Measuring Evidence for Mediation in the Presence of Measurement Error

Mediation analysis empirically investigates the process underlying effect of an experimental manipulation on a dependent variable interest. In simplest mediation setting, treatment can affect through mediator (indirect effect) and/or directly (direct effect). However, what appears to be indirect in standard may reflect data-generating without mediation, including possibility reversed causal ordering measured variables, regardless statistical properties estimate. To overcome this indeterminacy where possible, authors develop insight that statistically reliable total effect, combined with strong evidence for conditional independence and outcome given mediator, is unequivocal as model into operational procedure. This particularly helpful when theory insufficient definitely causally order or before believed mediator. The procedure combines Bayes factors principled measures degree support independence, latent modeling account measurement error discretization fully Bayesian framework. reanalyze set published studies illustrate how their approach facilitates stronger conclusions.