Mediation Analysis with an Ordinal Outcome

The article presents concepts and methods for mediation analysis for an ordinal outcome. We give definitions of natural direct and indirect effects using counterfactuals for ordinal outcomes; in this context, there are potentially different effects for any two levels of the outcome, and we consider difference and ratio scales. The confounding assumptions required for identification are similar to that in the existing mediation analysis literature. We discuss different modeling strategies for estimation. Under a proportional odds model with a reference category that is common, the direct and indirect effects on a ratio scale can each be summarized by a single estimate and are available in closed form; otherwise the effects may differ across categories compared and can be obtained by numeric simulation methods.