In program evaluation, randomized experiments are considered the gold standard.  While randomized experiments, can reveal whether an intervention work, it often can tell us little about why it might work.  In such cases, analysts often turn to mediation analysis. Traditionally in the social sciences, mediation analysis has been formulated, understood, and implemented within the framework of linear structural equation models.  The proposed paper demonstrates that this is problematic for three reasons; the lack of a general definition of causal mediation effects independent of a particular statistical model, the inability to specify the key identification assumption, and the difficulty of extending the framework to nonlinear models.  In this paper, we propose an alternative approach that overcomes these limitations.  Our approach is general because it offers the definition, identification, estimation, and sensitivity analysis of causal mediation effects without reference to any specific statistical model.  Further, the approach explicitly links these four elements closely together within a single framework, and suggests a sensitivity analysis that enables applied researchers to assess the robustness of their empirical conclusions to assumption violations.  The paper also considers various experimental designs could be used to estimate mediation effects.  Considering policy-based interventions, the paper illustrates the approach by applying it to the Job Search Intervention Study (JOBS II).