Content uploaded by Michael Sierra-Arévalo
Author content
All content in this area was uploaded by Michael Sierra-Arévalo on Jan 25, 2019
Content may be subject to copyright.
Content uploaded by Michael Sierra-Arévalo
Author content
All content in this area was uploaded by Michael Sierra-Arévalo on Dec 06, 2017
Content may be subject to copyright.
Crime & Delinquency
1 –22
© The Author(s) 2016
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0011128716635197
cad.sagepub.com
Article
Evaluating the Effect
of Project Longevity
on Group-Involved
Shootings and Homicides
in New Haven,
Connecticut
Michael Sierra-Arevalo1, Yanick Charette1,
and Andrew V. Papachristos1
Abstract
Beginning in November 2012, New Haven, Connecticut, served as the pilot
site for Project Longevity, a statewide focused deterrence gun violence
reduction strategy. The intervention brings law enforcement, social services,
and community members together to meet with members of violent street
groups at program call-ins. Using autoregressive integrated moving average
models and controlling for the possibility of a non-New Haven–specific
decline in gun violence, a decrease in group offending patterns, and the
limitations of police-defined group member involved (GMI) categorization
of shootings and homicides, the results of our analysis show that Longevity
is associated with a reduction of almost five GMI incidents per month. These
findings bolster research confirming the efficacy of focused deterrence
approaches to reducing gun violence.
Keywords
crime prevention, gangs, intervention, violence, deterrence
1Yale University, New Haven, CT, USA
Corresponding Author:
Michael Sierra-Arevalo, Department of Sociology, Yale University, 493 College, New Haven,
CT 06511, USA.
Email: michael.sierra-arevalo@yale.edu
635197CADXXX10.1177/0011128716635197Crime & DelinquencySierra-Arevalo et al.
research-article2016
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
2 Crime & Delinquency
Introduction
In 1991, New Haven, Connecticut, recorded a 30-year high of 36 homicides, mir-
roring the rates of violent crime that affected many American cities during the
rise of the “crack epidemic” (Blumstein & Rosenfeld, 1998; Cook & Laub, 1998;
Tonry & Moore, 1998). Following one of the largest and longest running crime
declines in American history (Zimring, 2007), the homicide count in New Haven
plummeted by 78% from 1991 to 2003. Shortly after 2003, however, homicides
in New Haven began to buck the national crime decline and began trending
upward. In 2011, New Haven came just shy of its 1991 high with 34 homicides.
With a murder rate of 26.2 per 100,000 in 2011, New Haven’s murder rate out-
paced Washington, D.C., and Chicago, and was on par with Oakland, California.1
In response to the mounting death toll, in 2011, state and local officials part-
nered with the New Haven Police Department, social service providers, and New
Haven community members to implement a data-driven gun violence reduction
strategy. With the formation of this partnership, New Haven became the pilot site
for a statewide gun violence reduction project that had shown success in cities
such as Boston and Chicago (Braga, Kennedy, Waring, & Piehl, 2001;
Papachristos, Meares, & Fagan, 2007). Drawing on the successful efforts of these
and other cities in reducing gun violence, New Haven’s Project Longevity began
in 2012, its strategy focused on targeting the small population of high-risk, repeat
offenders, often gang- or street group-involved, who account for the majority of
gun violence (Kennedy, 1997; Kennedy, Braga, & Piehl, 1997).
This article examines the efficacy of Project Longevity in reducing gun vio-
lence in New Haven after its first 3 years of continuous operation. Specifically,
we analyze whether or not the timing of Longevity affected the levels of group
member involved (GMI) shootings and homicides. Our results suggest that the
initiation of Project Longevity is associated with a significant decrease in GMI
shootings and homicides during the observation period. Using Hartford,
Connecticut, as a comparative case, we find that the observed decrease of GMI
gun violence in New Haven is not part of a general decline in gun violence, nor
a New Haven–specific decline in group crime. In addition, we test for the pos-
sibility that the police-defined category of “group member involved” biases our
results and find no such bias. In short, our findings provide evidence that the
observed decline in GMI shootings and homicides in New Haven is strongly
associated with the implementation of Longevity.
Background
Two parallel developments in American policing set the stage for the focused deter-
rence strategies that Project Longevity is modeled upon: problem-oriented policing
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 3
and community policing. Moving away from the professionalization, bureaucrati-
zation, and the “means over ends” emphasis that characterized the “reform era” of
policing (Goldstein, 1979; Kelling & Moore, 1988), problem-oriented policing
posits that police are better served by addressing problems through a process of
“identification, analysis, response, evaluation, and adjustment of the response”
(Braga, Kennedy, Waring, et al., 2001, p. 196). In conjunction with problem-
oriented policing, community policing gained popularity in the 1980s and 1990s as
a way for police to mend the relationships strained by the social upheaval of the
1960s and 1970s. Central to community policing are the principles of “proactive
problem-solving strategies” (Rosenbaum, 1988) and the inclusion of community
residents in the problem-solving process (Skogan, 1990).
Drawing from these developments, the group violence intervention (GVI)
pioneered in Boston in the 1990s is a focused deterrence strategy aimed at
reducing shootings and homicides driven by criminally active street groups
(Braga, Kennedy, Piehl, & Waring, 2001; Braga, Kennedy, Waring, et al.,
2001).2 Although deterrence is still integral to GVI—the strategy still uses
the promise and use of certain, swift, and severe punishment to dissuade
undesired behavior (Akers, 1999; Gibbs, 1975; Stafford & Warr, 1997;
Zimring & Hawkins, 1973)—the GVI approach departs markedly from tradi-
tional deterrence techniques such as increasing police presence in a general
area or engaging in mass police “crackdowns” (Weisburd, Telep, Hinkle, &
Eck, 2010). Instead, GVI focuses on the small number of individuals, often
involved in street gangs or groups, who account for the vast majority of gun
violence in cities, and makes use of a wide variety of legal “levers” on indi-
viduals who are often under state supervision, for example, probation or
parole (Braga, Hureau, & Winship, 2008; Kennedy, Piehl, & Braga, 1996).3
Rather than casting its net broadly and increasing penalties across large parts
of the population, the GVI approach identifies specific problems (e.g., street
group gun violence) and mobilizes law enforcement, social service, and com-
munity resources to address those problems and the individuals most likely to
be involved as perpetrators and victims.
Project Longevity—New Haven, Connecticut
The GVI strategy in Connecticut, branded as Project Longevity, began in
New Haven in August 2012. To ascertain the nature of the gun violence and
street group issue and guide future programmatic decisions, New Haven
began with a problem analysis composed of two parts: a group audit and an
incident review. The group audit is a focus group style meeting in which law
enforcement practitioners are guided through a mapping and survey exercise
to collect detailed information on street groups in a given municipality
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
4 Crime & Delinquency
(Sierra-Arevalo & Papachristos, 2015a).4 During the audit, information is
collected on the active street groups in the city, harnessing the unique “expe-
riential assets” of law enforcement to better understand the geographic loca-
tion of street groups, who is in them, and what activities members are engaged
in (Kennedy et al., 1997). In addition, because much of the violence perpe-
trated by street groups is part of a reciprocal process of intergroup conflict
(Papachristos, 2009; Papachristos, Hureau, & Braga, 2013; Tita & Radil,
2011), a key part of the audit is the collection of relational data detailing the
system of feuds and alliances between street groups (Kennedy et al., 1997;
Sierra-Arevalo & Papachristos, 2015a).
The incident review also leverages the experience of law enforcement
officers but is tailored to gather information on shootings and homicides.
Specifically, the incident review aims to assess which groups are most
actively involved in gun violence and the circumstances surrounding each
shooting. Officers are presented with information about past shootings and
homicides, including victim and offender information, location, and any
other information about the circumstances of the event. Officers are then
tasked with identifying whether the event was GMI based on whether the
victim or the perpetrator is a member of a group identified during the group
audit process.5
The initial problem analysis in New Haven showed the existence of 52
unique groups at the time of the audit, with 440 identified street group mem-
bers. Longevity’s focused approach meant that only those groups involved in
gun violence would be part of the intervention—less than 60% (n = 30) of the
52 identified groups were involved in shootings or homicides. Similarly, not
all identified groups were involved in active feuds with other street groups,
with only 42% (n = 22) of the groups engaged in conflict at the time of the
audit.
Using the intergroup conflict and shooting data in conjunction with data
on group membership, location, and violent activity gathered during the
group audit and incident review, Longevity personnel were able to identify
the most violent groups in the city and select group members to take part in
the intervention.
The Intervention: Group Call-Ins
Guided by data gathered during the group audit and incident review, Project
Longevity staff chose members from the two most violent street groups to be
invited to a call-in. Call-ins are meetings between street group members and
law enforcement, social service providers, and community members, function-
ing as an information dissemination tool for the law enforcement–community
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 5
partnership to deliver three key points to attendees. These points are (a) a com-
munity moral message against violence, (b) a credible law enforcement mes-
sage about the consequences of further violence, and (c) a genuine offer of help
for those who want it (National Network for Safe Communities [NNSC],
2013).
New Haven’s first call-in took place in November 2012, shortly after the
initial problem analysis. On the day of the call-in, members from the two
most violent groups were called into the aldermanic chambers housed in New
Haven’s City Hall. Attendees listened to the Longevity message over the
course of an hour, hearing from law enforcement, social service providers,
and community members. Representatives from law enforcement spoke to
attendees first, making sure to articulate the “new rules” (NNSC, 2013, p. 76)
being implemented, and reiterated that those who continued to engage in gun
violence (and their group) would meet with focused law enforcement atten-
tion. Next, social service providers showed call-in attendees that there is help
available to those who want it. Services offered in the New Haven call-ins
included housing assistance, high school diploma or general education devel-
opment (GED) classes, job training, and drug or alcohol recovery. Last, com-
munity representatives acted as “moral voices” known to and respected by
the call-in attendees, articulating to attendees the anti-violence message of
the program and drawing on their unique positions within the community to
help attendees connect with the message. For example, a formerly incarcer-
ated speaker offered testimony that change and a life away from guns and the
street life is possible, while the mother of a victim of gun violence spoke as a
“voice of pain,” her words concentrating on the tragic costs of street
violence.6
Subsequent call-ins continued to reach out to the other violent groups in
the city between November 2012 and June 2014, with a total of six call-ins.
The format of the call-ins remained consistent across this time.7 If and when
there was a law enforcement action against a violent group that had attended
a previous call-in, that law enforcement action was showcased to call-in
attendees as proof positive that the message they received was real, and that
continuing to engage in violence would have swift and certain
consequences.
Evaluation Design
The ideal implementation of Longevity would have allowed for a quasi-
experimental design that compared the rates of shooting incidents between
a treatment group and some comparison group (Braga, Hureau, &
Papachristos, 2014; Papachristos & Kirk, 2015). However, Project
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
6 Crime & Delinquency
Longevity was never designed as a quasi-experiment. Because of the con-
centration of gun violence within a small number of active street groups
and within a small number of neighborhoods, Longevity was designed as a
non-randomized and highly focused effort that selected groups based on
their involvement in gun violence. This left the project staff with little
choice but to “treat” nearly all of the groups identified in the group audit.
As a result, within the first year of implementation, virtually all of the iden-
tified groups in New Haven participated in at least one call-in, thereby leav-
ing no comparison groups.
The inability to have a true experimental design is a common limitation in
gun violence programs, including other GVI evaluations (Braga et al., 2014;
Engel, Corsaro, & Tillyer, 2010; Engel, Tillyer, & Corsaro, 2013). To analyze
crime trends within New Haven before and after the start of Longevity, we
use a series of interrupted time series regression models (Braga, Kennedy,
Waring, et al., 2001) designed to address three possible confounding factors
that, in light of our design, might bias our results: (a) a generalized decline in
gun violence, (b) a New Haven–specific decline in group-related criminality,
and (c) bias in the GMI identification process used by police.
Data
Data used in the present study were derived from fatal and non-fatal shooting
records collected by the New Haven and Hartford Police Departments
between January 2011 and April 2014. Data were aggregated to monthly
counts, creating a 40-month time series for both cities. Project Longevity was
active in New Haven during the last 18 months of the observation period,
starting November 2012 and continuing through April 2014. Given the spe-
cific focus of Longevity on gun violence driven by the small population of
criminally active street group members, the project aimed at addressing
shootings and homicides that were GMI—that is, fatal and non-fatal shoot-
ings police identified during shooting reviews as involving a street group
member as a victim or offender. All analyses are therefore conducted on total,
GMI, and non-GMI shootings.
Analytic Strategy
Taking into account the temporal nature of the data, we use a series of
AutoRegressive Integrated Moving Average (ARIMA) models, which
account for temporal dependencies of time series data. In our analysis,
the main dependent variable is the number of GMI shootings as a func-
tion of the treatment period.8 Our models include an autoregressive
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 7
parameter (Ar1 [ARIMA(1,0,0)]) to account for temporal autocorrela-
tion found between consecutive months.. When the number of shootings
is regressed on itself with a time lag of 1 month, no stationarity issues are
observed, confirming a key assumption required for the use of ARIMA
models. 9
As we have discussed, Longevity’s lack of experimental design makes it
difficult to discern whether any observed decrease in GMI incidents during
the treatment period is directly attributable to Longevity, or whether it is the
result of (a) a generalized decrease in shootings and homicides that extends
beyond New Haven, (b) a decrease in group-based criminality in New Haven,
or (c) bias in the categorization of fatal and non-fatal shootings as GMI or
non-GMI.
To address the first issue, we compare shootings and homicides in New
Haven with those in Hartford, Connecticut, a nearby city that did not receive
treatment, to account for general trends that might be occurring within the
same state.10 Because Longevity was active in New Haven but not Hartford
during the observation period, a Longevity-specific effect should manifest in
a greater reduction in GMI shootings in New Haven as compared with
Hartford over this time period.
A second series of analyses addresses the possibility that a decrease in
GMI shootings in New Haven is due to a decline in group-based criminality,
such as criminal activity of street groups (Braga, Papachristos, & Hureau,
2009). To account for this possibility, we aggregate the number of offenses in
New Haven at the month level using only those offenses for which police
records show more than one offender, and then compare trends in co-offend-
ing, that is, group crime, with GMI trends over the observation period. If the
reduction in GMI incidents is linked to Longevity’s implementation, we
expect any reduction in GMI shootings to be independent of trends in
co-offending.
Finally, because the identification of GMI incidents is necessarily
imperfect by nature of the incomplete information patrol officers and
investigators have at their disposal during investigations, it is not uncom-
mon for the shooter(s) to be unknown, even when a victim is known to not
be involved with an identified street group. Without this information,
police cannot conclusively say that a shooting incident is GMI, and must
conservatively list them as non-GMI, which could bias our measure of
GMI incidents in such a way that excludes shootings and homicides that
could likely be GMI. To address this potential problem, we construct a
more lenient possible-GMI category using logistic regression, predicting
shooting as non-GMI or possible-GMI based on a victim’s age, gender,
and race, whether the shooting was fatal or non-fatal, whether the suspect
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
8 Crime & Delinquency
was identified, and in which police district the event happened. Based on
this model, incidents that closely approximate events that police catego-
rized as GMI but were not officially labeled as such are included in the
category of possible-GMI. If a decrease in the number of incidents related
to this broader measure of GMI is still observed during the time of the
intervention, this decrease is not likely due to imperfections in the GMI
identification process.
A final set of analyses considers all of these possible issues simultane-
ously. If a decrease in GMI incidents during the treatment period can still be
observed after controlling for the trend in shootings in nearby Hartford, the
level of group criminality in New Haven, and variation in the identification
process of GMI incidents, then we can much more confidently claim that the
observed decrease in GMI shootings in New Haven is due to Project
Longevity.
Results
Trends in Shootings and Homicides in New Haven
Figure 1 shows the monthly distribution of shootings and homicides in
New Haven before and after the start of Project Longevity. In the 22
months leading up to the start of Longevity, there were 11.64 total shoot-
ings per month (SD = 4.17), 19.2% of which were homicides (M = 2.23,
SD = 1.63). Importantly, total shootings were trending downward in New
Haven even prior to implementation of Longevity: In the 22 months
before the first call-in, total shootings decreased 55.9% and fell to an
average monthly total of 7.3 shootings per month (SD = 3.34) after the
first call-in. This decline in total shootings—approximately four fewer
incidents per month—is reflected in both decreased homicides (before:
M = 2.23, SD = 1.63; after: M = 1.61, SD = 1.20, t(38) = 1.34, p = .190),
and non-fatal shootings (before: M = 9.41, SD = 3.43; after: M = 5.72,
SD = 2.74, t(38) = 3.69, p = .001).
The third panel of Figure 1 plots the monthly number of GMI incidents
before and after the start of Longevity. Approximately 59% (n = 230) of all
recorded shootings between January 2011 and April 2014 involved a mem-
ber of a street group as either victim or offender (GMI). Like non-GMI
shootings, GMI shootings have, overall, trended downward despite some
peaks during various months. Prior to Longevity, there were an average of
8.59 GMI shootings per month (SD = 1.78); this figure dropped to 2.28
GMI shootings per month after Longevity’s first call-in (SD = 1.78), an
almost 73% drop in average monthly GMI shootings, (t(38) = 6.63,
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 9
p < .001). At the same time that GMI shootings decreased, however, non-
GMI shootings increased approximately 66%. Before Longevity, there
were approximately 3.05 non-GMI shootings per month (SD = 2.08),
increasing to approximately 5.06 shootings per month (SD = 2.34) after the
first call-in. The marked decline in GMI shootings after the start of
Longevity suggests a negative programmatic effect. However, because the
GMI shootings and homicides were decreasing before Longevity’s imple-
mentation, we must investigate whether the decrease in GMI incidents
after the first call-in is attributable to the intervention, or whether it is
simply a continuation of a broader trend. To address this potential issue,
we use interrupted time series regressions that predict the total number of
GMI shootings as a function of the treatment period.
0
5
10
15
20
Number of incidents
Shootings and homicides Project Longevity
0
5
10
15
20
HomicidesShootings
0
5
10
15
20
2011.01
2011.03
2011.05
2011.07
2011.09
2011.11
2012.01
2012.03
2012.05
2012.07
2012.09
2012.11
2013.01
2013.03
2013.05
2013.07
2013.09
2013.11
2014.01
2014.03
GMINon-GMI
Figure 1. Monthly distribution of total incidents, shootings and homicides, and
GMI/non-GMI incidents in New Haven, Connecticut.
Note. GMI = group member involved.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
10 Crime & Delinquency
Table 1 presents the results of interrupted time series regressions predict-
ing the effect of Longevity on total and GMI shootings and homicides.
Results from these regressions show that the number of total shootings and
homicides decreased by just more than four during the intervention period.
By comparison, GMI shootings and homicides decreased by 5.33 during
months in which a call-in was performed.
Although these results are suggestive of a programmatic effect, it is diffi-
cult to ascertain whether the observed post-Longevity decline is directly
related to programmatic efforts without a true experimental design. In addi-
tion, there are at least three alternative explanations or potential issues that
might undermine the association between Longevity and the observed post-
Longevity decline: generalized downward shooting trends that extend beyond
New Haven, a decline in New Haven–specific group criminality, and bias in
which incidents are classified as “Group Member Involved” shootings. We
address each of these in turn.
Possible Issue 1: Generalized Decline in Shootings and
Homicides
Knowing that the lack of treatment and comparison groups within New
Haven makes it difficult to pinpoint programmatic effects within the same
city, an alternative approach to explore whether the post-Longevity decline in
GMI incidents is related to program implementation is to compare the trend
in New Haven with the trend of comparable cities that did not participate in
Longevity. One likely candidate for such a comparison is nearby Hartford,
Connecticut.
Hartford is approximately the same size as New Haven with a population
of 125,017, and is 39 miles away. More importantly, prior to Longevity, both
Hartford and New Haven had similar levels of fatal and non-fatal shootings
Table 1. Regression Model Predicting the Effect of Project Longevity on Number
of Incidents Per Month in New Haven.
Shootings and homicides
(New Haven)
Total (n = 388) GMI (n = 230)
b SE p b SE p
Intercept 11.53 1.04 < .001 8.05 1.09 < .001
Ar1 0.28 0.15 .064 0.50 0.15 .001
Intervention −4.05 1.54 .009 −5.33 1.61 .001
Note. GMI = group member involved; Ar1 = autoregressive parameter.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 11
and were both experiencing a downward trend. Figure 2 shows the number of
total monthly shootings in both cities before and after the start of Longevity
in New Haven. Prior to Longevity, both cities had a comparable number of
monthly shootings, although New Haven’s rate began trending below
Hartford’s as early as November 2011—a trend that accelerated after
Longevity began. Still, Figure 2 shows the trend of shootings in Hartford is
correlated with the trend in New Haven (r = .44, p = .004).
Table 2 shows the same regression model presented in the previous analy-
sis, but this time predicting shootings and homicides that occurred in Hartford.
We can observe that during Longevity’s intervention period, the decrease in
GMI incidents observed in New Haven is not seen in Hartford, suggesting the
intervention period is unrelated to incidents in Hartford.
Next, we predict the overall level of shootings and homicides in New Haven,
as well as GMI incidents specifically, while accounting for fatal and non-fatal
shooting trends in Hartford. The results of interrupted time series regressions
predicting total shootings and homicides and GMI incidents in New Haven,
while controlling for shootings and homicides in Hartford, are displayed in the
bottom half of Table 2. After controlling for shootings and homicides (b = 0.28,
SE = 0.12, p = .025) and GMI incidents in Hartford (b = 0.16, SE = 0.09,
p = .079), a significant decrease in GMI shootings and homicides is still found
in New Haven during the intervention period (total: b = −3.47, SE = 1.36,
p = .011; GMI: b = −5.04, SE = 1.51, p = .001), suggesting that reduction in
GMI incidents in New Haven are independent of trends in Hartford.
Figure 2. Monthly distribution of the shootings and homicides in New Haven and
in Hartford before and during Project Longevity.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
12 Crime & Delinquency
Possible Issue 2: Decline in Group Offending
The number of shootings and homicides in a city is closely tied to the crimi-
nal activity of street groups (Braga et al., 2009). If street group activity gener-
ally decreased in New Haven, the observed decrease in GMI incidents might
not be related to the implementation of Project Longevity, but instead to a
reduction in criminal activity that is unrelated to violence reduction efforts.
As a robustness check, we operationalize New Haven’s level of street
group criminality using police arrest data listing multiple offenders for a sin-
gle incident—namely, co-arrests. Although a crude measure of street group
criminality, the group nature of gangs and their activities is well documented
(Decker, 1996; Klein & Crawford, 1967; Reiss, 1988), and, as such, the num-
ber of co-arrest incidents provides an indicator of New Haven’s group crime
patterns. In particular, the leveraging of data on the underlying group pro-
cesses of crime in New Haven matches up well with the group focus of
Longevity, as well as the relatively fluid nature of New Haven street groups
(see Sierra-Arevalo & Papachristos, 2015a).
Over the observation period, the individual arrest rate is stable (b = −1.00,
SE = 1.43, p = .491), suggesting that the overall level of “criminality” in New
Haven (or police enforcement) showed little change over the observation
period. In contrast, there is a slight but statistically significant decrease in the
level of co-offending for each month (b = −0.72, SE = 0.20, p = .001). Prior
to Longevity, there were approximately 110.82 (SD = 13.84) co-offending
arrests per month, but this figure dropped 14.4% after the start of the program
to approximately 94.00 (SD = 15.89) co-offending arrests per month.
Table 2. Regression Model Predicting the Effect of Project Longevity on Number
of Incidents Per Month in Hartford and New Haven.
Hartford
bSE p b SE p
Total (n = 443) GMI (n = 311)
Intercept 12.28 1.46 < .001 8.80 1.21 < 0.001
Ar1 0.40 0.14 .005 0.28 0.15 .064
Intervention −2.60 2.11 .219 −2.33 1.77 .189
New Haven Total (n = 388) GMI (n = 230)
Intercept 8.20 1.74 < 0.001 6.68 1.26 < 0.001
Ar1 0.21 0.16 .011 0.48 0.15 .002
Hartford 0.28 0.12 .025 0.16 0.09 .079
Intervention −3.47 1.36 .011 −5.04 1.51 .001
Note. GMI = group member involved; Ar1 = autoregressive parameter.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 13
The number of monthly shootings and homicides in New Haven is not
related to individual arrest rates (r = .20, p = .217) but is correlated with the
number of recoded co-offenses (r = .52, p = .001); the relationship is even
stronger for GMI incidents (r = .59, p < .001). To parse out the relationship
between co-offending and GMI shootings, we estimate a model predicting
shootings and homicides in New Haven, controlling for co-offending. The
results shown in Table 3 suggest that the overall level of co-offending in
New Haven has a statistically significant effect on the total number of
shootings: An increase of 100 recorded co-offenses is related to an increase
of nine shootings. However, Table 3 also shows a statistically significant
intervention effect of Longevity on GMI shootings and homicides, even
when controlling for the effect of co-offending. This further suggests that
the observed decrease in the number of GMI incidents observed during
Project Longevity cannot be completely explained by a decrease in group
offending.
Possible Issue 3: Bias in GMI Identification
Another threat to the validity of our findings stems from the possibility that
our dependent variable undercounts the total number of GMI incidents in
New Haven. For example, the incomplete information inherent to ongoing
investigations might prevent some shooting from being conclusively identi-
fied as GMI during shooting reviews, ultimately resulting in an artificially
low number of GMI shootings. Under such a condition, a significant decrease
in GMI incidents might only be observed because the larger, total number of
GMI incidents is not taken into account.
We examine this concern using a logistic regression that identifies possible-
GMI shootings based not on shooting review classification, but instead on the
demographic characteristics of the victim (race, age, gender), the identification
Table 3. Regression Model Predicting the Effect of Project Longevity on the
Number of Incidents Per Month in New Haven Using Co-Offenses as a Control.
Shootings and homicides
(New Haven)
Total (n = 388) GMI (n = 230)
b SE p b SE p
Intercept 1.65 4.50 .714 3.16 3.58 .378
Ar1 0.24 0.15 .121 0.42 0.17 .011
Co-offenses 0.09 0.04 .025 0.05 0.03 .157
Intervention −2.60 1.53 0.09 −4.86 1.41 .001
Note. GMI = group member involved; Ar1 = autoregressive parameter.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
14 Crime & Delinquency
of the suspect, whether the shooting was fatal or non-fatal, and the location of
the shooting.11 This strategy identified 323 possible-GMI incidents—93 more
than were identified as GMI by police during the shooting review process.
Next, we subject the police-defined GMI and possible-GMI trend to the
same time series regression models described above to determine the effect
of Longevity on these two GMI formulations. Table 4 shows that, after
including the 93 additional cases of possible-GMI incidents with the police-
identified cases, there is still a statistically significant decrease in the number
of incidents after the start of Longevity. Even with a more lenient identifica-
tion of GMI, there is a significant negative programmatic effect of Longevity
on GMI shootings and homicides.
Multiple Explanations and Summary Model
As a final robustness check, we take into account the three threats to validity
we have discussed and control for them in a single model. Table 5 presents
the results from a series of models predicting total shootings and homicides,
GMI (as defined by police), and possible-GMI shootings that use the shoot-
ing trends in Hartford and the level of co-offending in New Haven as statisti-
cal controls. The results find a continued, but somewhat reduced, Longevity
effect when controlling for these additional parameters. As seen in the last
row in Table 5, even after controlling for these additional parameters, the
implementation of Project Longevity is associated with 2.4 total shootings
(fatal and non-fatal) per month, 4.6 fewer GMI incidents, and 3.1 fewer pos-
sible-GMI incidents after the start of the program. Although such an analysis
still lacks true causal power, the robustness of the observed intervention
effect to different statistical conditions and parameters strongly suggests that
the observed decline in GMI shootings and homicides in New Haven can be
attributed to the enactment of Project Longevity.
Table 4. Regression Model Predicting the Effect of Project Longevity on GMI and
Possible-GMI Incidents Per Month in New Haven.
Shootings and
homicides (New Haven)
GMI
(n = 230)
Possible-GMI
(n = 323)
b SE p b SE p
Intercept 8.05 1.09 < .001 10.20 0.93 < .001
Ar1 0.50 0.15 .001 0.31 0.14 .027
Intervention −5.33 l.61 .001 −4.94 1.26 .001
Note. GMI = group member involved; Ar1 = autoregressive parameter.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 15
Conclusion
Project Longevity is a focused deterrence effort aimed at reducing gun vio-
lence in New Haven, Connecticut. Emulating previously evaluated programs
from other cities (e.g., Braga, Apel, & Welsh, 2013; Braga, Kennedy, &
Waring, et al., 2001; Corsaro & Engel 2015; Engel et al., 2013; Papachristos
& Kirk 2015; Papachristos et al., 2007), Longevity performed call-ins with
street group members between November 2012 and April 2014 to leverage
group dynamics and curb the violence on New Haven streets. In these call-
ins, a combination of law enforcement, social service providers, and com-
munity members spoke with street group members to deliver a unified
message to group members that the gun violence must stop, there is help for
those who want it, and those who choose to continue committing acts of vio-
lence will meet with swift legal consequences (Crandall & Wong, 2012;
NNSC, 2013).
To test whether Project Longevity had a significant, negative effect on
GMI shootings and homicides in New Haven, we examined data on lethal
and non-lethal shootings in the city from January 2011 until April 2014 using
a series of ARIMA models. The results of our analysis suggest that, even
accounting for a variety of alternative explanations, the implementation of
Project Longevity in New Haven, Connecticut, was associated with a reduc-
tion of nearly five GMI shootings and homicides per month. These results
support a growing body of empirical research that confirms the efficacy of
focused deterrence strategies for reducing gun violence in American cities
(Braga & Weisburd, 2012, 2015). Moving away from traditional deterrence
and broken windows approaches that privilege broadly applied police sweeps
or enforcement of minor offenses, New Haven’s Project Longevity is one
Table 5. Regression Model Predicting the Effect of Project Longevity on Number
of Incidents Per Month in New Haven, Using Hartford Incidents and New Haven
Co-Offenses as Controls.
Shootings and
homicides
(New Haven)
Total
(n = 388)
GMI
(n = 230)
Possible-GMI
(n = 323)
b SE p b SE p b SE p
Intercept 0.88 4.32 .839 2.34 3.45 .498 −0.18 3.98 .963
Ar1 0.21 0.16 .170 0.44 0.16 .007 0.28 0.16 .068
Hartford 0.22 0.12 .069 0.14 0.09 .137 0.14 0.11 .196
Co-offenses 0.07 0.39 .066 0.04 0.03 .220 0.08 0.04 .028
Intervention −2.38 1.45 .101 −4.58 1.44 .001 −3.12 1.42 .029
Note. GMI = group member involved; Ar1 = autoregressive parameter.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
16 Crime & Delinquency
more instance of how targeting specific offenders, in this case members of
violent street groups, can significantly enhance public safety.
To be sure, the design of Project Longevity is not ideal for programmatic
evaluation. As we have described, the pressing nature of the gun violence
problem and the relatively small number of actively violent street groups
identified during New Haven’s group audit led to nearly all of the groups
being invited to a call-in during the intervention period. Because of this, we
are unable to compare the effect of the Longevity intervention on a set of
treatment and control groups or neighborhoods as is more common in quasi-
experimental designs. Despite this issue with the design of the intervention,
we account for three alterative explanations for the observed decrease in GMI
shootings and homicides in New Haven: (a) a general decrease in gun vio-
lence that extends beyond New Haven, (b) a reduction in group offending
patterns in New Haven, and (c) bias resulting from an imperfect police-
defined GMI measure. Even after accounting for these alternative explana-
tions, a significant decrease in GMI incidents after the implementation of
Project Longevity in New Haven is observed.
Even accounting for these plausible alternative explanations, we cannot
be unequivocally sure that the effects we attribute to the implementation of
Longevity are not caused by unobserved changes during the intervention
period, such as in activities by pre-existing social service programs. Before
and during the implementation of Longevity, New Haven had dozens of
local social service organizations providing job training, drug and alcohol
counseling, mental health services, and adult education. In fact, the existing
network of social service providers was an integral part of the Longevity
strategy, with several local agencies and programs partnering with
Longevity. Although previous work finds that provision of social services
is not responsible for the observed declines in gun violence attributed to
focused deterrence initiatives (Engel et al., 2013), we cannot conclude that
the observed Longevity effect does not overlap with other unmeasured pro-
grams, policies, or services.
We believe that our results bolster an already strong case for future imple-
mentations of focused deterrence strategies in cities across the United States,
especially similar medium-sized cities that are less likely to have their gun
violence problems discussed in the same breath as metro areas such as
Chicago, Los Angeles, or Indianapolis (McGarrell, Chermak, Wilson, &
Corsaro, 2006; Papachristos & Kirk, 2015; Tita, Riley, Ridgeway, Grammich,
& Abrahamse, 2010). Similar to other smaller cities such as Rockford, Illinois
(Corsaro, Brunson, & McGarrell, 2013), and Lowell, Massachusetts (Braga,
McDevitt, & Pierce, 2006), the results of New Haven’s Project Longevity
indicate that focused deterrence strategies can effectively bolster public
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 17
safety outside big city contexts, and suggest that the underlying street group
dynamics that drive gun violence are not unique to major cities.
However, as discussed in previous evaluations of focused deterrence ini-
tiatives (e.g., Braga, Kennedy, Waring, et al., 2001), each city in which such
a strategy is implemented faces a unique set of challenges. Although the
underlying logic of the focused deterrence strategy is relatively stable
between sites, the implementation of the strategy is necessarily adapted to the
unique constellation of street groups in each city. Knowing that the structure
of street groups is not consistent across cities (McGloin, 2005; Sierra-Arevalo
& Papachristos, 2015b), we echo calls by other scholars (Braga & Weisburd,
2015) and suggest that the success of Project Longevity in New Haven,
Connecticut, should spur the adoption of focused deterrence into the public
safety repertoires of other cities, but also a deeper exploration of the underly-
ing mechanisms and group processes that generate the “spillover effects” of
focused deterrence strategies (Braga et al., 2013).
With New Haven as only the first of three cities that are part of the state-
wide Longevity plan, Connecticut is a promising place to continue the study
of focused deterrence strategies’ effects on gun violence, as well as their
implementation. Although Hartford and Bridgeport are 20 and 39 miles
from New Haven, respectively, and each city has gun violence in need of
attention, each city also has a unique collection of law enforcement, service
providers, community members, and street groups. How the differences in
the context of where a focused deterrence strategy is carried out affects pro-
gram implementation presents an exciting area for future research to explore,
with such research providing useful information for law enforcement, com-
munities, and policy makers alike as they work together to reduce urban gun
violence.
Keeping these future avenues of research in mind, our findings provide
evidence that focused deterrence strategies such as Project Longevity are a
viable and efficacious step away from overly broad policies and policing
practices such as stop and frisk or police sweeps. With public, academic,
and policy-making attention firmly trained on the need for change in how
the criminal justice system acts on the lives of community residents, an
approach that funnels limited resources toward those most likely to be
involved in gun violence as victims and offenders provides a promising
way forward. Although such programs are not panaceas for the underlying
issues that engender gun violence—ineffectual schools, broken homes,
unemployment, poverty—they are at the very least an effective way to
address the gun violence that is symptomatic of these broader social ills,
while also minimizing the number of community members caught up in the
criminal justice system.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
18 Crime & Delinquency
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research,
authorship, and/or publication of this article: This project was supported by
Cooperative Agreement Number 2012CKWXK039 awarded by the Office of
Community Oriented Policing Services, U.S. Department of Justice. The opinions
contained herein are those of the author(s) and do not necessarily represent the official
position or policies of the U.S. Department of Justice. References to specific agencies,
companies, products, or services should not be considered an endorsement by the
author(s) or the U.S. Department of Justice. Rather, the references are illustrations to
supplement discussion of the issues.
Notes
1. Homicide rates were drawn from the FBI’s Uniform Crime Reports for 2011.
2. Group Violence Intervention (GVI; at one time called the “group violence
reduction strategy” or GVRS) is the name of the strategy currently being
implemented in dozens of U.S. cities with the support of the National Network
for Safe Communities (NNSC) and John Jay College of Criminal Justice.
For more information, see http://nnscommunities.org/our-work/strategy/
group-violence-intervention
3. See Kennedy (1997) for a thorough discussion of the “pulling levers” approach.
4. In the case of New Haven, researchers from Yale University and University of
New Haven worked with representatives from the New Haven Police Department,
Connecticut Probation and Parole, and the U.S. Attorney’s Office.
5. The definition of “group” does not necessarily overlap with official legal or
departmental definitions of a “gang.” Instead, it can be “any set, clique, or crew
of individuals” that commit crimes together (NNSC, 2013, p. 36). See Sierra-
Arevalo and Papachristos (2015a) for a discussion of the importance of this
broader definition of groups/gangs for avoiding “nation conflation” in audits.
6. For a much more thorough description of the call-in, see Crandall and Wong
(2012).
7. The location of the call-in did shift during this period, moving from the alder-
manic chambers to the basement meeting room of neighborhood churches. The
structure and content of the call-in itself remained constant.
8. The AutoRegressive Integrated Moving Average (ARIMA) model is written as Ŷt
= µ + bTxTx + b1X1 . . . bnXn + ϕ1Yt − 1, which is Y regressed on itself lagged by one
period (t − 1), and estimated by ϕ1. The effect of other parameters (X), including
treatment, can be estimated by b.
9. The Autocorrelation Function has a value of .47 for total shootings, .77 for GMI
incidents. Stationarity is not detected when using the Augmented Dickey-Fuller
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 19
test (Total: = −3.70, p = .038; GMI: ADF = −3.59, p = .046).
10. Although Hartford was selected to participate in Project Longevity, the program
did not begin in the city until April 2014. Although our analysis does include
April 2014, Hartford has hosted only a single call-in and sustained efforts at con-
tinuing them in the same systematic fashion as New Haven ebbed and flowed;
no other call-ins in Hartford were conducted during the observation period of
this study. As such, we do not consider Hartford to have received the full Project
Longevity “treatment.” Regardless, all models were run with and without inci-
dents from April 2014; the results of our analysis are not sensitive to the inclu-
sion of these shootings and homicides during that month.
11. Results from the logistic regression used to predict possible-GMI incidents
according to the characteristics of shooting are available upon request.
References
Akers, R. L. (1999). Criminological theories: Introduction and evaluation (2nd ed.).
Chicago, IL: Fitzroy Dearborn.
Blumstein, A., & Rosenfeld, R. (1998). Explaining recent trends in U.S. homicide
rates. The Journal of Criminal Law and Criminology, 88, 1175-1216.
Braga, A. A., Apel, R., & Welsh, B. C. (2013). The spillover effects of focused deter-
rence on gang violence. Evaluation Review, 37, 314-342.
Braga, A. A., Hureau, D. M., & Papachristos, A. V. (2014). Deterring gang-involved
gun violence: Measuring the impact of Boston’s operation ceasefire on street
gang behavior. Journal of Quantitative Criminology, 30, 113-139.
Braga, A. A., Hureau, D., & Winship, C. (2008). Losing faith? Police, Black churches,
and the resurgence of youth violence in Boston. Cambridge, MA: Rappaport
Institute for Greater Boston, Harvard Kennedy School.
Braga, A. A., Kennedy, D. M., Piehl, A. M., & Waring, E. J. (2001). Reducing
gun violence: The Boston gun project’s operation ceasefire. Washington, DC:
National Institute of Justice.
Braga, A. A., Kennedy, D. M., Waring, E. J., & Piehl, A. M. (2001). Problem-oriented
policing, deterrence, and youth violence: An evaluation of Boston’s operation
ceasefire. Journal of Research in Crime & Delinquency, 38, 195-225.
Braga, A. A., McDevitt, J., & Pierce, G. L. (2006). Understanding and preventing gang
violence: Problem analysis and response development in Lowell, Massachusetts.
Police Quarterly, 9, 20-46.
Braga, A. A., Papachristos, A. V., & Hureau, D. (2009). The concentration and stabil-
ity of gun violence at micro places in Boston, 1980-2008. Journal of Quantitative
Criminology, 26, 33-53.
Braga, A. A., & Weisburd, D. L. (2012). The effects of focused deterrence strate-
gies on crime: A systematic review and meta-analysis of the empirical evidence.
Journal of Research in Crime & Delinquency, 49, 323-358.
Braga, A. A., & Weisburd, D. L. (2015). Focused deterrence and the prevention of
violent gun injuries: Practice, theoretical principles, and scientific evidence.
Annual Review of Public Health, 36, 55-68.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
20 Crime & Delinquency
Cook, P. J., & Laub, J. H. (1998). The unprecedented crime epidemic in youth vio-
lence. Crime & Justice, 24, 27-64.
Corsaro, N., Brunson, R. K., & McGarrell, E. F. (2013). Problem-oriented policing
and open-air drug markets examining the Rockford pulling levers deterrence
strategy. Crime & Delinquency, 59, 1085-1107.
Corsaro, N., & Engel, R. S. (2015). Most challenging of contexts: Assessing the
impact of focused deterrence on serious violence in New Orleans. Criminology
& Public Policy, 14, 471-505.
Crandall, V., & Wong, S.-L. (2012). Practice brief: Call-in preparation and execu-
tion. New York, NY: National Network for Safe Communities, Center for Crime
Prevention and Control, John Jay College of Criminal Justice.
Decker, S. H. (1996). Collective and normative features of gang violence. Justice
Quarterly, 13, 243-264.
Engel, R. S., Corsaro, N., & Tillyer, M. S. (2010). Evaluation of the Cincinnati
Initiative to Reduce Violence (CIRV). Cincinnati, OH: University of Cincinnati
Policing Institute.
Engel, R. S., Tillyer, M. S., & Corsaro, N. (2013). Reducing gang violence using
focused deterrence: Evaluating the Cincinnati Initiative to Reduce Violence
(CIRV). Justice Quarterly, 30, 403-439.
Gibbs, J. P. (1975). Crime, punishment, and deterrence. New York, NY: Elsevier.
Goldstein, H. (1979). Improving policing: A problem-oriented approach. Crime &
Delinquency, 25, 236-258.
Kelling, G. L., & Moore, M. H. (1988). The evolving strategy of policing (No. 4).
Washington, DC: U.S. Department of Justice, Office of Justice Programs,
National Institute of Justice.
Kennedy, D. M. (1997). Pulling levers: Chronic offenders, high-crime settings, and a
theory of prevention. Valparaiso University Law Review, 31, 449-484.
Kennedy, D. M., Braga, A. A., & Piehl, A. M. (1997). The (un)known universe:
Mapping gangs and gang violence in Boston. In D. L. Weisburd & J. T. McEwen
(Eds.), Crime mapping and crime prevention (pp. 219-262). New York, NY:
Criminal Justice Press.
Kennedy, D. M., Piehl, A. M., & Braga, A. A. (1996). Youth violence in Boston:
Gun markets, serious youth offenders, and a use-reduction strategy. Law and
Contemporary Problems, 59, 147-196.
Klein, M. W., & Crawford, L. Y. (1967). Groups, gangs, and cohesiveness. Journal of
Research in Crime & Delinquency, 4, 63-75.
McGarrell, E. F., Chermak, S., Wilson, J. M., & Corsaro, N. (2006). Reducing homi-
cide through a “lever-pulling” strategy. Justice Quarterly, 23, 214-231.
McGloin, J. M. (2005). Policy and intervention considerations of a network analysis
of street gangs. Criminology & Public Policy, 4, 607-635.
National Network for Safe Communities. (2013). Group violence intervention: An
implementation guide. Washington, DC: U.S. Department of Justice, Office of
Community Oriented Policing Services.
Papachristos, A. V. (2009). Murder by structure: Dominance relations and the social
structure of gang homicide. American Journal of Sociology, 115, 74-128.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
Sierra-Arevalo et al. 21
Papachristos, A. V., Hureau, D. M., & Braga, A. A. (2013). The corner and the crew:
The influence of geography and social networks on gang violence. American
Sociological Review, 78, 417-447.
Papachristos, A. V., & Kirk, D. S. (2015). Changing the street dynamic: Evaluating
Chicago’s group violence reduction strategy. Criminology & Public Policy, 14,
525-558.
Papachristos, A. V., Meares, T. L., & Fagan, J. (2007). Attention felons: Evaluating proj-
ect safe neighborhoods in Chicago. Journal of Empirical Legal Studies, 4, 223-272.
Reiss, A. J., Jr. (1988). Co-offending and criminal careers. Crime & Justice, 10, 117-170.
Rosenbaum, D. P. (1988). Community crime prevention: A review and synthesis of
the literature. Justice Quarterly, 5, 323-395.
Sierra-Arevalo, M., & Papachristos, A. V. (2015a). Applying group audits to prob-
lem-oriented policing. In G. Bichler & A. Malm (Eds.), Preventing crime with
social network analysis (pp. 27-46). Boulder, CO: Lynne Rienner.
Sierra-Arevalo, M., & Papachristos, A. V. (2015b). Social network analysis and
gangs. In D. C. Pyrooz & S. H. Decker (Eds.), Handbook of gangs (pp. 157-177).
New York, NY: Wiley.
Skogan, W. G. (1990). Disorder and decline: Crime and the spiral of decay in American
neighborhoods. Berkeley and Los Angeles, CA: University of California Press.
Stafford, M. C., & Warr, M. (1993). A reconceptualization of general and specific
deterrence. Journal of Research in Crime and Delinquency, 30, 123-135.
Tita, G. E., & Radil, S. M. (2011). Spatializing the social networks of gangs to explore
patterns of violence. Journal of Quantitative Criminology, 27, 521-545.
Tita, G. E., Riley, K. J., Ridgeway, G., Grammich, C. A., & Abrahamse, A. (2010).
Reducing gun violence: Results from an intervention in East Los Angeles. Santa
Monica, CA: RAND.
Tonry, M., & Moore, M. H. (1998). Youth violence in America. In M. Tonry & M.
H. Moore (Eds.), Crime and justice (Vol. 24, pp. 1-26). Chicago, IL: University
of Chicago Press.
Weisburd, D., Telep, C. W., Hinkle, J. C., & Eck, J. E. (2010). Is problem-oriented
policing effective in reducing crime and disorder? Criminology & Public Policy,
9, 139-172.
Zimring, F. E. (2007). The great American crime decline. Oxford, UK: Oxford
University Press.
Zimring, F. E., & Hawkins, G. J. (1973). Deterrence: The legal threat in crime con-
trol. Chicago, IL: University of Chicago Press.
Author Biographies
Michael Sierra-Arevalo is a PhD candidate in sociology at Yale University, and an
affiliate fellow at the Institution for Social and Policy Studies. His research interests
include police, gangs, violence prevention, and how police officers’ perceptions of
danger structure their behavior and interactions with the public.
Yanick Charette is a post-doctoral associate in the Department of Sociology and the
Yale Institute for Network Science. His research focuses on gun violence patterns and
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
22 Crime & Delinquency
how this violence can be conceptualized as a form of interaction between offenders.
His main interest is the dynamicity of the collaborative process and retaliation. He
submitted his doctoral thesis in criminology at the University of Montreal, under the
supervision of Prof. Carlo Morselli and Prof. Pierre Tremblay. His thesis was on the
effect of unreported crimes on our understanding of criminal behaviors.
Andrew V. Papachristos is an associate professor in the Department of Sociology at
Yale University, director of the Center for Research on Inequalities and the Life
Course (CIQLE), and a faculty affiliate at the Institution for Social and Policy Studies
at Yale University. His research focuses on social networks, neighborhoods, street
gangs, and interpersonal violence.
at Yale University Library on May 2, 2016cad.sagepub.comDownloaded from
