Indirect Effects in Path Analysis

Abstract

The indirect effect in path analysis has been defined as equal to the difference between the correlation coefficient and the path coefficient. This definition has led to ambiguous and inconsistent interpretations of indirect effects across various applications. This paper shows that what was earlier defined as the indirect effect may, in fact, be composed of three parts: (1) the indirect effect, (2) spurious association due to joint dependence on a prior variable, and (3) association due to the correlation between predetermined variables. We argue for a redefinition of the indirect effect so that only those paths which can be traced through intervening variables are included in the indirect effect term. Indirect effects can then be interpreted consistently across all applications as being indirect causal effects.