Integrating Recursive and Non-Recursive Models in Experimental Mediation Analysis

Experimental mediation analysis within experimental or quasi-experimental designs enhances internal validity by ensuring that treatment assignment is exogenous through randomization. Yet even in such designs, the possibility of reverse causation cannot be fully ruled out. Existing causal mediation frameworks typically impose a unidirectional causal structure, leaving the directionality between mediator and outcome unresolved. This paper integrates recursive and non-recursive modeling approaches to show that, in the absence of confounding, recursive models can parsimoniously represent reciprocal processes at their steady state. We further introduce an instrument-free method for estimating reciprocal effects at equilibrium and propose a unified framework linking causal mediation analysis with reciprocal modeling traditions. Together, these contributions provide a clearer conceptual and methodological foundation for interpreting mediation mechanisms under ambiguous causal ordering, illuminating dynamic mutual influence in processes such as attitudes, affect, and self-regulation.