Mediation analysis is a widely used technique within the psychological sciences and has been shown to be an effective tool to evaluate explanatory pathways between predictors and outcomes. Multiple effect size metrics have been developed; however, mediation analysis has been slow to develop accessible, interpretable effect size metrics in the cases of categorical (or otherwise non-normally distributed) mediators and/or outcomes. Herein, we propose the use of average marginal effects within mediation analysis to alleviate these issues—termed Marginal Mediation Analysis. The method provides interpretable indirect and direct effect size estimates in the same units as the outcome even when mediators and/or outcomes are categorical, a count measure, or another non-normal distribution. The approach is shown to fit the causal definitions of mediation analysis. We further present results of Monte Carlo simulations that show the utility of the proposed method in psychological research. We also discuss the assumptions inherent in the approach. We conclude by showing an application of it to adolescent health-risk behavior data (n = 13,600), demonstrating the increased interpretability and information provided compared to other common approaches.