Gang Membership, Drug Selling, and Violence in Neighborhood Context

Abstract

A prominent perspective in the gang literature suggests that gang member involvement in drug selling does not necessarily increase violent behavior. In addition it is unclear from previous research whether neighborhood disadvantage strengthens that relationship. We address these issues by testing hypotheses regarding the confluence of neighborhood disadvantage, gang membership, drug selling, and violent behavior. A three‐level hierarchical model is estimated from the first five waves of the 1997 National Longitudinal Survey of Youth, matched with block‐group characteristics from the 2000 U.S. Census. Results indicate that (1) gang members who sell drugs are significantly more violent than gang members that don’t sell drugs and drug sellers that don’t belong to gangs; (2) drug sellers that don’t belong to gangs and gang members who don’t sell drugs engage in comparable levels of violence; and (3) an increase in neighborhood disadvantaged intensifies the effect of gang membership on violence, especially among gang members that sell drugs.

Keywords:

• neighborhood disadvantage
• social learning
• gang membership
• drug selling
• violence
• conditional effects
• fixed‐effects
• hierarchical modeling

Acknowledgments

This paper was supported by grant #5 R03 DA15717‐02 from the National Institute of Drug Abuse, NIH. We also thank the Bureau of Labor Statistics and the Center for Human Resources Research (CHRR) at The Ohio State University for their assistance with data.

Notes

1. Throughout the manuscript we frequently refer to the relationship between “neighborhood disadvantage,” “gang membership,” “drug selling,” and “violence.” When those terms are used we are actually referring to the relationship between current gang membership and/or current drug selling and the current frequency of violence. We avoid using the word “current” in the text because its continued use throughout the manuscript is repetitious.

2. When we refer to the effect of neighborhood disadvantage in the text we are referring to the socio‐economic conditions in proximity to the residential address, but not necessarily the socio‐economic conditions where drugs are being sold.

3. Goldstein (1985) defines systemic violence as “traditionally aggressive patterns of interaction within the system of drug distribution and use.”

4. Subculture is defined herein as a network of individuals involved in drug selling or who are gang members. Sometimes drug and gang subcultures are partially overlapping such as when gang members sell drugs.

5. NLSY97 over‐sampled African American and Hispanic respondents. We elected against using the over‐sample because the analytic focus of the paper does not specifically address race/ethnicity. The so called “cross‐sectional” sample, which we analyze here, is also preferred because it is self‐weighted‐the US population 12–16 is sampled with probability of selection proportionate to size. The results are substantively identical when the over‐sample is included in our analysis.

6. The violence measure available to us does not distinguish whom attacks are directed towards or why they were carried out. For some research purposes, such as determining the precise proportion of violence that is gang‐related or instrumental, this could be an important limitation. However, the issue is not particularly relevant in this analysis because the SSSL model predicts increased violence upon the confluence of gang membership, drug selling, and neighborhood disadvantage and is agnostic about whom it is directed towards or why.

7. For example, Y_ijk is the number of violent attacks committed during an interval of time with length m_ijk termed the “exposure.” In our case, Y_ijk is the number of violent attacks committed during one year for each person j within each neighborhood k, so that m_ijk = 1 (i.e., constant exposure). According to our Level‐1 model, the predicted value of Y_ijk when m_ijk = 1 will be the event rate, λ_ijk . Specifically, to denote that Y_ijk has a Poisson distribution with exposure m_ijk and event rate per time period of λ_ijk we write (Raudenbush & Bryk, 2002):

(1)

The expected value and variance of Y_ijk , given the event rate λ_ijk, are then:

(2)

Thus, the expected number of events, Y_ijk , per unit of time i for person j within neighborhood k is the event rate, λ _ijk , multiplied by the exposure, m_ijk . At Level‐1 we model:

(3)

where η_ijk is the log of the event rate. Note that while λ_ijk is constrained to be non‐negative, log(λ_ijk ) can take on any value. The predicted log event rate can be converted to an event rate by generating λ_ijk = exponential{η_ijk }.