| In recent years, criminologists have become increasingly interested in how the social organization of urban neighborhoods can contribute to the reduction of crime (see Sampson, Raudenbush, and Earls, 1997, for a review). This line of research poses challenging problems of causal inference. While changes in social organizational properties of neighborhoods may cause changes in criminal offending and crime victimization, such changes may also reflect the demographic composition of the neighborhood. Moreover, past crime events may affect the motivation and capacity of residents to organize effectively to combat crime. Thus, selection bias and endogeneity challenge the validity of causal inferences. In this paper, we adapt recent statistical methods for causal inference to the multilevel setting in which potential offenders (or crime victims) are nested within neighborhoods. We illustrate how propensity score stratification, inverse probability of treatment weighting, selection modeling, and sensitivity analysis may be adapted to improve the validity of causal inferences about the effects of neighborhood social processes on crime. We illustrate these methods with longitudinal data from the Project on Human Development in Chicago Neighborhoods.
(Return to Program Resources)