At the request of Chicago Public Schools (CPS) leadership, we, as well as the Boston Consulting Group, were asked to build models to predict which Chicago high school students were most at risk of being shot. We based these predictions on retrospective data covering shootings between September 2007 and October 2009. Working with CPS, we were able to assemble a wide range of covariates. These included fixed student characteristics such as race and gender, a large number of behavioral variables (e.g., school misconduct, past shootings, test scores, progress toward graduation), and a handful of school-level controls, such as the type of school (military, charter, alternative, etc.) and each school’s per capita shooting rate from the previous two years. All of these variables reflect information that was known prior to the fall of 2009.
Controlling for other factors, being male is the single most important predictor—virtually all students shot are male. Despite the fact that blacks have much higher raw victimization rates than other groups (black males have a mean victimization rate of 0.0080 compared to 0.0027 for Hispanics and 0.0004 for all others), the coefficient on black is small and statistically insignificant in the regression, as is the Hispanic coefficient. This implies that the differences in risk across race are largely being captured by other covariates. A number of behavioral factors predict violent victimization: serious misconduct at school, suspensions, and incarceration history. The strongest predictor is having spent time in a juvenile detention center, which raises the likelihood of being shot by 2.5 percentage points—a tenfold increase over the mean probability. Other factors correlated with victimization are being overage (implying that the student previously failed a grade), low test scores, suspensions, absences, and an indicator for whether the student had spent time in an adult detention center. The school-related variable with the greatest explanatory power is per capita shooting rate. A one-standard-deviation change in this variable increases the likelihood of being shot by 0.00075, or about one-fourth of the average victimization rate.
|Country:||United States of America|
|Institution:||American Economic Review: Papers and Proceedings |
|Author:||Dana Chandler, Steven D. Levitt, and John A. List|