Abstract
The group-based violence intervention model is predicated on the assumption that individuals who hear credible messages of consequences for further violence will deliver the message to other group members. Using social network analysis, we develop an algorithm of who should receive the message to maximize the spread of the message among the remaining group members. Using a sample of gangs in four different cities we show how the reach of actual call-ins were suboptimal compared to our suggested algorithm. Using simulations, we further show that typically only around a third of the group needs to be delivered the message to achieve complete coverage of the network. We find that even when limiting possible invitees to those under supervision large proportions of groups can be reached if the invitee list is data driven.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Note this is not about sanctions in terms of the probability of being caught given a particular act, which offenders do appear to update their sanction risk (Anwar & Loughran, Citation2011; Piquero et al., Citation2011). This statement is in regards to the actual sanctions themselves, for example, if you changed a mandatory minimum for an aggravated assault from 5 to 10 years offenders would unlikely be aware of the change, or if you asked a group member with a prior felony what the punishment for carrying a firearm is they would be unlikely to know.
2 The particular sites under study were not implemented directly by Kennedy’s National Network for Safe Communities (NNSC), but his research group was instrumental in motivating the project but his research group was instrumental in motivating the project to upstate New York police agencies (via grants from the New York State Division of Criminal Justice Services). In some cases NNSC conducted group audit sections directly, and in some cases NNSC conducted group audit sections. As such we believe the sites to hold quite true to Kennedy’s vision for how a group involved intervention should be conducted.
3 Although even those under supervision are not always mandated to attend (Hamilton et al., Citation2018).
4 Although to date the majority of such evaluations are based on quasi-experimental designs (Roman, Klein, & Wolff, Citation2018).
5 While McCord (Citation2003) discussed this in terms of outreach workers unintentionally solidifying or creating new gang connections, call-ins by law enforcement provide the same opportunity.
6 This is a high cost in the case that criminal justice practitioners actually visit the group involved member in their home. In the form of a letter a custom notification would be less costly (Hunt, Parast, & Weinberger, Citation2017). Though not tested we imagine the trade-off in a lesser cost could be lesser impact as well.
7 While this prevents those in state prison from being included as a potential node to be called-in, it does not eliminate those with the majority of time in local jails.
8 The nodes are drawn with slight transparency, as the force-based layout used in SPSS sometimes produced nodes that lie on coincident lines or in misleading places. While overall the network is portrayed in a fairly nice planar graph, there are a few misleading examples. For example, in City 2 gang 1, node 20 is connected to node 3, but not connected to nodes 29 or node 39.
9 Though we note custom notifications have utility beyond simply spreading the deterrence message. A thorough presentation of the benefits of incorporating custom notifications into GVI can be found in Kennedy and Friedrich, (Citation2014).
10 This is an estimate by the police department. They did not retain records on who was invited and did not show up, only a record of who showed up.
11 This can be calculated by noting that given a set of n individuals invited to a call-in, there are a total number of potential 2n observed sets of people who actually showed up to a call-in. For any particular permutation, you can calculate the probability of observing that exact outcome by calculating where n is the number of individuals invited, k is the number of individuals who showed up, and p is the probability of showing up. One just calculates the reach for each permutation, and then calculates the expected reach using the probability as the weight.
12 Given two nodes a and b, the Jaccard coefficient is where
is the neighbor set of node x and
and
denote the union and intersection of those sets, respectively (Liben-Nowell & Kleinberg, Citation2007).
13 We hope that the algorithm has utility beyond just criminal justice settings though. For one example, it could be used by epidemiologists identifying particular users in needle exchange networks who may be used to spread information to other drug users (Davidson, Scholar, & Howe, Citation2011; Valente, Foreman, Junge, & Vlahov, Citation1998).
14 There have been several successful focused deterrence initiatives at smaller cities (under 150,000 population) more similar to our sample of cities in upstate New York (Corsaro, Brunson, & McGarrell, Citation2013; Corsaro et al., Citation2012; Sierra-Arevalo, Charette, & Papachristos, Citation2017).
Additional information
Notes on contributors
Andrew P. Wheeler
Andrew P. Wheeler is an Assistant Professor of criminology at the University of Texas at Dallas in the School of Economic, Political and Policy Sciences. His research focuses on the spatial analysis of crime at micro places, evaluating crime reduction policies by police departments, and practical problems faced by crime analysts.
Sarah J. McLean
Sarah J. McLean is the Associate Director and the Director of Research and Technical Assistance at the Finn Institute. She holds a Ph.D. in criminal justice from the University at Albany, with a specialization in policy and process. At the Institute she designs and manages evaluative research on criminal justice strategies and interventions.
Kelly J. Becker
Kelly J. Becker is a Research Analyst with the Finn Institute. She holds dual Bachelor's degrees in Criminal Justice and Psychology from the University at Albany and obtained her Master's degree from The School of Criminal Justice at Albany. Her areas of interest include gang violence, focused deterrence, and police decision making.
Robert E. Worden
Robert E. Worden is the Director of the Finn Institute. He is also an associate professor of criminal justice and public policy at the University at Albany, State University of New York, on whose faculty he has served since 1990. He holds a Ph.D. in political science from the University of North Carolina at Chapel Hill, with specializations in public administration and public policy.