QUANTIFYING GANG LOCATIONS: SYSTEMATICALLY TESTING VALIDITY USING A PARTIAL TEST OF MESSICK’S UNIFIED PERSPECTIVE

Abstract

Gangs pose a serious problem in 21st century policing. After spending forty years studying them, Klein describes why we still have a lot to learn about gangs. Specifically, he suggests: “Too little attention has been paid to the communities in which gangs appear. Observing and understanding neighborhoods is far more complex than studying their gangs; yet it is communities that spawn gangs and must inevitably be the proximal focus for controlling them” (Klein 2007: xiv). The neighborhood complexity to which Klein refers has not been adequately addressed. The field lacks strong theoretical development to guide decisions about conceptualizing and operationalizing gangs from an ecological perspective. As a reflection of that lack of guidance, researchers today employ several alternate indicators of gang ecologies. This study seeks to identify the consequences of variations in how gangs are conceptualized and measured at an ecological level. Researchers model gangs in substantially different ways, using dissimilar indicators, spatial scales, and levels of measurement. These variations may generate disparities in empirical results and different estimates of gang ecologies without a clear rationale for, or understanding of, the implications of their selection. The analysis examines two central research questions: (1) Do indicators of gang ecologies identify gangs in similar ways and with results that are consistent across spatial scales? (2) Can indicators predict the presence versus absence of gangs—as a binary outcome—as well as predict continuous gang outcomes, such as the number of gangs present or the geographic size of the gang ecology, with results consistent across spatial scales? Arrest data and gang data provided by the Philadelphia Police Department (PPD) included information on 3,996 gang members who belonged to 113 gangs. PPD data indicated (1) where gang members live, (2) gang arrests (N=7,488 from 2012-2015), (3) crime incidents that involve a gun (N=26,865 from 2012-2015), and (4) PPD defined gang set space boundaries. This study examined the validity of each of these indicators when each is used to define gang ecologies. The analysis plan was guided by Messick’s unified perspective of construct validity (Messick 1995) and included two types of analyses. The first analysis employed a series of 60 regression models. Model comparisons tested various aspects of construct validity as proposed by Messick. The second analysis developed an algorithm creating gang set space polygons using either the locations where gang members live or the locations of gang-related crime. The set space polygons created by this algorithm were compared to the PPD set space polygons. The degree to which the gang set space polygons created by the algorithm overlapped with the gang set space polygons defined by PPD functioned as another validity test. The results of the regression analysis revealed the home address and arrest variables better explained the spatial distribution of the PPD gang set space locations than the gun crime variable. The link between the gang indicators and the PPD identified set space polygons, however, was complex. Oftentimes, the home address data and the arrest variables significantly predicted a binary gang outcome—whether one or more gangs existed in an area; but those variables could not significantly estimate a continuous gang outcome, i.e. how many gangs were present, or geographic size of gang set space. This means the home address and arrest variables have limited ability to explain the spatial distribution of the gang set space boundaries defined by PPD. The spatial analysis used an algorithm to approximate gang set space locations. The results indicate that locations of gang members’ homes and of gang arrests both can approximate the PPD reported gang set space locations equally well. However, the spatial overlap between the PPD reported set space and the approximated set space locations proved relatively small. Although the approximated gang set space polygons usually did overlap with the PPD reported gang set space polygons to some extent, the mean overlap was 10% using the home address locations and 7% using the arrest data. The policy or practice usefulness of the indicators used here could be minimal. Although the overlap wasn't perfect, the algorithm was able to identify the general locations where gangs exist using only the home address of gang members or gang arrest data. This study contributes to our knowledge about gang measurement at the ecological level. Conceptualizing and measuring gangs in a theoretically driven way is critical to the development of effective policies to control and prevent violence, fear, and other social harms caused by gangs. The results of this study will pave the way for future research to build on our understanding of how gangs link to crime, how community level dynamics work to foster or prevent gang activity, and ultimately, how to reduce and prevent gang problems.