Modeling the Gang Ecosystem

Predictive analytics has become one of the hottest trends in crime reduction. In recent years, US cities including Los Angeles, Chicago, Santa Cruz, and Memphis have received attention for their efforts to build models that predict crime before it happens. One fascinating element of this trend is the possible applicability of models from disciplines such as geology and biology to urban policy.

In “The Ecology of Gang Territorial Boundaries,” researchers P. Jeffrey Brantingham, George E. Tita, Martin B. Short, and Shannon E. Reid argue that gang territory development and violence can be forecast using models of animal competition. The authors of the paper, which was published in the July 2012 issue of Criminology, argue that this has surprising implications for how we understand gangs and leads to counterintuitive policy implications for reducing gang violence.

Brantingham et al. developed a model of gang competition based on a well-established model of interspecies competition. The model they use, the Lotka-Volterra model, was developed in the early 20^th century to predict how the populations of predator species and prey species influenced one another.

To test whether this model could predict gang behavior, the researchers used data on Los Angeles gangs from surveys done in 2003 and 2005. These earlier studies identified 29 active street gangs in the Hollenbeck neighborhood of East Los Angeles, determined the rivalries between the gangs, and found each gang’s geographic “anchor point.” The studies also examined the 1,208 violent crimes attributable to street gangs that occurred in Hollenbeck between 1999 and 2002 and determined which gangs were involved, which gang was the “suspect,” and which the “victim.”

Brantingham et al. focus on 13 gangs from the earlier research to test their model. Using the “anchor points” in Hollenbeck as starting points for gangs, they run the model and compare the results against the actual data. They asked two questions: Does gang-related violence occur where the model says it should, specifically, on the border between gang territories? And can the model predict the distribution of violence around the gang border?

To answer the former, the authors calculated the distance from the predicted boundary between two gangs to each crime involving that pair of gangs. The median distance was less than one-tenth of a kilometer, meaning that the crimes were in fact centered on the predicted boundaries. To address the latter, the authors looked at the spread of gang-related violence around the border. The shape of the distribution coincided with key predictions from the model. In the press release announcing the article, UCLA put this latter finding in layman’s terms: “[T]he theory predicted that 58.8 percent of the crimes would occur within one-fifth of a mile of the border between two gangs — or just under two blocks — and 87.5 percent within two-fifths of a mile of the border — or just over three blocks. …In fact, the team found that 58.2 percent occurred within two blocks of the border and 83.1 percent within just over three blocks of the border.”

The accuracy of the model has surprising implications for understanding how gangs function. In the model, competition between gangs is the sole factor driving the development of gang territory. While gangs in the model will grow without competition, they will not form discrete territories. However, if there is even slight competition between two gangs, the researchers’ model predicts that the gangs will quickly expand until their borders meet.

A more provocative implication – and one that may have greater relevance to policy – is that efforts to reduce rivalry between gangs may actually increase gang violence. According to the model, gangs with more intense rivalries have sharper, more abrupt borders, which limit interaction between the gangs. The authors suggest that attempts to lessen rivalry between gangs could have the perverse effect of increasing overlap between gang territories and increasing violence.

While the implications of this study are quite interesting, there are a number of questions that must be answered before the findings can be more broadly applied. The authors concede that Hollenbeck is somewhat unique because it is bounded on all sides by major roads; it forms a limited gang ecosystem, with little interaction between Hollenbeck gangs and outside gangs. The study is also fairly limited in scope. It does not attempt to model when violence will occur, how many people will join gangs, or any gang criminal activity beyond inter-gang violence.

Caveats aside, this study provides a proof point for the general concept that cities can be understood through models and methods from other disciplines – an approach that has implications well beyond crime policy. Hopefully this study will lead to further research that allows policymakers see the invisible relationships that connect urban ecosystems.

Feature Photo: cc/(substance)