24. SPSS AMOS | Concept of Mediation Analysis (Part 1)

Mediation analysis in AMOS is built to answer a simple but powerful question: does the effect of an independent variable on an outcome happen only directly, or does it travel indirectly through a third variable (the mediator)? In this framework, the mediator “intervenes” in the relationship between X and Y by receiving influence from X and then transmitting it to Y. That setup matters because it distinguishes mechanisms (how effects occur) from mere associations (whether effects exist).

The core bookkeeping in mediation uses three effect types. The total effect of X on Y is the combined impact when no mediator path is specified—often represented as the relationship between X and Y alone (labeled C in the lecture). Once the mediator M is included, the direct effect becomes the remaining X→Y link while controlling for M (labeled C′, or “C complement”). The indirect effect captures the mediated pathway X→M→Y and is computed as the product of two path coefficients: A (X→M) and B (M→Y). In other words, indirect effect = A×B, and total effect = direct effect (C′) + indirect effect (A×B).

Different mediation patterns follow from whether the direct and indirect components are significant and whether they move in the same or opposite directions. Partial mediation occurs when both the indirect effect (A×B) and the direct effect (C′) are significant, meaning some influence runs through the mediator while some still passes directly. Full mediation describes the case where the indirect effect is significant but the direct effect is insignificant—so the effect of X on Y is carried entirely through M. The lecture also distinguishes complementary mediation (direct and indirect effects share the same sign, such as both positive) from competitive mediation (direct and indirect effects have opposite signs, such as a negative direct effect paired with a positive indirect effect).

Testing mediation has evolved. The classic Baron and Kenny (1986) approach used a four-step sequence: first verify that X affects Y (the C path) without the mediator; then show X affects M (A path); then show both X→M and M→Y are significant (A and B); finally test the model with X, M, and Y together to determine whether mediation exists and what type. Critiques targeted the reliance on significance testing of unstandardized coefficients and, in particular, the use of the Sobel test. The lecture notes that the C path no longer needs to be significant for mediation to occur, because suppressor effects can hide the total effect even when the indirect effect is real. It also rejects the requirement that A and B must each be individually significant, since the indirect effect depends on their product.

The revised, more accepted strategy focuses directly on the indirect effect A×B and its sampling distribution. Because Sobel-style assumptions are considered flawed, bootstrap methods are recommended: repeatedly resample the data with replacement (often thousands of times) to build a confidence interval for the indirect effect. If the interval excludes zero, mediation is supported. The lecture then sets up an AMOS example: testing whether authentic leadership influences life satisfaction indirectly through self-efficacy, while also including a direct path from authentic leadership to life satisfaction to determine whether the mediation is partial, full, complementary, or competitive.