Concerning the c′ Path: How PROCESS-based Mediation Undermines Experimental Methods

This paper critiques the widespread use of PROCESS Model 4 for testing mediation in experimental designs, arguing that the inclusion of the direct effect (c') path can undermine the very causal inferences experiments are designed to establish. In an experimental context, the independent variable (X) is an instrumental manipulation intended to induce systematic variance in a mediator (M), which subsequently influences a dependent variable (Y). By including the c' path, PROCESS Model 4 statistically controls for the exact variance the research design intentionally induces, often leaving Path b to model only residual, non-experimental covariance. To demonstrate these shortcomings, the paper compares PROCESS Model 4 with a serial causal chain approach using structural equation modeling (SEM), which excludes the c' path to allow for a test of model fit. Formal analyses of four simulated datasets reveal that while both methods converge when data align with theory, they diverge sharply when data are inconsistent. Specifically, results indicate that PROCESS Model 4 consistently produces significant indirect effects even when the underlying data structure contradicts the theoretical model, whereas the SEM-based serial causal chain accurately identifies such inconsistencies through fit diagnostics and residual covariance. The paper concludes by providing specific best-practice recommendations, suggesting that researchers prioritize models that align with the methodological assumptions of experimental control to ensure more reliable social scientific evidence.