Network Exposure and Homicide Victimization in an African American Community
β57%
reduction in homicide odds per degree of separation from victim (OR = 0.43)
In a high-crime Chicago community (N = 3,718; 2006β2011), 41% of all gun homicides occurred in a co-offending network comprising <4% of the neighborhood population. Being in a network component with a homicide victim increases the annual homicide rate by 900% (from 55.2 to 554.1 per 100,000). Each social tie removed from a homicide victim decreases one's odds of victimization by 57% (OR = 0.430). Network variables absorb most individual risk factor effects, reducing model AIC from 910 to 415. Effects robust to neighborhood fixed effects. Source of the "900%" figure often misattributed to later papers.
Papachristos, A. V., & Wildeman, C. (2014). Network exposure and homicide victimization in an African American community. American Journal of Public Health, 104(1), 143β150.
Methodology
Logistic regression; co-offending network analysis
Sample
3,718 high-risk individuals; Chicago 2006β2011
Key finding
OR = 0.430 per degree; 900% rate increase for network members
Tragic, but Not Random: The Social Contagion of Nonfatal Gunshot Injuries
OR = 3.13
per 1Β° of network exposure to gunshot victims (N = 169,620, Chicago)
In Chicago's city-wide co-offending network (N = 169,620; 2006β2012), 70% of all nonfatal gunshot injuries occurred in networks comprising less than 6% of the city's population. Network exposure to gunshot victims predicts individual victimization after controlling for demographics and gang membership: OR = 3.13 at 1 degree, rising to OR = 14.68 at β€3 degrees (best-fitting model). Every 1% increase in exposure increases odds by 1.1%. Gang membership independently triples victimization risk (OR = 3.30). Effects extend to 2β3 degrees of separation. Rate for network members: 740.5 per 100,000 vs. city average of 62.1 (12Γ higher).
Papachristos, A. V., Wildeman, C., & Roberto, E. (2015). Tragic, but not random: The social contagion of nonfatal gunshot injuries. Social Science & Medicine, 125, 139β150.
Methodology
Logistic regression; affiliation exposure models
Sample
169,620 individuals; Chicago 2006β2012
Exposure OR at β€3 degrees
14.68 (AIC-best model)
Social Networks and the Risk of Gunshot Injury
β25%
reduction in gunshot odds per network degree from victim (OR = 0.754; Boston)
In Boston's Cape Verdean community, 85% of all gunshot injuries in a network of 763 individuals occurred within a single connected component. Each network step removed from a gunshot victim decreases odds of victimization by 25% (OR = 0.754; 95% CI, 0.654β0.869), controlling for age, arrest history, gang membership, and network density. Effect levels off after approximately 5 degrees of separation. Gang members show more pronounced social distance effects. Prior arrest more than doubles victimization odds (OR = 1.85). First study to use formal network models to analyze gunshot risk in a defined high-risk population.
Papachristos, A. V., Braga, A. A., & Hureau, D. M. (2012). Social networks and the risk of gunshot injury. Journal of Urban Health, 89(6), 992β1003.
Methodology
Rare-event logistic regression; network geodesic analysis
Sample
763 individuals; Boston Cape Verdean community 2008β2009
Key finding
OR = 0.754 per degree (25% reduction); 85% injuries in one network
Modeling Contagion Through Social Networks to Explain and Predict Gunshot Violence in Chicago, 2006β2014
63.1%
of gunshot violence episodes attributable to social contagion (N = 138,163)
Using a Hawkes process epidemic model fitted to 8 years of Chicago arrest and shooting data (N = 138,163), social contagion accounts for 63.1% of all gunshot violence episodes. Subjects were shot an average of 125 days after their network infector (median: 83 days). A combined contagion + demographics model identifies 53.3% more gunshot subjects than demographics alone when targeting the top 1% highest-risk individuals daily. 4,107 distinct contagion cascades detected; largest cascade: 469 subjects. Strongest evidence to date that violence follows epidemic-like transmission dynamics through interpersonal networks.
Green, B., Horel, T., & Papachristos, A. V. (2017). Modeling contagion through social networks to explain and predict gunshot violence in Chicago, 2006 to 2014. JAMA Internal Medicine, 177(3), 326β333.
Methodology
Hawkes point process; probabilistic contagion model
Sample
138,163 individuals; Chicago 2006β2014; 11,123 episodes
Contagion interval
Mean 125 days; median 83 days after infector
Connected in Crime: The Enduring Effect of Neighborhood Networks on the Spatial Patterning of Violence
Co-offending networks among 172,714 individuals (~6% of Chicago's population) link all city neighborhoods through short chains of co-arrest ties β all neighborhoods are "connected in crime" by just a few handshakes, regardless of geographic or social distance. Using exponential random graph models (ERGMs) and network autoregressive models with PHDCN data, the authors find: (1) neighborhood co-offending networks are stable over time; (2) generated by concentrated disadvantage, collective efficacy deficits, and endogenous network properties; (3) better predictors of crime distribution than spatial adjacency models. Bridges Social Disorganization theory with Network Contagion by showing how neighborhood structure generates the network topology that enables violence diffusion.
Papachristos, A. V., & Bastomski, S. (2018). Connected in crime: The enduring effect of neighborhood networks on the spatial patterning of violence. American Journal of Sociology, 124(2), 517β568.
Methodology
ERGM; network autoregressive models; PHDCN + CPD data
Sample
172,714 individuals; all Chicago neighborhoods; 1999β2004
Key finding
Network models outperform spatial adjacency for predicting crime distribution