Reporting SmartPLS3 Structural Model - Hypotheses Testing, Mediation, and Moderation Analysis

Q: How should R², Q², and SRMR be interpreted in this reporting approach?

R² indicates explained variance in endogenous constructs based on incoming paths; the transcript lists R² values such as PC = 0.277 (27.7%), RC = 0.196, DC = 0.292, and TP = 0.206. Q² assesses predictive relevance and must be greater than zero; it’s obtained via blindfolding. SRMR is used for model fit and should be less than about 0.10 (lower is better). Together, these metrics establish whether the model both explains and predicts.

Q: What statistical thresholds are used to decide whether a path hypothesis is supported?

A path is treated as significant when p ≤ 0.05 and the t statistic is at least 1.96 (the transcript also notes that t values below 1.96 and p values above 0.05 indicate insignificance). The write-up should explicitly label hypotheses as supported or not supported based on these criteria, even if the beta direction is negative or positive.

Q: How is mediation reported, and what does it mean when indirect effects are insignificant?

Mediation reporting requires three components for each mediator: total effect, direct effect, and specific indirect effect. The transcript’s example shows SL → TP has a significant total effect (β ≈ 0.396, p = 0), but the specific indirect effects through TC, RC, and PC are all insignificant (p > 0.05). That pattern means the effect of SL on TP does not operate through those mediators in this model, so mediation is not supported.

TL;DR

Separate direct path-coefficient reporting from moderation reporting; don’t mix interaction effects into the direct-effects table.

Briefing Cornell Notes

Briefing

Structural equation modeling in SmartPLS3 is reported through a disciplined sequence: clean the path-coefficient output for hypothesis testing, then document predictive power (R² and Q²) and model fit (SRMR), and finally report mediation and moderation with the same level of statistical detail (betas, t values, p values, and—when requested—bias-corrected confidence intervals). The core takeaway is that results should be presented in a reader-friendly table that separates direct effects from mediation and moderation effects, while still tying every claim to the underlying statistics.

For the direct relationships, the transcript recommends copying the path coefficients (original sample mean/β, t statistics, and p values) into Excel, formatting values to three decimals, and removing rows that are not relevant to the specific hypothesis set being reported. Moderation effects are handled separately rather than mixed into the direct-effect table. Hypotheses are then evaluated using common SmartPLS thresholds: significance is indicated when p ≤ 0.05 and the t statistic exceeds 1.96. Relationships that fail these thresholds are treated as insignificant (even if the beta direction is meaningful), and the write-up should explicitly distinguish supported versus not supported hypotheses.

After direct effects, the reporting shifts to the structural model’s explanatory and predictive performance. R² values are taken from the bootstrapping “quality criteria” section for endogenous constructs (in the transcript: PC1 = 0.277, RC = 0.196, DC = 0.292, and TP = 0.206). Each R² is interpreted as the proportion of variance explained by the incoming paths—e.g., the 27.7% change in PC attributed to the set of predictors feeding into it (including the moderator’s influence where applicable). Predictive relevance is assessed via Q² using blindfolding; Q² must be greater than zero for predictive relevance. Model fit is documented using SRMR, with the guideline that lower is better and values under about 0.10 (preferably closer to 0.08) are acceptable.

Hypothesis reporting is illustrated with examples: servant leadership (SL) shows a significant negative effect on relationship conflict (RC) and a significant negative effect on relationship conflict-related outcomes, while at least one SL-to-PC path is treated as insignificant due to p > 0.05 and t < 1.96. The transcript also emphasizes that some reviewers may require 95% confidence intervals, which are obtained from bootstrapping “path coefficients → confidence intervals → bias-corrected” (typically the 2.5% and 97.5% bounds) and added to the results table.

Mediation is reported by listing total, direct, and indirect effects for each mediator (TC, RC, and PC). The total effect of SL on team performance (TP) is significant (β ≈ 0.396, p = 0), but the specific indirect effects through TC, RC, and PC are all insignificant—meaning the influence does not pass through these mediators in this model. Moderation is then reported by focusing on interaction effects involving the moderator CH (community hostility). Only the SL → RC relationship is moderated in a statistically significant way; other moderation paths are insignificant. To interpret the moderation substantively, slope analysis is used: when community hostility is low, servant leadership more strongly reduces relationship conflict; when community hostility is high, the beneficial effect is dampened and relationship conflict can increase slightly, as reflected by the changing slope lines.

Cornell Notes

SmartPLS3 results should be reported in a structured order: (1) direct path coefficients for hypothesis testing, (2) structural model quality using R², Q², and SRMR, then (3) mediation and (4) moderation. Direct effects are judged significant when p ≤ 0.05 and t ≥ 1.96, and the transcript recommends separating moderation effects into their own reporting section. Predictive relevance requires Q² > 0 (from blindfolding), while SRMR should be below about 0.10 (lower is better). In the example model, servant leadership has a significant effect on relationship conflict, mediation through TC/RC/PC is not supported, and community hostility moderates only the SL → RC link. Slope analysis clarifies that low community hostility strengthens SL’s reduction of relationship conflict, while high community hostility dampens that effect.

What is the recommended workflow for reporting SmartPLS3 structural model results?

Start by formatting and copying the path coefficients (original sample mean/β, t values, p values) into Excel/Word for direct hypothesis testing. Clean the table by removing rows not needed for the direct-effect hypotheses and report moderation separately. Next, report structural model quality: R² for endogenous constructs, Q² from blindfolding for predictive relevance (Q² > 0), and SRMR from fit indices (ideally < 0.10, preferably closer to 0.08). After that, report mediation using total, direct, and indirect effects for each mediator. Finish with moderation by reporting interaction effects (β, t, p) and interpreting them using slope analysis.

How should R², Q², and SRMR be interpreted in this reporting approach?

R² indicates explained variance in endogenous constructs based on incoming paths; the transcript lists R² values such as PC = 0.277 (27.7%), RC = 0.196, DC = 0.292, and TP = 0.206. Q² assesses predictive relevance and must be greater than zero; it’s obtained via blindfolding. SRMR is used for model fit and should be less than about 0.10 (lower is better). Together, these metrics establish whether the model both explains and predicts.

What statistical thresholds are used to decide whether a path hypothesis is supported?

A path is treated as significant when p ≤ 0.05 and the t statistic is at least 1.96 (the transcript also notes that t values below 1.96 and p values above 0.05 indicate insignificance). The write-up should explicitly label hypotheses as supported or not supported based on these criteria, even if the beta direction is negative or positive.

How is mediation reported, and what does it mean when indirect effects are insignificant?

Mediation reporting requires three components for each mediator: total effect, direct effect, and specific indirect effect. The transcript’s example shows SL → TP has a significant total effect (β ≈ 0.396, p = 0), but the specific indirect effects through TC, RC, and PC are all insignificant (p > 0.05). That pattern means the effect of SL on TP does not operate through those mediators in this model, so mediation is not supported.

How is moderation reported and interpreted using slope analysis?

Moderation is reported by examining the interaction term involving the moderator CH (community hostility) for each relevant path. The transcript finds CH significantly moderates only the SL → RC relationship; other moderation paths are insignificant. Slope analysis then shows how the SL → RC relationship changes at different CH levels: when CH is low, higher SL corresponds to lower relationship conflict; when CH is high, the slope flattens or turns less favorable, indicating dampened or reversed benefit.

Review Questions

In SmartPLS reporting, what specific values must be documented to establish explanatory power, predictive relevance, and model fit—and what thresholds are used for each?
If the total effect of X on Y is significant but all specific indirect effects through mediators are insignificant, what mediation conclusion should be drawn?
For moderation, how do you decide which interaction effects to interpret, and what does slope analysis add beyond the beta/t/p results?

Key Points

1
Separate direct path-coefficient reporting from moderation reporting; don’t mix interaction effects into the direct-effects table.
2
Use p ≤ 0.05 and t ≥ 1.96 as the practical significance rule for supported versus unsupported hypotheses.
3
Report R² for endogenous constructs, Q² from blindfolding (require Q² > 0), and SRMR for fit (lower than ~0.10, preferably ~0.08).
4
When reviewers request it, add bias-corrected 95% confidence intervals (2.5% and 97.5%) from bootstrapping to the hypothesis table.
5
Mediation requires total, direct, and specific indirect effects; insignificant indirect effects mean the hypothesized mediation route is not supported.
6
Moderation interpretation should combine interaction significance (β/t/p) with slope analysis to describe how the relationship changes across low vs high moderator values.

Highlights

A reader-friendly SmartPLS write-up starts with cleaned path coefficients for direct hypotheses, then moves to R²/Q²/SRMR before tackling mediation and moderation.

Even with a significant total effect (SL → TP), mediation can fail if all specific indirect effects through TC, RC, and PC are insignificant.

Community hostility (CH) moderates only the SL → RC link; slope analysis shows SL reduces relationship conflict mainly when CH is low.

Topics

SmartPLS3 Reporting
Hypotheses Testing
Mediation Analysis
Moderation Analysis
R Square Q Square SRMR

Mentioned

PLS
SRMR
R²
Q²
CH
TC
RC
PC
TP
SL