This study introduces a new set of weighting procedures for revealing the mediation mechanism in multi-level settings. These methods are illustrated through an investigation of whether the impact of a system-wide policy change on student outcomes is mediated by policy-induced peer composition change. When the policy changed not only lower-achieving students' course-taking but also the ability composition of math classes among other concurrent changes, unpacking the overall policy impact on math achievement is challenging. To illustrate, the causal questions focus on decomposing the total policy effect into the indirect effect mediated by peer composition change and the direct effect of the policy for the subpopulation of lower achieving students. Specifically, the authors ask: (1) Did the increase in peer ability mediate the policy effect on these students' math achievement? (2) Would the policy have a direct effect on these students' math achievement if their peer composition had remained unchanged by the policy? (Contains 1 table and 1 figure.)