#SmartPLS4 Series 26 - Mediation Model Analysis with Multiple Mediators

TL;DR

Mediation in SmartPLS 4 is assessed by first testing specific indirect effects for each mediator pathway (IV→mediator→DV).

Briefing Cornell Notes

Briefing

The session lays out a practical workflow for testing mediation with multiple mediators in SmartPLS 4, then demonstrates how to classify each mediator pathway as full mediation, partial mediation, or no mediation using bootstrapping results. The core takeaway is straightforward: mediation hinges on whether the indirect effect (the product of the IV→mediator path and mediator→DV path) is statistically significant, and the final label depends on whether the direct effect (IV→DV while mediators are included) is also significant.

The model used in the example links Internal Marketing (IM) to Organizational Performance (OP) through four mediators: Organizational Support, Organizational Commitment, Organizational Culture, and Internal Service Quality (ISQ). Hypotheses are framed so each mediator carries the effect of IM to OP—e.g., Organizational Support mediates IM’s relationship with OP; Organizational Commitment does the same; Organizational Culture acts as a mediator; and ISQ also mediates. The session emphasizes that mediation hypotheses are built by first identifying the mediator and then specifying the IV→mediator and mediator→DV relationships that form the indirect effect.

Methodologically, the procedure starts with the indirect effect. In SmartPLS terms, the indirect effect for a given mediator is computed as P1×P2, where P1 is the path coefficient from the independent variable to the mediator and P2 is the path coefficient from the mediator to the dependent variable. With multiple mediators, SmartPLS reports “specific indirect effects” for each mediator pathway separately. If an indirect effect is significant, the mediator is considered to carry an indirect influence from IM to OP.

After confirming which indirect effects are significant, the analysis checks the direct effect: the IV→DV path coefficient in the presence of all mediators. This direct effect is the key to mediation type. If both the indirect effect and the direct effect are significant, the result is partial mediation. The session further distinguishes partial mediation as “complimentary” when the signs of the relevant path coefficients align positively (no negative sign among the multiplied components), producing a positive indirect effect alongside a significant direct effect. If the indirect effect is significant but the direct effect is not significant, the result is full mediation—meaning IM’s impact on OP operates entirely through the mediator(s). If the indirect effect is not significant, mediation is absent; if the direct effect is also not significant, there is effectively no effect.

In the worked example, SmartPLS bootstrapping is run with 5,000 samples (complete one-tailed percentile bootstrap). The results show that the indirect effects for ISQ→OP, along with the other mediator pathways (IM→Organizational Support→OP, IM→Organizational Commitment→OP, and IM→Organizational Culture→OP), are significant—so each mediator carries an indirect effect from IM to OP. The direct effect IM→OP is also significant, leading to partial mediation. Using the ISQ pathway as the illustration, the session multiplies the relevant coefficients (e.g., P1=0.680, P2=0.245, and P3=0.552) and notes that the absence of negative signs indicates complimentary partial mediation. The session closes by summarizing that this combination—significant specific indirect effects plus a significant direct effect—yields partial mediation, while other significance patterns produce full mediation or no mediation.

Cornell Notes

The mediation workflow in SmartPLS 4 starts by testing specific indirect effects for each mediator pathway. For any mediator, the indirect effect equals the product of the IV→mediator path coefficient and the mediator→DV path coefficient (P1×P2). After identifying significant indirect effects, the analysis checks the direct effect (IV→DV) while mediators are included. If both indirect and direct effects are significant, the result is partial mediation; if only the indirect effect is significant, it is full mediation; if the indirect effect is not significant, mediation is absent (and if the direct effect is also not significant, there is no effect). In the example model, Internal Marketing affects Organizational Performance through four mediators, with significant indirect effects and a significant direct effect indicating complimentary partial mediation via Internal Service Quality.

How is the indirect effect for a mediator computed in SmartPLS mediation analysis?

For each mediator pathway, SmartPLS uses the product of two path coefficients: P1 (IV→mediator) and P2 (mediator→DV). The indirect effect is P1×P2. Conceptually, this represents the extent to which the IV’s influence on the DV runs through that mediator. With multiple mediators, each mediator has its own P1×P2 product, reported as a specific indirect effect.

What decision rule determines whether a mediator is supported?

A mediator is supported when its specific indirect effect is statistically significant. SmartPLS provides these specific indirect effects directly after bootstrapping. If the indirect effect for a mediator is significant, the IV’s impact on the DV passes through that mediator; if it is not significant, that mediator pathway does not show mediation.

How does SmartPLS distinguish full mediation from partial mediation?

After checking indirect effects, the analysis tests the direct effect (IV→DV) in the presence of the mediators. If the indirect effect is significant and the direct effect is also significant, the result is partial mediation. If the indirect effect is significant but the direct effect is not significant, the result is full mediation—meaning the IV’s effect on the DV operates entirely through the mediator(s).

What does “complimentary partial mediation” mean in the example?

In the worked example, the indirect effect and the direct effect are both significant, so the mediation is partial. It is labeled “complimentary” because the multiplied path coefficients for the indirect effect carry positive alignment (no negative sign among the components used to compute the indirect effect). Using the ISQ pathway as the illustration, the session multiplies coefficients such as 0.680 and 0.245 and then considers the direct path coefficient (e.g., 0.552) as significant, with no negative sign indicated, leading to complimentary partial mediation.

What bootstrapping setup is used to test mediation significance?

Bootstrapping is run with 5,000 samples using a complete one-tailed percentile bootstrap. The procedure then uses the resulting significance tests for specific indirect effects and the direct effect to classify mediation outcomes.

What does “no mediation” look like when indirect and direct effects differ?

If the indirect effect is not significant, mediation is not present for that pathway. If the direct effect is also not significant, the model indicates no effect of the IV on the DV. If the direct effect is significant while the indirect effect is not, the IV influences the DV directly without mediation.

Review Questions

In a multiple-mediator model, what are the two quantities you must check to classify mediation type, and how do their significance patterns map to full vs partial mediation?
How would you interpret a significant specific indirect effect but an insignificant direct effect for the same mediator pathway?
Why does the indirect effect depend on multiplying two path coefficients, and what do those two paths represent in the model?

Key Points

1
Mediation in SmartPLS 4 is assessed by first testing specific indirect effects for each mediator pathway (IV→mediator→DV).
2
Compute the indirect effect as the product of the IV→mediator path coefficient (P1) and the mediator→DV path coefficient (P2).
3
Run bootstrapping (in the example: 5,000 samples, complete one-tailed percentile bootstrap) to obtain significance for indirect and direct effects.
4
Classify mediation using both indirect and direct effects: significant indirect + significant direct = partial mediation; significant indirect + insignificant direct = full mediation.
5
If the indirect effect is not significant, that mediator pathway does not show mediation, regardless of the direct effect.
6
When both direct and indirect effects are significant, the session’s example labels the result complimentary partial mediation when the multiplied path signs align positively.
7
With multiple mediators, SmartPLS reports each mediator’s specific indirect effect separately, then evaluates the direct effect with all mediators included.

Highlights

Each mediator pathway’s mediation strength is determined by a specific indirect effect, built from the product of two path coefficients (IV→mediator and mediator→DV).

Full vs partial mediation hinges on whether the direct effect (IV→DV) remains significant after including mediators.

The example uses bootstrapping with 5,000 samples and then finds significant indirect effects for all four mediators, plus a significant direct effect—yielding partial mediation.

Complimentary partial mediation is identified when both indirect and direct effects are significant and the indirect-effect path coefficients align without negative sign conflicts.