MEDIATING AND MODERATING EFFECTS
Another statistical method that researchers use to understand the mechanism through which IVs affect the DVs is the moderation and mediation tests. A classic discussion of moderator and mediator analyses can be found in Baron and Kenny (1986), which has recently been expanded upon by Fairchild and MacKinnon (2014). A moderator variable is one that affects the strength of the relationship between two other variables. The moderator variable can either be qualitative in nature (e.g., gender) or quantitative (e.g., level of reinforcement) that influences the association between two other variables and points to why these effects might hold and its inclusion changes the strength of association between the two other variables.
In contrast, a mediator variable explains the mechanism or process that underlies the relationship between the two variables. For example, level of education might be seen as a mediator variable if it explained the relationship between SES and health-related behaviors. In many cases, mediators may describe underlying psychological processes (e.g., beliefs, emotions) and may explain the relationship between two variables. Trauma might affect someone’s ability to return to work after a natural disaster. However, this effect may be mediated by autonomic reactivity that may explain the effect of trauma on the ability to return to work. With the help of newer statistical software, we are now able to test more complicated hypotheses that test the simultaneous influence of multiple mediators and moderated mediation effects as well as mediated moderation effects (see Preacher, Rucker, & Hayes, 2007). For example, a home-based intervention study used SEM techniques to simultaneously model multiple mediators and found that identification of personal goals, enhanced depression knowledge, and reduced anxiety each independently mediated the effects of the intervention program on depression (Gitlin, Roth, & Huang, 2014).