Our framework extends prior work on

Our framework extends prior work on moderation and mediation in several key respects. First, the framework incorporates each of the logical possibilities that result when a moderator variable influences one or more of the paths of the basic mediated model in Figure 1A. Prior discussions of moderation and mediation have addressed only a subset of these possibilities. Second, we show that combining moderation and mediation does not yield a single path model but instead produces a set of models that each portray direct,indirect, and total effects at a particular level of the moderator variable. This perspective emphasizes that evidence for mediation varies according to the level of the moderator variable under consideration. Third, we point out that models that specify moderation of both paths of an indirect effect implicitly introduce a nonlinear effect for the moderating variable. This point has not been mentioned in discussions of such models (Baron & Kenny, 1986; James & Brett, 1984; Muller et al., 2005), yet it has important implications for conceptualizing and interpreting moderation of indirect effects. Finally, we demonstrate how to derive confidence intervals and conduct significance tests for direct, indirect, and total effects at selected levels of the moderator variable. These procedures have been discussed for mediated models that exclude moderation (MacKinnon et al., 2002; Shrout & Bolger, 2002; Sobel, 1982) but havenot been addressed for models that combine moderation and mediation.