SEMinR - Quick Guide - Mediation Analysis using SEMinR in R

TL;DR

Bootstrapping is used to test mediation significance by evaluating indirect effects and their confidence intervals.

Briefing Cornell Notes

Briefing

The mediation test hinges on one practical question: does collaborative culture carry the effect of vision onto organizational performance? After bootstrapping the SEMinR model, the indirect pathway from Vision to organizational performance through the mediator (collaborative culture) comes out significant—its estimate is 0.165, with results indicating it exceeds the 1.96 threshold and the confidence interval does not include zero. That combination is the statistical green light for mediation in this setup.

With the indirect effect confirmed, the analysis turns to whether the mediation is partial or full. That depends on the direct effect from Vision to organizational performance: if the direct path remains significant alongside the indirect path, mediation is partial; if the direct path is insignificant, mediation is complete (often called full mediation). Running the bootstrap results for the direct path shows Vision → organizational performance is significant as well, with a t statistic reported as significant and a bootstrap interval that again does not span zero.

Taken together, both pathways—(1) the indirect effect through collaborative culture and (2) the direct effect from Vision to organizational performance—are significant. The conclusion is therefore partial mediation: collaborative culture explains part of how vision translates into performance, but vision also has a remaining direct influence on organizational performance that is not fully absorbed by the mediator.

Methodologically, the workflow is straightforward and repeatable. First, the model is estimated in SEMinR with a measurement model and a structural model that includes three core links: Vision → collaborative culture (the IV-to-mediator path), collaborative culture → organizational performance (the mediator-to-DV path), and Vision → organizational performance (the direct effect). Next, the estimated model is bootstrapped (using a summary object for the bootstrap output) with iteration limits kept under 300. The bootstrap output is then queried for total indirect effects and, when needed, specific indirect effects. In this illustration, there is only one mediator, so total indirect effects and specific indirect effects coincide; the significance check is performed by verifying that the indirect effect is greater than 1.96 and that zero is not contained in the interval.

The key takeaway is that mediation analysis in SEMinR is less about running a single test and more about reading two layers of bootstrap evidence: significance of the indirect effect establishes that the mediator matters, while significance (or not) of the direct effect determines whether the mediator fully accounts for the relationship or only partially does so.

Cornell Notes

The analysis tests whether collaborative culture mediates the relationship between Vision and organizational performance. After estimating the SEMinR model and bootstrapping it, the indirect effect from Vision to organizational performance through collaborative culture is significant: the estimate is 0.165, it exceeds the 1.96 benchmark, and the confidence interval does not include zero. The next step checks the direct effect Vision → organizational performance using bootstrap results. Because the direct effect is also significant (interval excludes zero), the mediation is classified as partial rather than full. This workflow relies on bootstrapped significance for both indirect and direct paths, with total indirect effects sufficient here because there is only one mediator.

How does the bootstrap output determine whether the indirect effect is significant in this mediation model?

The indirect effect is read from the bootstrap-based summary object under total indirect effects. Significance is assessed by checking whether the indirect effect estimate clears the 1.96 threshold and whether the confidence interval does not include zero. In the example, the indirect effect from Vision to organizational performance via collaborative culture is 0.165, and the interval contains no zero, so the indirect effect is significant.

Why do total indirect effects and specific indirect effects match when there is only one mediator?

With a single mediator, there is only one indirect pathway from the exogenous variable to the dependent variable. That means the “total indirect effect” and the “specific indirect effect” refer to the same indirect route, so both quantities coincide. The transcript notes that multiple mediators would require specific indirect effects to separate different mediation channels.

What rule distinguishes partial mediation from full mediation?

Partial mediation occurs when both the indirect effect (through the mediator) and the direct effect (IV → DV) are significant. Full mediation occurs when the indirect effect is significant but the direct effect is not significant. The transcript applies this by first confirming the indirect effect is significant, then checking whether the direct Vision → organizational performance path is significant.

How is the direct effect significance used to classify the mediation outcome here?

After confirming the indirect effect is significant, the direct path Vision → organizational performance is tested using bootstrap results. The direct effect is significant, with a t statistic reported as significant and a bootstrap interval that does not include zero. Because both indirect and direct effects are significant, the mediation is partial.

What structural paths must be included in the SEMinR model for this mediation test?

The structural model includes three links: Vision → organizational performance (direct effect), Vision → collaborative culture (IV-to-mediator path), and collaborative culture → organizational performance (mediator-to-DV path). These correspond to the exogenous variable, the mediator, and the dependent variable used in the mediation checks.

Review Questions

In this example, what two bootstrap checks are required to label mediation as partial versus full?
If the model had multiple mediators, what additional quantities would need to be examined beyond total indirect effects?
Why does the confidence interval including or excluding zero matter for both indirect and direct effects?

Key Points

1
Bootstrapping is used to test mediation significance by evaluating indirect effects and their confidence intervals.
2
The indirect effect from Vision to organizational performance through collaborative culture is significant (estimate 0.165; exceeds 1.96; interval excludes zero).
3
Mediation classification depends on whether the direct effect Vision → organizational performance is significant after accounting for the mediator.
4
Because the direct effect is also significant (interval excludes zero; t statistic significant), the result is partial mediation.
5
With only one mediator, total indirect effects and specific indirect effects are effectively the same; multiple mediators would require specific indirect effects.
6
The SEMinR mediation workflow follows a repeatable sequence: estimate the SEM model, summarize results, bootstrap the estimated model, then read indirect and direct effects from the bootstrap summary objects.

Highlights

Collaborative culture mediates the Vision → organizational performance relationship: the indirect effect is 0.165 and is significant because it clears 1.96 and the interval excludes zero.

The mediation is partial, not full, because the direct Vision → organizational performance path remains significant in the bootstrap results.

The workflow relies on bootstrapped inference: significance is judged by thresholds (1.96) and whether zero falls inside the confidence interval.

With one mediator, total indirect effects are sufficient; specific indirect effects become necessary when multiple mediators exist.