Assessing causal effects under treatment heterogeneity: an evaluation of a CCTV program in Detroit