Mediation, moderation and moderated mediation

Many PLS-SEM papers go beyond direct effects. Mediation explains how an effect works (through which construct), and moderation explains when it is stronger or weaker. The course model has both, and they combine into moderated mediation.

Mediation: the indirect effects

An indirect effect is the product of the paths along a route. For PE → BI → UB it is 0.351 × 0.440 = 0.155. Because the product of two coefficients is not normally distributed, it is tested with its bootstrap confidence interval, which must exclude zero.

Types of mediation

Following Zhao, Lynch and Chen (2010), the type depends on the indirect and direct effects:

• Indirect-only (full) mediation: significant indirect effect, no (or non-significant) direct effect.
• Complementary (partial) mediation: both significant and in the same direction.
• Competitive (partial) mediation: both significant but in opposite directions.
• Direct-only: significant direct effect, no indirect effect.
• No effect: neither is significant.

Effort expectancy shows complementary mediation: it affects use partly through intention (0.099) and partly directly (0.107); the indirect part accounts for 48% of the total effect (variance accounted for, VAF = 0.479). For PE, SI and TR, the model has no direct path to UB, so their effects on use are indirect by design. To claim full mediation in your paper, you would need to include the direct paths and show that they are not significant.

Moderation: when is PE more important?

The interaction PE × TR → BI has β = 0.211, 95% CI [0.120, 0.297], p < .001: trust significantly strengthens the effect of performance expectancy on intention. Its f² is 0.075; for interactions, Kenny (2018) suggests 0.005, 0.01 and 0.025 as small, medium and large, so this is a large interaction effect.

The interaction coefficient alone is hard to communicate. Simple slopes show the PE → BI path at different levels of trust:

• low trust (−1 SD): β = 0.140, p = .027
• average trust: β = 0.351, p < .001
• high trust (+1 SD): β = 0.562, p < .001

Among users with high trust, a one-SD increase in performance expectancy raises intention by more than half an SD; among users with low trust, the effect is a quarter as large. The managerial message: a useful app is not enough; users must also trust it. The slope plot in the Moderation tab shows the lines fanning out, which is how you should present the interaction in your paper.

Two technical notes. When a moderator is in the model, the main effect of PE (0.351) is its effect at the average level of trust. And moderators should be justified by theory, like any path; testing every possible interaction inflates false positives.

Moderated mediation

Because the moderated path (PE → BI) is part of the mediation route PE → BI → UB, the indirect effect itself depends on trust. The index of moderated mediation (Hayes, 2015) tests this: here it is 0.093, 95% CI [0.049, 0.140], so the indirect effect changes significantly with trust. The conditional indirect effects make it concrete: 0.062 at low trust, 0.155 at average trust and 0.248 at high trust. For users with high trust, performance expectancy is four times as effective in driving actual use.

Reporting

Bootstrapped indirect effects showed that behavioural intention mediated the effects of performance expectancy (β = 0.155, 95% CI [0.108, 0.205]), effort expectancy (β = 0.099, [0.055, 0.146]), social influence (β = 0.081, [0.040, 0.129]) and trust (β = 0.080, [0.038, 0.127]) on use behaviour. For effort expectancy, the significant direct effect indicated complementary mediation. Trust positively moderated the effect of performance expectancy on intention (β = 0.211, 95% CI [0.120, 0.297], f² = 0.075): the effect was 0.140 at low, 0.351 at mean and 0.562 at high trust. The index of moderated mediation was significant (0.093, 95% CI [0.049, 0.140]).