Many state and municipal governments have efforts underway to create meta-analytic models that incorporate state/local costs. These models are intended to synthesize program effects from the extant literature and model the cost-effectiveness of local implementation. While this evidence-based approach to policymaking is an important policy advance, using these models to formulate crime policy requires careful attention to each of several sources of heterogeneity from variation in model inputs. Meta analytic model inputs that have important variation in effect size (commonly modeled), costs (rarely modeled), and benefits (never modeled). In particular, variation in benefits is a critical source of heterogeneity in cost-benefit modeling and has a known bias. Recent studies have shown that harms from criminal victimization (and thus the benefits from preventing criminal victimization) are highly skewed such that for some crime types the median harm is as much as an order of magnitude smaller than the mean. Thus, meta-analyses that use mean benefit as an input are likely to greatly overestimate benefits in small samples (e.g. where few program replications are undertaken). We report on a Bayesian approach that incorporates the full distribution of costs, effect sizes and benefits to model the cost-effectiveness of community-based substance abuse treatment in Washington, DC and, using the same data as prior studies, find smaller benefit-cost ratios and substantial increases in the risk that the costs of the program will exceed the benefits.