine immediate and sustained changes in violent crime.12
Figure 1 shows the unconditional bivariate change in the various types of outcomes that
the focused deterrence strategy was intended to impact (as well as the non-GMI homicides,
which we use as a point of comparison). Specifically, we find that after implementation, the
mean monthly count of total homicides decreased from 15.2 pretest monthly incidents to
12.4 posttest monthly incidents or by 18.6%. A similar percentage reduction (17.4%) in
firearm-related incidents was observed during this same period—from 13.8 pretest mean

incidents reported to police (Black, 1970). Although it is important to consider the appropriate
limitations of official incident level data, police incident reports have been widely used to assess trends
and patterns of lethal gun violence (Blumstein and Rosenfeld, 1998) and previous focused deterrence
initiatives (Braga and Weisburd, 2012).
12.

16

Alternative specifications are examined in the sensitivity analyses.

Criminology & Public Policy

 Corsaro and Engel
monthly offenses to 11.4 posttest monthly incidents. The largest and most substantive
reduction was specifically observed for GMI homicides; a 30.1% decline was detected (a
change from 8.8 GMI homicides per month to 6.2 in the postintervention period). Comparatively, non-GMI incidents also declined but the reduction was much less pronounced
(–10.3%) between the preintervention and postintervention periods (6.9 average monthly
incidents to 6.2 postmonthly incidents). Finally, firearm assaults experienced a sizeable
reduction from 33.4 per month to 28.0 per month (–16.2%).
Figure 2 displays the monthly counts of homicides (all types) and firearm assaults, which
were distributed as event counts. These events do not approximate a normal, continuous
distribution (King, 1988). Because crime rates are discrete events that can suffer from
low counts during a specified time period, there are a variety of problems when analyzing
such data by using ordinary least-squares regression estimation (Osgood, 2000). Most
importantly, normal or symmetrical error distributions cannot be expected when crime
counts are small because the error distribution becomes rapidly skewed. A similar assumption
is required when using Autoregressive Integrated Moving Average time-series analysis (Box
and Jenkins, 1976).
The conventional analytical approach in criminology for analyzing event counts, and
in particular crime events, has been to rely on Poisson regression via maximum-likelihood
estimation (Long, 1997; Osgood, 2000). The Poisson distribution as it relates to this analysis
is expressed as follows:
P (Yi = y i  x i ) =

exp (−λ) λ y i
yi !

(2)

where Yi is a random variable representing a violent crime count (i.e., homicide event or
firearm assaults in this sample), yi is a count value that denotes the number of monthly
events observed for a discrete time period, and λi represents different values in violent crime
counts at distinct points (i.e., months) in time. To predict λi , we relied on the loglinear
function of the following model:
l n (λi ) = x iT β

(3)

where x iT β is a linear combination of predictors for each case (i). When estimating the
interrupted time-series models, this combination of measures included a postintervention
variable (1 = November 2012 onward), trend measures that account for a general decline
(i.e., fluctuations) in the time series, and monthly dummy variables to control for consistent
seasonal shifts in the data.
It is important to note the conditional Poisson process assumes equidispersion between
the expected mean and variance for the outcome variables modeled (Long, 1997). We
reestimated each Poisson regression model by relying on the conditional negative binomial
distribution because the overdispersion in the models that were estimated could lead to

Volume 14 r Issue 3

17

 18
2

Monthly Count of Targeted Crime Outcomes (Total Homicides, Firearm Homicides, GMI Homicides, and Firearm Assaults)
Preintervention and Postintervention

F I G U R E

Research Article

Criminology & Public Policy

Focused Deterrence in New Orleans

 Corsaro and Engel
TA B L E

4

Annual Changes in Homicides Relative to Changes in Overall Violent and
Property Offenses (2006–2012)
Offense Type

Homicide
Percentage change
Overall Violent Crime
Percentage change
Overall Property Crime
Percentage change

2006

2007

2008

2009

2010

2011

2012*

162
—
2,225
—
12,178
—

209
+29.01
3,451
+53.03
15,583
+27.96

179
–14.35
2,869
–16.68
14,880
–4.51

174
–2.79
2,614
–8.08
12,940
–13.04

175
+0.57
2,593
-0.83
12,645
–2.28

200
+14.28
2,748
+6.00
14,013
+10.81

161
2.42
2,240
–0.04
11,431
–1.06

*
The 2012 total only includes incidents between January 1, 2012 and December 31, 2012. The percentage change for 2012 is also
relative to the total number of incidents between January 1, 2011 and October 31, 2011. In 2011, 165 homicides, 2,241 overall violent
crimes, and 11,553 total property crimes occurred between January 1 and October 31.

biased statistical inferences (Hilbe, 2007; Osgood, 2000). In each outcome examined, we
used the Kolmogorov–Smirnov goodness-of-fit test and found that none of the distributions
examined were overdispersed. Thus, we display the results from the conventional Poisson
regressions with monthly fixed-effects parameters as a way to control for omitted static
influences on specified outcomes that were not included in our models (Allison and Waterman, 2002). We also include the Huber–White robust sandwich estimator to ensure the
coefficient variances were robust to violations of homoscedastic error distributions. STATA
12.0 SE software (StataCorp, College Station, TX) was used in all analyses presented herein.
However, as with any interrupted time-series estimation, we note that statistical inferences
in the subsequent analyses should be tempered because of the limitations of the design (see
McDowall, McCleary, Meidinger, and Hay, 1980: 7).
It is first necessary to assess whether potential changes in targeted crime outcomes,
such as homicides, occurred simply as a result of a broader fluctuation in crime within
New Orleans during the period of examination. Table 4 displays annual Uniform Crime
Reports crime counts from 2006 to 2012, which serves as a 7-year preintervention window
that illustrates the relative stability between annual changes in homicides with changes
in overall violent crime (homicides, rapes, robberies, and assaults combined) and changes
in property crime (burglaries, larcenies, and motor vehicle thefts). When examining
year-by-year variations, the annual changes in homicides followed extremely similar
patterns to changes in overall violent crime and property crime. The shifts (increases or
decreases) were virtually identical among homicides, overall violent crimes, and property
crimes in 6 of the 7 years from 2006 to 2012. In the year where there was a minor
degree of relative variation (2010), crimes of all types were stable relative to the earlier
baseline year (2009), with a difference less than two percentage points between each

Volume 14 r Issue 3

19

 Research Article

Focused Deterrence in New Orleans
TA B L E

5

Poisson Regression Results and Percentage Change Estimates on Targeted
Outcomes (January 1, 2008 to March 31, 2014)
Total Homicides
Coefficient
(St. Error)

Variable

Intervention
February
March
April
May
June
July
August
September
October
November
December
Intercept
*

*

–.191
(.085)
–.215
(.169)
.161
(.168)
.170
(.166)
.030
(.166)
.051
(.141)
.132
(.164)
–.120
(.203)
–.145
(.165)
–.158
(.207)
–.267
(.193)
–.017
(.199)
2.69**
(.152)

(IRR-1) × 100

–17.38%
19.50%
–16.99%
–18.29%
–2.63%
–4.70%
–13.31%
12.10%
13.32%
14.61%
23.73%
2.73%
—

Overall Violence
Coefficient
(St. Error)
**

.128
(.035)
–.045
(.102)
.086
(.090)
.075
(.067)
.100
(.079)
.023
(.066)
.085
(.077)
–.007
(.069)
–.055
(.086)
.002
(.072)
–.065
(.077)
–.041
(.067)
5.41**
(.063)

(IRR-1) × 100

13.65%
–4.40%
8.98%
7.88%
10.51%
2.32%
8.87%
–0.69%
–5.35%
0.20%
–6.29%
–4.01%
—

Overall Property
Coefficient
(St. Error)
**

.085
(.021)
–.151*
(.061)
–.020
(.041)
–.020
(.040)
.125*
(.049)
.070
(.045)
.136*
(.031)
.096*
(.032)
.010
(.046)
.016
(.034)
–.051
(.040)
.027
(.043)
6.48**
(.049)

(IRR-1) × 100

8.87%
–14.01%
–1.98%
–1.98%
13.31%
7.25%
14.65%
10.08%
1.00%
1.61%
–4.97%
2.73%
—

p < .05. ** p < .01.

outcome examined. In short, from 2006 through almost all of 2012, homicides, overall
violent crimes, and property crimes followed extremely similar patterns of change in New
Orleans.
Multivariate Results
Moving to the multivariate framework, Table 5 presents the results of the interrupted
time-series analyses that examine the impact of the GVRS on homicides while controlling
for other important covariates (i.e., seasonal monthly shocks). The targeted gang and
criminally active group members that were called into notification sessions were identified
20

Criminology & Public Policy

 Corsaro and Engel
based on their participation in gun violence; thus, it was anticipated that the strategy
would have the greatest impact on lethal violence. Comparative estimates are also
provided for changes in overall violent crimes and property crimes, which were beyond
the specific scope of the focused deterrence strategy. Unstandardized coefficients and
standard errors are presented within the table, along with the estimated percentage change
in the monthly incidents, which is expressed as incident rate ratios (IRRs). The IRRs
are exponentiated coefficients given the use of the logarithmic transformation in the
Poisson regression models and percentage changes ([IRR – 1] × 100) in the intervention
estimates.
The findings presented in Table 5 indicate that the GVRS was associated with a statistically significant 17.4% decline (b = –.191, SE = .085) in the mean monthly number
of homicides in the postintervention period relative to the preintervention period. Additionally, none of the monthly dummy variables were statistically significant, which suggests
relative seasonal stability in homicide incidents in New Orleans. Comparatively, both overall
violence and overall property crimes experienced statistically significant increases during the
same period that homicides experienced a significant decline. A significant 13.6% increase
in violent crimes and an 8.8% increase in property crimes occurred after November 2012.
These findings, combined with the earlier year-by-year trends (displayed in Table 4), suggest
that the New Orleans strategy had a crime prevention impact that was unique to homicides
that occurred beyond chance, whereas overall Part I crimes significantly increased during
the postintervention period.
Table 6 provides more precise details on the types of homicides and firearm-related
outcomes that were impacted by the GVRS. Several prior focused deterrence evaluations
showcase a consistent effect: The greatest and most sizeable decreases in homicides are
those that are group and gang related, which is expected because the most chronically
violent gangs are the specific target of the multiagency task force (see Braga et al., 2008;
Corsaro and McGarrell, 2009; Engel et al., 2013). In New Orleans, GMI homicides
significantly reduced by 32.1% in the postintervention period (b = –.387, SE = .116)—
net of seasonal controls. Comparatively, non-GMI homicide incidents (i.e., lethally violent incidents that were considerably more likely to be domestically related and involve
nonchronic offenders) did not experience a statically significant mean difference. Specifically, an 8.9% nonsignificant decline occurred in non-GMI homicides (b = –.094, SE
= .093). When compared with a significant reduction in GMI homicides by 32.0% in
the postintervention period, these results highlight that the driving force behind the overall homicide decline was the specific reduction in lethal violence that was group or gang
involved.
Table 6 also illustrates that both lethal and nonlethal firearm-related incidents
experienced similar changes that corresponded with the New Orleans GVRS. The mean
number of monthly firearm-related homicides significantly declined by 16.3% (b = –.178,
SE = .092). Likewise, the mean monthly number of nonlethal firearm assaults significantly
Volume 14 r Issue 3

21

 Research Article

Focused Deterrence in New Orleans
TA B L E

6

Poisson Regression Results and Percentage Change Estimates on Targeted and
Comparison Outcomes
GMI Homicidesa
Variable

Intervention
February
March
April
May
June
July
August
September
October
November
December
Intercept

Non-GMI Homicidesa

Firearm Homicidesa

Firearm Assaultsb

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

–.387**
(.116)
–.120
(.260)
–.023
(.220)
–.149
(.187)
–.180
(.240)
–.247
(.239)
–.003
(.278)
–.570*
(.257)
–.088
(.231)
–.213
(.249)
–.304
(.250)
.165
(.218)
2.31**
(.154)

–32.09%

–.094
(.093)
–.575**
(.216)
.446*
(.230)
.239
(.167)
–.121
(.243)
–.121
(.216)
.381**
(.162)
–.037
(.160)
–.121
(.243)
.001
(.335)
–.055
(.265)
.024
(.282)
1.89
(.159)

–8.97%

–.178*
(.092)
–.165
(.210)
.169
(.205)
.217
(.203)
.031
(.201)
.100
(.175)
.144
(.197)
–.168
(.219)
–.097
(.199)
–.153
(.247)
–.248
(.222)
.022
(.231)
2.62**
(.173)

–16.30%

–.177**
(.067)
–.136
(.117)
–.033
(.112)
.021
(.177)
–.097
(.146)
–.207
(.110)
–.224
(.146)
–.207
(.145)
–.495**
(.124)
.005
(.117)
–.173
(.183)
–.327*
(.157)
3.65**
(.099)

–16.22%

–11.30%
–2.27%
–13.84%
–16.47%
–21.88%
–0.29%
–43.44%
–8.42%
–19.18%
–26.21%
17.93%

–43.72%
56.20%
26.99%
–11.39%
–11.39%
46.37%
–3.63%
–11.39%
0.10%
–5.35%
2.43%

–15.21%
18.14%
24.23%
3.14%
10.51%
15.48%
–15.46%
9.24%
14.18%
–21.96%
2.24%

–12.71%
–3.24%
2.12%
–9.24%
–18.69%
–20.06%
–18.69%
–39.04%
0.50%
–15.88%
–27.39%

a

Time series: January 1, 2008 to March 31, 2014.
Time series: January 1, 2010 to March 31, 2014.
*
p < .05. ** p < .01.
b

declined by 16.2% (b = –.177, SE = .067). Firearm assaults seemed to have more
seasonal fluctuations (particularly during the late summer and fall months) than all other
targeted offenses examined, as evidenced by the significance levels of the monthly dummy
variables.
In an effort to assess whether changes in homicides were observed for the most “at-risk”
groups for homicide victimization and to minimize concerns related to gang “definitional
issues” (see Maxson, 1999), a series of time-series analyses is presented and distinguished
by victim age and race demographics (given that such demographic classifications are not
22

Criminology & Public Policy

 Corsaro and Engel
TA B L E

7

Poisson Regression Results and Percentage Change Estimates on Race and
Age-Specific Homicides (January 1, 2010 to March 31, 2014)
Black Male Victims
20–29 Years Old
Variable

Intervention
February
March
April
May
June
July
August
September
October
November
December
Intercept
*

Black Male Victims
30+ Years Old

All Other Victims
20–29 Years Old

All Other Victims
30+ Years Old

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

–.311**
(.104)
–.234
(.243)
–.097
(.219)
–.030
(.277)
–.085
(.212)
–.030
(.238)
.072
(.210)
–.207
(.299)
–.273
(.272)
–.239
(.282)
–.225
(.252)
–.159
(.253)
1.89**
(.188)

–26.72%

.100
(.129)
.213
(.360)
.593
(.342)
.380
(.348)
.213
(.381)
.300
(.360)
.618
(.375)
.257
(.376)
.117
(.424)
.166
(.377)
.100
(.384)
.240
(.429)
1.06*
(.341)

10.51%

–.012
(.337)
−2.079*
(.990)
–.287
(.549)
.558
(.419)
–.317
(.463)
–.135
(.395)
–.540
(.567)
−1.233
(.661)
–.828
(.521)
−1.926*
(.977)
−1.924
(.977)
–.272
(.534)
.136
(.317)

–1.19%

–.388
(.290)
-1.466*
(.503)
–.262
(.353)
–.842*
(.385)
.111
(.326)
–.506
(.435)
–.255
(.421)
–.373
(.362)
–.506
(.356)
–.054
(.454)
–.784
(.487)
–.431
(.411)
.715
(.224)

–32.15%

–20.86%
–9.24%
–2.95%
–8.15%
–2.95%
7.46%
–18.69%
–23.89%
–21.25%
–20.15%
–14.70%

23.73%
60.94%
46.22%
23.78%
34.95%
65.52%
29.30%
12.41%
18.05%
10.51%
27.12%

–87.49%
–24.94%
74.71%
–27.16%
–12.63%
–41.73%
–70.85%
–56.30%
–85.42%
–85.39%
–23.81%

–76.91%
–23.04%
–56.91%
11.73%
–39.70%
–22.50%
–31.13%
–39.71%
–5.26%
–54.34%
–35.01%

p < .05. ** p < .01.

subject to misclassification). Table 7 shows that homicides involving Black male victims
between the ages of 20 and 29 years old experienced a statistically significant decline of
26.7% (b = –.311, SE = .104). This observed intervention effect was found only for
victims with this race and age classification. No significant decline occurred in homicides
that involved older Black male victims (30 years old or older), and there were no significant
changes in homicides that included pooled White and Hispanic male and female victims,
or Black female victims (i.e., all other victims) between 20 and 29, as well as 30 years old or
Volume 14 r Issue 3

23

 Research Article

Focused Deterrence in New Orleans

older. Thus, the lone observed intervention effect among New Orleans homicides was for
Black males between the ages of 20 and 29 years old.
Taken together, the analyses presented in this section consistently indicate that homicides experienced significant declines and that overall violence and overall property crimes
experienced increases. Comparing trends among specific types of homicides, the observed
significant reductions were specific only to GMI and firearm homicides, whereas non-GMI
homicides (i.e., those that did not involve group and gang members) remained relatively stable during the period examined here. Finally, homicides that involved Black males between
the ages of 20 and 29 years old were the homicides that experienced significant changes.
These results indicate that targeted violent crime incidents had observed significant declines,
whereas there was no evidence of a general reduction in overall crime or lethal violence that
did not involve group or gang members.
Sensitivity and supplemental time-series analyses. A series of sensitivity tests was
examined to determine whether the estimated intervention point estimates were robust
against rival explanations.13 We conducted several placebo and sensitivity tests related to
the intervention date. The placebo tests were designed to assess the relationship between
the timing of the intervention estimate and the change in the targeted outcomes. We randomly selected four different placebo intervention dates for the preintervention time series
available across each targeted outcome. As shown in Table 8, no placebo test resulted in
a statistically significant intervention point estimate among the various homicide models
(i.e., total homicides, GMI homicides, and firearms homicides), which indicates the observed significant intervention effect on the various types of homicides was found only
when the modeled postintervention period paralleled the true postintervention period in New
Orleans. This would suggest that the observed intervention effect on homicides occurred
immediately after the initial notification session in New Orleans, which took place in late
October 2012. In terms of firearm assaults, the placebo postintervention model (i.e., placebo
model 4, or a postintervention period of January 2012 onward) was statistically significant

13.

24

We reestimated each model using negative binomial regression (to control for overdispersion), and the
results mirrored those that were presented in this study in all meaningful ways. We also included a
series of additional control variables such as linear and curvilinear trend measures, as well as annual
dummy variables to absorb random annual fluctuations in violence. None of the linear, curvilinear, or
use of dummy estimates changed the intervention estimates in any substantive way. We likewise
examined each model for potential autoregressive processes because time-series data are likely to be
influenced by closest proximity empirical values (McCleary and Hay, 1980). We did not find
autoregressive processes to be an issue of concern based on the following results: First, we conducted
Autoregressive Integrated Moving Average analyses on the preintervention time series, where we
detected no statistically significant first- or second-order autoregressive parameters in the model
identification stage. Second, we estimated first-order autoregressive parameters into each model that is
presented in this article using the ARPOIS package available in STATA (Tobias and Campbell, 1998),
where the results did not diverge substantively from those presented herein.

Criminology & Public Policy

 Corsaro and Engel
TA B L E

8

Preintervention Time-Series Intervention Placebo Tests on Targeted Crime
Outcomes (Total Homicides, GMI Homicides, Firearm Homicides, and Total
Firearm Assaults)
Total Homicidesa
Variable

Coefficient
(St. Error)

Placebo
Model 1
Placebo
Model 2
Placebo
Model 3
Placebo
Model 4

–.092
(.074)
.012
(.072)
–.103
(.074)
–.095
(.078)

p Value

.695
.859
.167
.219

Firearm Homicidesa
Coefficient
(St. Error)

–.047
(.081)
.002
(.078)
–.101
(.080)
–.086
(.084)

p Value

.559
.976
.210
.306

GMI Homicidesb
Coefficient
(St. Error)

–.039
(.144)
–.030
(.110)
–.057
(.109)
–.173
(.119)

p Value

.787
.780
.599
.144

Firearm Assaultsb
Coefficient
(St. Error)

.002
(.081)
–.093
(.113)
–.124
(.067)
–.195
(.058)

p Value

.976
.196
.088
.001

Note. Monthly dummy variables used to capture systematic seasonal fluctuations are estimated in each placebo model (4 models ×
4 outcomes = 16 in total).
a
Time series: January 1, 2008 to March 31, 2014, and random placebo preintervention and postintervention dates (1 = July 2009,
2 = February 2010, 3 = March 2012, 4 = May 2012).
b
Time series: January 1, 2010 to March 31, 2014, and random placebo preintervention and postintervention dates (1 = March 2010,
2 = February 2011, 3 = August 2011, 4 = January 2012).

(IRR = .822), which indicates that the reduction in nonlethal firearms violence occurred
in midyear 2013.14,15

14.

There are two potential interpretations of this finding for firearm assaults. One possibility is that firearm
assaults oscillated in an unsystematic way between January 2012 and March 2014 and that the
intervention estimate in the time-series models (despite the use of relevant and suitable control
variables) is influenced by such nonsystematic fluctuations. The second possibility, which seems more
plausible given the visual display of the data in Figure 2, is that the statistically significant reduction in
the average number of monthly firearm assaults was not precisely associated with the specific timing of
the intervention onset in the time-series models. In this instance, the movement of the postintervention
period to different dates does not influence the mean reduction in firearm assaults because the
observed decline was less consistent with the timing of the true postintervention period; rather, it was
influenced by the large shift in the magnitude of firearm assaults at some point in the true
postintervention period. Figure 2 shows that beginning in early summer 2013, firearm assaults declined
from reasonably stable 30–40 incidents per month to 15–25 incidents per month. In short, the timing of
the intervention in the model becomes less important because of the very large decrease in firearm
assaults that occurred in 2013.

15.

As a second set of sensitivity tests, we followed procedures used by Cook and MacDonald (2011) and
Piehl et al. (2003) to assess whether model fit parameters and the intervention point estimate
coefficients had the largest estimated effect sizes (and optimal model fit indices) at the time of the
intervention, or in the postintervention period (i.e., potential lagged effects) relative to potential
preintervention effects (i.e., lead effects). The results presented in Appendix A illustrate that for total
homicides, GMI homicides, and firearm homicides, the Wald chi-square statistic was smallest at the

Volume 14 r Issue 3

25

 Research Article

Focused Deterrence in New Orleans

Finally, it is important to isolate unique programmatic effects where possible. Although
the combination of GVRS (focused deterrence) and street workers for conflict mediation
(CURE Violence, which was implemented in Central City) has been shown to impact
crime in previous settings (Engel et al., 2013; see also Webster, Whitehill, Vernick, and
Curriero, 2013), we specifically partitioned the estimated impacts of GVRS and CURE
Violence within Central City as well as the remainder of the city (i.e., excluding Central
City).16 The results are displayed in Appendix B and show that the overall city (minus
Central City) experienced statistically significant declines in both total homicides (–17.5%,
p < .05) and GMI homicides (–32.9%, p < .05). Comparatively, although both total
homicides and GMI homicides were also reduced in Central City, this specific area of New
Orleans did not have statistically significant declines that corresponded with November
2012 onset date. Thus, as will be discussed in more detail, there was no indication that the
observed reduction in homicides that corresponded with the citywide focused GVRS was
driven by the supplemental CURE Violence strategy within Central City.
Discussion
The current study illustrates that focused deterrence strategies can have a significant impact
even in the most challenging of contexts, which in the City of New Orleans included
extremely high murder rates, political and police corruption, and a local culture seemingly
more tolerant of violence. Furthermore, overall homicides in New Orleans significantly
declined between 17% and 31% when compared with similar high-trajectory homicide
cities. More refined interrupted time-series analyses within New Orleans show that significant reductions in violence were observed specifically for overall homicides (–17%), GMI
homicides (–32%), homicides that involved young Black male victims (–26%), and both
lethal and nonlethal firearms violence (–16%). A series of sensitivity tests and supplemental
analyses provide more support that these observed intervention effects were robust, were
unlikely to have been caused by extraneous circumstances (e.g., a general overall crime shift),
and were consistent with the timing of the GVRS.
This study has two noteworthy limitations. First, the evaluation design was of secondary
consideration to the implementation of the strategy given the widespread and persistent
diffusion of violence within New Orleans. In an effort to overcome this limitation, we relied

point of the true intervention (November 2011). There seemed to be a 2-month lagged effect (lag – 2)
for firearm assaults. Thus, there is no evidence the intervention estimates were influenced in any
discernible way by preexisting (preintervention) declines.
16.

26

The City of Baltimore implemented a conflict mediation-based program (Safe Streets) within the
McElderry Park community. However, police districts within Baltimore also conducted offender
notification sessions, which were shown to have a significant and direct relationship with the observed
reduction in homicides and nonfatal shootings associated with Safe Streets (see Webster et al., 2013:
36). Webster et al. controlled for this relationship when examining the impact of the conflict mediation
strategy within McElderry Park.

Criminology & Public Policy

 Corsaro and Engel
on a series of analytical models to assess (and isolate) program impact; this methodology is
of moderate strength and would have been enhanced considerably with a stronger quasiexperimental or pure experimental design, such as the use of a place-based focus at the onset
to use a matching or randomization process (see Braga and Weisburd, 2014). However, prior
criminological research has illustrated that geographic locations that have the highest levels of
crime are also more likely to experience greater variability in crime. By modeling the change
in homicides in New Orleans against a set of approximately matched control cities, it was
apparent the decline was significantly sharper for New Orleans during the period examined
here than all other sites. Second, the current study does not allow us to disentangle which
of the observed aggregate crime effects are associated with deterrence (Braga and Weisburd,
2012), incapacitation effects (Levitt, 1998), changes in perceived legitimacy (Papachristos
et al., 2007), or enhancements in social services—which can lead to greater institutional
engagement (McCall et al., 2013). More precise measures (particularly at the individual
and group levels) that capture these regulatory theoretical principles through surveys and
narratives with notified offenders and groups would enhance the literature considerably.
Limitations notwithstanding, this study contributes to the literature in several important ways. Our findings illustrate that it might be possible to alter the mindset of gang and
criminally active group members in settings where retaliatory violence has been a common
occurrence. As found in a growing body of literature, gang and groups and individuals who
receive the focused deterrence message participate in fewer documented cases of gun violence (as both victims and offenders) and less overall crime (across multiple outcomes) after
call-in sessions relative to highly comparable gangs and groups as well as individuals who do
not receive the deterrent-based message (Braga et al., 2013, 2014; Papachristos and Kirk,
2015; Wallace et al., forthcoming). In short, it might be possible to alter in a tangible way
persistent cultures of violence.
The findings presented in this study also suggest that outreach workers used to interrupt violence in one specific neighborhood had no independent effect that empirically
corresponded with the reduction in violence in New Orleans. Such findings are similar
with prior focused deterrence evaluations that have attempted to describe and examine the
observable impact of street workers on violence (see Engel et al., 2013; Tillyer, Engel, and
Lovins, 2012). Although not definitive, the results in this study add to the growing body
of literature questioning the effectiveness of the use of street outreach workers rooted in
the public-health model that are not connected to a larger criminal-justice–based violencereduction initiative. As noted by Papachristos (2011), although the initial Chicago CeaseFire
strategy showed mostly promising results, replications have shown less support for conflict
mediators to combat serious lethal violence in alternative settings, such as in Pittsburgh
(see Wilson and Chermak, 2011). Where conflict mediation has been implemented in
settings that have shown changes in crime, it has corresponded with offender notification strategies that have been implemented in nearby geographic areas (e.g., Baltimore Safe
Streets—see Webster et al., 2013) or occurred simultaneously with mediation strategies
Volume 14 r Issue 3

27

 Research Article

Focused Deterrence in New Orleans

(e.g., Cincinnati—see Engel et al., 2013, and now New Orleans). Thus, the literature has
suggested that conflict mediation might complement focused deterrence, or its influence
could be conditional in unique social settings, but there is little evidence to indicate that
conflict mediation can function reliably as a stand-alone strategy to reduce violence. Future
replications and evaluations are needed in this area.
Finally, it should be recognized that policy transfer was greatly enhanced by the persistence and commitment to the strategy by political and police officials in New Orleans. The
research team, staff, and officers associated with GVRS received clear direction and unwavering political and police support from the highest levels of government, including Mayor
Landrieu and NOPD Superintendent Serpas. This provided a clear mandate for the City
of New Orleans, empowered those associated with the GVRS team, and ultimately led to
successful implementation and reduced violence. However, members from the New Orleans
GVRS working group (and future potential replicating agencies) would be well advised to
understand the potential toward deterioration of treatment in strategies that approach their
second, third, and fourth years (where applicable). Often, treatment deterioration is driven
by staffing changes, changes in the characteristics of the target population, some decline in
enthusiasm among working group officials, or other systematic factors (Kennedy, 2011; see
also Land, McCall, and Williams, 1992). As noted by Brunson et al. (2013), the deepest
collaborative relationships in focused deterrence working groups occur most often between
individuals and not institutions, and thus staffing changes can undermine the collaborative
capacity necessary for sustainability. Thoughtful efforts to rejuvenate the focused deterrence
model among the key providers are critical to long-term sustainability (Tillyer et al., 2012).
Additionally, maintaining a clear focus on problem identification (gangs and groups) is vital
to sustained success.
Factors such as defining gang violence can potentially undermine such momentum.
Even after recommending, measuring, and training NOPD supervisors multiple times
to use an expanded definition of gang-member involved violent incidents, the agency
routinely reverted back to its standard (conservative) definition of gang involvement in
violence, requiring multiple recoding of the violent incidents by the research team and
ongoing consultation and technical assistance to refocus the NOPD on the individuals and
gangs most responsible for violence in the city. We conclude by reiterating that researcher–
practitioner partnerships are imperative not only for problem identification, implementing
effective strategic approaches, resource management, and evaluation purposes, but also
for helping law enforcement keep track as to which individuals and groups are driving
a city’s violent crime problems (see Kennedy, 2009b) and for maintaining programmatic
sustainability. Each of these tasks is important for successful policy transfer and program
implementation. Although researchers have long acknowledged that measurement matters
for evaluation purposes, it should now be clear that measurement matters a great deal for
programmatic implementation.

28

Criminology & Public Policy

 Corsaro and Engel
Appendix A: Time-Series Intervention Estimate Sensitivity Tests on Targeted
Crime Outcomes (Total Homicides, GMI Homicides, Firearm Homicides, and
Total Firearm Assaults)
Total Homicidesa
Variable

Lead + 2
Lead + 1
Intervention
(T0 )
Lag – 1
Lag – 2

Firearm Homicidesa

GMI Homicidesb

Firearm Assaultsb

Coefficient
(St. Error)

Model Wald
Statistic

Coefficient
(St. Error)

Model Wald
Statistic

Coefficient
(St. Error)

Model Wald
Statistic

Coefficient
(St. Error)

Model Wald
Statistic

–.166
(.086)
–.166
(.087)
–.191*
(.085)
–.171*
(.090)
–.219**
(.086)

25.65

–.092
(.095)
–.153
(.092)
–.178*
(.092)
–.160*
(.098)
–.228**
(.091)

29.24

–.251*
(.124)
–.323**
(.114)
–.387**
(.115)
–.353**
(.115)
–.398**
(.119)

22.68

–.150*
(.068)
–.157*
(.067)
–.177**
(.067)
–.188**
(.073)
–.213**
(.076)

35.99

25.65
25.43
25.52
26.57

24.91
24.05
22.31
26.42

29.02
18.35
29.15
34.41

35.51
37.06
35.38
37.70

Note. Monthly dummy variables used to capture systematic seasonal fluctuations are estimated in each sensitivity model.
a
Time series: January 1, 2008 to March 31, 2014, and random placebo preintervention and postintervention dates (1 = July
2009, 2 = February 2010, 3 = March 2012, 4 = May 2012).
b
Time series: January 1, 2010 to March 31, 2014, and random placebo preintervention and postintervention dates (1 = March
2010, 2 = February 2011, 3 = August 2011, 4 = January 2012).
*
p < .05. ** p < .01.

Appendix B: Impact on Targeted Violence Within Geographic Settings with
Competing Strategic Interventions (January 1, 2008 to March 31, 2014)
Total Homicides
City Minus Central City
Variable

Intervention
February
March
April
May
June

GMI Homicides

Central City Only

City Minus Central City

Central City Only

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

–.193**
(.008)
–.188
(.179)
.158
(.179)
.181
(.177)
.048
(.159)
.048
(.177)

–17.55%

–.131
(.432)
–1.386
(1.055)
.223
(.740)
–.148
(.634)
–.554
(.755)
.139
(.664)

–12.27%

–.400*
(.115)
–.097
(.266)
–.047
(.224)
–.192
(.195)
–.192
(.244)
–.261
(.198)

–32.96%

–.161
(.414)
–1.648
(1.030)
.693
(1.081)
.893
(1.033)
.199
(1.253)
.199
(1.253)

–14.87%

–17.13%
17.11%
19.84%
4.91%
4.91%

–74.99%
24.98%
–13.75%
–42.53%
14.91%

–9.24%
–4.59%
–17.46%
–17.46%
–22.97%

Volume 14 r Issue 3

–80.75%
99.97%
144.24%
22.01%
22.01%

29

 Research Article

Focused Deterrence in New Orleans
Total Homicides

City Minus Central City
Variable

July
August
September
October
November
December
Intercept

*

GMI Homicides

Central City Only

City Minus Central City

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

(IRR-1)
× 100

Coefficient
(St. Error)

.162
(.176)
–.172**
(.224)
–.158
(.170)
–.158
(.231)
–.242
(.200)
–.025
(.194)
2.70**
(.147)

17.58%

–1.247
(1.036)
.698
(.544)
.362
(.712)
.139
(.666)
–1.226
(1.048)
.160
(.759)
–.523
(.506)

–71.26%

.017
(.238)
–.885**
(.293)
–.192
(.209)
–.261
(.285)
–.321
(.374)
.138
(.202)
2.29**
(.187)

–15.80%
–14.62%
–14.62%
–21.49%
–2.46%

100.97%
43.62%
14.91%
–70.65%
17.35%

(IRR-1)
× 100

1.71%
–58.72%
–17.47%
–22.97%
–27.45%
14.79%

Central City Only
Coefficient
(St. Error)

(IRR-1)
× 100

–1.650
(1.043)
1.991**
(.924)
1.586
(1.089)
.892
(1.033)
.239
(1.292)
.932
(1.287)
–1.548
(.899)

–80.79%
632.28%
388.41%
144.00%
26.99%
153.95%

p < .05. ** p < .01.

References
Allison, Paul D. and Richard P. Waterman. 2002. Fixed-effects negative binomial regression
models. Sociological Methodology, 32: 247–265.
Associated Press. 2005. New Orleans murder rate on the rise again. NBCNews.com. August
18. Retrieved June 17, 2015 from nbcnews.com/id/8999837#.VAHQd_ldXng.
Azrael, Deborah, Anthony A. Braga, and Mallory O’Brien. 2013. Developing the Capacity
to Understand and Prevent Homicide: An Evaluation of the Milwaukee Homicide Review
Commission. Washington, DC: U.S. Department of Justice. Retrieved June 17, 2015
from ncjrs.gov/pdffiles1/nij/grants/240814.pdf
Black, Donald J. 1970. Production of crime rates. American Sociological Review, 35: 733–
748.
Blumstein, Alfred and Richard Rosenfeld. 1998. Explaining recent trends in US homicide
rates. Journal of Criminal Law and Criminology, 88: 1175–1216.
Box, George and Gwilym M. Jenkins. 1976. Time Series Analysis: Forecasting and Control.
San Francisco, CA: Holden Day.
Braga, Anthony A. 2008. Pulling levers focused deterrence strategies and the prevention of
gun homicide. Journal of Criminal Justice, 36: 332–343.
Braga, Anthony A., Robert Apel, and Brandon C. Welsh. 2013. The spillover effect of
focused deterrence on gang violence. Evaluation Review, 37: 314–342.
Braga, Anthony A., David M. Hureau, and Andrew V. Papachristos. 2014. Deterring ganginvolved gun violence: Measuring the impact of Boston’s Operation Ceasefire on street
gang behavior. Journal of Quantitative Criminology, 30: 113–139.
30

Criminology & Public Policy

 Corsaro and Engel
Braga, Anthony A., David M. Kennedy, Elin J. Waring, and Anne M. Piehl. 2001. Problemoriented policing, deterrence, and youth violence: An evaluation of Boston’s Operation
Ceasefire. Journal of Research in Crime and Delinquency, 38: 195–226.
Braga, Anthony A., Andrew V. Papachristos, and David Hureau. 2010. The concentration and stability of gun violence at micro-places in Boston, 1980–2008. Journal of
Quantitative Criminology, 26: 33–53.
Braga, Anthony A., Glenn L. Pierce, Jack McDevitt, Brenda J. Bond, and Shea Cronin.
2008. The strategic prevention of gun violence among gang-involved offenders. Justice
Quarterly, 25: 132–162.
Braga, Anthony A., and David L. Weisburd. 2012. The effects of focused deterrence
strategies on crime: A systematic review and meta-analysis of the empirical evidence.
Journal of Research in Crime and Delinquency, 49: 323–358.
Braga, Anthony A., and David L. Weisburd. 2014. Must we settle for less rigorous evaluations
in large area-based crime prevention programs? Lesson from a Campbell review of
focused deterrence. Journal of Experimental Criminology. E-pub ahead of print.
Brunson, Rod K., Anthony A. Braga, David M. Hureau, and Kashea Pegram. 2013. We trust
you, but not that much: Examining police-black clergy partnerships to reduce youth
violence. Justice Quarterly. E-pub ahead of print. doi:10.1080/07418825.2013.868505
Cohen, Jacob. 1988. Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ:
Lawrence Earlbaum Associates.
Cook, Phillip. 1980. Research in criminal deterrence: Laying the groundwork for the
second decade. In (Norval Morris and Michael Tonry, eds.), Crime and Justice: An
Annual Review of Research. Chicago, IL: University of Chicago Press.
Cook, Thomas D. and Donald T. Campbell. 1979. Quasi-experimentation: Design and
Analysis for Field Settings. Boston, MA: Houghton-Mifflin.
Cook, Philip J. and John MacDonald. 2011. Public safety through private action: An
economic assessment of BIDS. The Economic Journal, 121: 445–462.
Corsaro, Nicholas. 2013. The High Point drug market intervention: Examining impact
across target areas and offense types. Victims & Offenders, 8: 1–31.
Corsaro, Nicholas, Rod K. Brunson, and Edmund F. McGarrell. 2010. Evaluating a policing
strategy intended to disrupt an illicit street-level drug market. Evaluation Review, 34:
513–548.
Corsaro, Nicholas, Eleizer D. Hunt, Natalie K. Hipple, and Edmund F. McGarrell. 2012.
The impact of drug market pulling levers policing on neighborhood violence: An
evaluation of the High Point drug market intervention. Criminology & Public Policy,
11: 165–199.
Corsaro, Nicholas and Edmund F. McGarrell. 2009. Testing a promising homicide reduction
strategy: Re-assessing the impact of the Indianapolis “pulling levers” intervention.
Journal of Experimental Criminology, 5: 63–82.
Durlauf, Steven N. and Daniel S. Nagin. 2011. Imprisonment and crime: Can both be
reduced? Criminology & Public Policy, 10: 9–54.
Engel, Robin S., S. Gregory Baker, Marie S. Tillyer, John Eck, and Jessica Dunham. 2008.
Implementation of the Cincinnati Initiative to Reduce Violence (CIRV): Year 1 Report.
Volume 14 r Issue 3

31

 Research Article

Focused Deterrence in New Orleans

Cincinnati, OH: University of Cincinnati Policing Institute. Retrieved June 17, 2015
from police.pittsburghpa.gov/publicsafety/pirc/files/cirv.1.pdf
Engel, Robin S., Marie Skubak Tillyer, and Nicholas Corsaro. 2013. Reducing gang violence
using focused deterrence: Evaluating the Cincinnati Initiative to Reduce Violence
(CIRV). Justice Quarterly, 30: 403–439.
Gibbs, Jack P. 1975. Crime, Punishment, and Deterrence. New York: Elsevier.
Goldstein, Herman. 1979. Improving policing: A problem-oriented approach. Crime &
Delinquency, 25: 236–258.
Griffiths, Elizabeth and Jorge M. Chavez. 2004. Communities, street guns, and homicide
trajectories in Chicago, 1980–1995: Merging methods for examining homicide trends
across space and time. Criminology, 42: 941–978.
Haviland, Amelia M. and Daniel S. Nagin. 2005. Causal inference with group-based
trajectory models. Psychometrika, 70: 1–22.
Haviland, Amelia M. and Daniel S. Nagin. 2007. Using group-based trajectory modeling
in conjunction with propensity scores to improve balance. Journal of Experimental
Criminology, 3: 65–82.
Haviland, Amelia, Daniel S. Nagin, and Paul R. Rosenbaum. 2007. Combining propensity score matching and group-based trajectory analysis in an observational study.
Psychological Methods, 12: 247–267.
Hilbe, Joseph M. 2007. Negative Binomial Regression. Cambridge, U.K.: Cambridge University Press.
Hirsch, Arnold R. and Joseph Logsdon. 1992. Creole New Orleans: Race and Americanization.
Baton Rouge: Louisiana State University Press.
Jones, Bobby L., Daniel S. Nagin, and Kathryn Roeder. 2001. A SAS procedure based
on mixture models for estimating developmental trajectories. Sociological Research and
Methods, 29: 374–393.
Kennedy, David M. 1997. Pulling levers: Chronic offenders, high-crime settings, and a
theory of prevention. Valparaiso University Law Review, 31: 449–484.
Kennedy, David M. 2006. Old wine in new bottles: Policing and the lessons of pulling levers.
In (David L. Weisburd and Anthony A. Braga, eds.), Police Innovation: Contrasting
Perspectives. New York: Cambridge University Press.
Kennedy, David M. 2009a. Deterrence and Crime Prevention: Reconsidering the Prospect of
Sanction. New York: Routledge.
Kennedy, David M. 2009b. Gangs and public policy: Constructing and deconstructing
gang databases. Criminology & Public Policy, 8: 711–716.
Kennedy David M. 2011. Don’t shoot: One man, a street fellowship, and the end of violence
in inner-city America. New York: Bloomsbury
Kennedy, David M. and Michael A. Friedrich. 2014. Custom Notifications: Individualized Communication in the Group Violence Intervention. Washington, DC: Office of
Community Oriented Policing Services.

32

Criminology & Public Policy

 Corsaro and Engel
King, Gary. 1988. Statistical models for political science event counts: Bias in conventional
procedures and evidence for the exponential Poisson regression. American Journal of
Political Science, 32: 838–863.
Kirk, David S. 2009. A natural experiment on residential change and recidivism: Lessons
from Hurricane Katrina. American Sociological Review, 74: 484–505.
Land, Kenneth C., Patricia L. McCall, and Jay R. Williams. 1992. Intensive supervision
of status offenders: Evidence on continuity of treatment effects for juveniles and
a “Hawthorne Effect” for counselors. In (Joan McCord and Richard E. Tremblay,
eds.), Preventing Antisocial Behavior: Interventions from Birth to Adolescence. New York:
Guilford Press.
Levitt, Steven D. 1998. Why do increased arrest rates appear to reduce crime: Deterrence,
incapacitation, or measurement error? Economic Inquiry, 36: 353–372.
Long, J. Scott. 1997. Regression Models for Categorical and Limited Dependent Variables.
Thousand Oaks, CA: Sage.
Maxson, Cheryl. 1999. Gang homicide: A review and extension of the literature. In (M.
Dwayne Smith and Margaret A. Zahn, eds.), Homicide Studies: A Sourcebook of Social
Research. Newbury Park, CA: Sage.
McCall, Patricia L., Kenneth C. Land, Cindy Brooks Dollar, and Karen F. Parker. 2013. The
age structure-crime relationship: Solving a long-standing puzzle. Journal of Quantitative
Criminology, 29: 167–190.
McCall, Patricia L., Kenneth C. Land, and Karen F. Parker. 2011. Heterogeneity in the rise
and decline of city-level homicide rates, 1976–2005: A latent trajectory analysis. Social
Science Research, 40: 363–378.
McCleary, Richard and Richard A. Hay. 1980. Applied Time-Series Analysis for the Social
Sciences. Beverly Hills, CA: Sage.
McDowall, David, Richard McCleary, Errol E. Meidinger, and Richard A. Hay, Jr. 1980.
Interrupted Time Series Analysis. Thousand Oaks, CA: Sage.
McGarrell, Edmund F., Steven Chermak, Jeremy M. Wilson, and Nicholas Corsaro. 2006.
Reducing homicide through a “lever-pulling” strategy. Justice Quarterly, 23: 214–231.
McGarrell, Edmund F., Nicholas Corsaro, Natalie K. Hipple, and Timothy S. Bynum.
2010. Project safe neighborhoods and violent crime trends in U.S. cities: Assessing
violent crime impact. Journal of Quantitative Criminology, 26: 165–190.
Morgan, Stephen L. and Christopher Winship. 2007. Counterfactuals and Causal Inference:
Methods and Principles for Social Research. New York: Cambridge University Press.
Moore, Leonard. 2010. Black Rage in New Orleans: Police Brutality and African American Activism from World War II to Hurricane Katrina. Baton Rouge: Louisiana State
University Press.
Nagin, Daniel S. 1998. Criminal deterrence research at the outset of the twenty-first century.
In (Michael Tonry, ed.), Crime and Justice: A Review of Research. Chicago, IL: University
of Chicago Press.
Nagin, Daniel S. 2005. Group-Based Modeling of Development. Cambridge, MA: Harvard
University Press.
Volume 14 r Issue 3

33

 Research Article

Focused Deterrence in New Orleans

Nagin, Daniel S. 2013. Deterrence in the twenty first century: A review of the evidence.
In (Michael Tonry, ed.), Crime and Justice: An Annual Review of Research. Chicago, IL:
University of Chicago Press.
National Research Council. 2004. Fairness and Effectiveness in Policing: The Evidence. Washington, DC: National Academies Press.
Osgood, D. Wayne. 2000. Poisson-based regression analysis of aggregate crime rates. Journal
of Quantitative Criminology, 16: 21–43.
Papachristos, Andrew V. 2011. Too big to fail: The science and politics of violence prevention. Criminology & Public Policy, 10: 1053–1061.
Papachristos, Andrew V. and David S. Kirk. 2015. Changing the street dynamic: Evaluating
Chicago’s group violence reduction strategy. Criminology & Public Policy. This issue.
Papachristos, Andrew V., Tracey L. Meares, and Jeffery Fagan. 2007. Attention felons:
Evaluating Project Safe Neighborhoods in Chicago. Journal of Empirical Legal Studies,
4: 223–272.
Paternoster, Raymond. 1987. The deterrent effect of the perceived certainty and severity of
punishment: A review of the evidence and issues. Justice Quarterly, 4: 173–218.
Piehl, Anne Morrison, Suzanne J. Cooper, Anthony A. Braga, and David M. Kennedy.
2003. Testing for structural breaks in evaluation of programs. The Review of Economics
and Statistics, 85: 550–558.
Saunders, Jessica, Russell Lundberg, Anthony A. Braga, Greg Ridgeway, and Jeremy Miles.
2014. A synthetic control approach to evaluating place-based crime interventions.
Journal of Quantitative Criminology. E-pub ahead of print.
Sherman, Lawrence W., Denise C. Gottfredson, Doris L. MacKenzie, John Eck, Peter
Reuter, and Shawn D. Bushway. 1998. Preventing Crime: What Works, What Doesn’t,
What’s Promising. Washington DC: National Institute of Justice.
Skogan, Wesley G., Susan M. Hartnett, Natalie Bump, and Jill Dubois. 2009. Evaluation
of CeaseFire-Chicago. Washington, DC: U.S. Department of Justice, Office of Justice
Programs, National Institute of Justice.
Tillyer, Marie Skubak, Robin S. Engel, and Brian Lovins. 2012. Beyond Boston: Applying
theory to understand and address sustainability issues in focused deterrence initiatives
for violence reduction. Crime & Delinquency, 58: 973–997.
Tita, George K., Jack Riley, Greg Ridgeway, Clifford A. Grammich, Allan Abrahamse, and
Peter W. Greenwood. 2004. Reducing Gun Violence: Results from an Intervention in East
Los Angeles. Santa Monica, CA: RAND Corporation.
Tobias, Aurelio and Mike Campbell. 1998. Arpois Stata Package. Retrieved June 17, 2015
from stata.com/stb/stb46/sts13/arpois.ado
Tyler, Tom R. 1990. Why People Obey the Law. New Haven, CT: Yale University Press.
Uniform Crime Reports. 2013. Table 8: Offenses known to law enforcement by state
by city, 2008–2013. Retrieved December 10, 2014 from fbi.gov/about-us/cjis/ucr/
crime-in-the-u.s/2013/crime-in-the-u.s.-2013
Wallace, Danielle, Andrew V. Papachristos, Tracey Meares, and Jeffrey Fagan. Forthcoming.
Desistance and legitimacy. The impact of offender notification meetings on recidivism
among high risk offenders. Justice Quarterly. E-pub ahead of print.
34

Criminology & Public Policy

 Corsaro and Engel
Warner, Barbara and Glenn Pierce. 1993. Reexamining social disorganization theory using
calls to the police as a measure of crime. Criminology, 31: 493–518.
Webster, Daniel W., Jennifer M. Whitehill, Jon S. Vernick, and Frank C. Curriero. 2013.
Effects of Baltimore’s Safe Streets program on gun violence: A replication of Chicago’s
CeaseFire program. Journal of Urban Health, 90: 27–40.
Weisburd, David L., Shawn D. Bushway, Cynthia Lum, and Sue-Ming Yang. 2004. Trajectories of crime at places: A longitudinal study of street segments in the city of Seattle.
Criminology, 42: 283–321.
Weisburd, David L., Cody W. Telep, Joshua C. Hinkle, and John Eck. 2010. Is problemoriented policing effective in reducing crime and disorder? Findings from a Campbell
systematic review. Criminology & Public Policy, 9: 139–172.
Wellford, Charles, Brenda J. Bond, and Sean Goodison. 2011. Crime in New Orleans:
Analyzing Crime Trends and New Orleans’ Responses to Crime. Washington, DC: Office
of Justice Programs, Bureau of Justice Assistance.
Williams, D. J., D. Currie, W. Linden, and P. D. Donnelly. 2014. Addressing gang-related
violence in Glasgow: A preliminary pragmatic quasi-experimental evaluation of the
Community Initiative to Reduce Violence (CIRV). Aggression and Violent Behavior,
19: 686–691.
Wilson, Jeremy M. and Steven Chermak. 2011. Community-driven violence reduction
programs: Examining Pittsburgh’s One Vision One Life. Criminology & Public Policy,
10: 993–1027.
Zimring, Frank E. and Gordon Hawkins. 1973. Deterrence: The Legal Threat in Crime
Control. Chicago, IL: University of Chicago Press.
Nicholas Corsaro is an assistant professor in the School of Criminal Justice at the University
of Cincinnati and the research director at the Police Foundation in Washington, DC. His
research focuses on collaborating with police agencies to develop efficient and effective
approaches to problem analysis and crime prevention. His previous research appears in
Crime & Delinquency, Criminology & Public Policy, Journal of Criminal Justice, Journal of
Experimental Criminology, Journal of Quantitative Criminology, Journal of Urban Health, and
Justice Quarterly. He received his Ph.D. from the School of Criminal Justice at Michigan
State University.
Robin S. Engel is a professor in the School of Criminal Justice and the director of the
Institute of Crime Science at the University of Cincinnati. She works extensively in partnership with police agencies to enhance their effectiveness, efficiency, and equity. Her research
includes empirical assessments of police behavior, police/minority relations, police supervision and management, criminal gangs, and violence-reduction strategies. Previous research
has appeared in Criminology, Justice Quarterly, Journal of Research in Crime and Delinquency,
Journal of Criminal Justice, Crime & Delinquency, and Criminology & Public Policy.

Volume 14 r Issue 3

35