9.8 Take-Home Points

• A causal or path model without causal feedback loops can be estimated as a series of regression models: one regression model for each variable that has at least one predictor in the path model.

• Unstandardized regression coefficients, standardized regression coefficients, and correlations can be multiplied to obtain indirect effects and indirect correlations.

• An indirect effect is a mediated effect. Variables that are at the same time predicted and predictors in an indirect effect are called mediators, intermediary variables, or intervening variables.

• Statistical inference on an indirect effect—its confidence interval and significance level—requires a sampling distribution of the size of the indirect effect. This distribution can be bootstrapped with the PROCESS macro (Hayes, 2013).

• Mediation is an intuitively appealing concept but it is difficult to establish. A causal interpretation of a regression coefficient requires a clear time order between predictor, mediator, and dependent variable, a clear theoretical and conceptual difference between these three variables, and the inclusion of all variables that may confound the effects of the predictor and mediator(s) in the regression models.

Read the little but very helpful book on the logic of causal order by James A. Davis (1985) for more information on causality and correlational analysis.