Mediators, Moderators, and Tests for Mediation.

TLDR: In this paper, it is suggested that no middle ground exists between exploratory and confirmatory (causal) analysis, and that attempts to explain how mediation processes occur require well-specified causal models.

Abstract: : The following points are developed. First, mediation relations are generally thought of in causal terms. Influences of an antecedent are transmitted to a consequence through an intervening mediator. Second, mediation relations may assume a number of functional forms, including nonadditive, nonlinear, and nonrecursive forms. Special attention is given to nonadditive forms, or moderated mediation, where it is shown that while mediation and moderation are distinguishable processes, a particular variable may be both a mediator and a moderator within a single set of functional relations. Third, current procedures for testing mediation relations in industrial and organizational psychology need to be updated because these procedures often involve a dubious interplay between exploratory (correlational) statistical tests and causal inference. It is suggested that no middle ground exists between exploratory and confirmatory (causal) analysis, and that attempts to explain how mediation processes occur require well-specified causal models. Given such models, confirmatory analytic techniques furnish the more informative tests of mediation. (Author)