Report PLS-SEM results in two stages: first the measurement model (outer loadings of .708 or higher, composite reliability between .70 and .95, AVE of .50 or higher and HTMT below .85 or .90), then the structural model (VIF below 3 to 5, path coefficients with bootstrap t values, p values and confidence intervals, R², f² and Q²). This sequence and these values follow the reporting guidelines of Hair, Risher, Sarstedt and Ringle (2019) and the PLS-SEM primer (Hair et al., 2022). Examiners look for this order, so use it even if your software prints results differently.
Before the results: describe your set-up
State the software and version, the algorithm settings (path weighting scheme, maximum iterations), and the bootstrapping settings. Hair et al. (2022) recommend 10,000 bootstrap subsamples for final results. Say whether constructs are reflective or formative and why, and report sample size against your minimum sample calculation.
Stage 1: measurement model (reflective)
| Criterion | Threshold | Source |
|---|---|---|
| Outer loadings | ≥ .708 | Hair et al. (2019) |
| Cronbach's alpha, rho_A, composite reliability | .70 – .95 | Hair et al. (2019); Dijkstra & Henseler (2015) |
| Average variance extracted (AVE) | ≥ .50 | Fornell & Larcker (1981) |
| HTMT | < .85 (or < .90 for similar constructs) | Henseler et al. (2015) |
Present one table with loadings, alpha, rho_A, composite reliability and AVE for every construct, followed by an HTMT matrix. If you deleted items, say which and why. Formative constructs are reported differently: indicator weights and their significance, indicator VIF, and convergent validity with a global single item (Hair et al., 2022).
Stage 2: structural model
- Collinearity: inner VIF values; 5 or above is critical, ideally 3 or below (Hair et al., 2019).
- Path coefficients: β, t, p and the 95% bias-corrected or percentile confidence interval. A path is supported when the interval excludes zero.
- R²: .75, .50 and .25 are described as substantial, moderate and weak (Hair et al., 2019), although what counts as high depends on the discipline.
- f²: .02, .15 and .35 indicate small, medium and large effects (Cohen, 1988).
- Q² (predictive relevance): values above 0 indicate predictive relevance; PLSpredict is now preferred for out-of-sample prediction (Hair et al., 2019).
Worked example
A study of 312 users tests whether Perceived Usefulness (PU) and Trust (TR) predict Intention to Use (IU). Bootstrapping with 10,000 subsamples gives:
| Hypothesis | Path | β | t | p | 95% CI | f² | Decision |
|---|---|---|---|---|---|---|---|
| H1 | PU → IU | .46 | 8.21 | < .001 | [.35, .57] | .29 | Supported |
| H2 | TR → IU | .21 | 3.42 | .001 | [.09, .33] | .06 | Supported |
R² for Intention to Use is .38. Following Cohen (1988), the effect of PU (f² = .29) is medium, close to large, and the effect of TR (f² = .06) is small. Inner VIFs are 1.41 for both predictors. A results paragraph could read:
The structural model showed no collinearity concerns (VIF = 1.41, below 3; Hair et al., 2019). Perceived Usefulness had a positive effect on Intention to Use, β = .46, t = 8.21, p < .001, 95% CI [.35, .57], supporting H1, and Trust had a smaller positive effect, β = .21, t = 3.42, p = .001, 95% CI [.09, .33], supporting H2. Together the predictors explained 38% of the variance in Intention to Use (R² = .38).
Note that β, R², f² and p have no leading zero in APA 7 because they are bounded by 1 in this context, while t can exceed 1 and keeps its digits (American Psychological Association, 2020).
Mediation and moderation
For mediation, report the specific indirect effect with its bootstrap confidence interval, not a Sobel test, and classify the mediation type (Hair et al., 2022). For moderation, report the interaction term's path, its interval and a simple slope plot.
Run it free
The free PLS-SEM tool produces both stages with sourced cut-offs, APA tables and a draft results paragraph; the results doctor checks a finished results section for missing statistics.
FAQ
Do I need to report model fit (SRMR) in PLS-SEM?
Hair et al. (2019) caution that fit indices are of limited use in PLS-SEM. Report SRMR if your examiner expects it, but do not rely on it.
Should I report the Fornell-Larcker criterion?
Report HTMT as the main criterion; add Fornell-Larcker only as a supplement (Henseler et al., 2015).
How many bootstrap samples?
10,000 for final results (Hair et al., 2022).
Can I report one-tailed p values?
Only if your hypotheses are directional and you say so before testing. Two-tailed is the safer default.
References
- American Psychological Association. (2020). Publication manual of the American Psychological Association (7th ed.). https://doi.org/10.1037/0000165-000
- Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.
- Dijkstra, T. K., & Henseler, J. (2015). Consistent partial least squares path modeling. MIS Quarterly, 39(2), 297–316.
- Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104
- Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A primer on partial least squares structural equation modeling (PLS-SEM) (3rd ed.). Sage.
- Hair, J. F., Risher, J. J., Sarstedt, M., & Ringle, C. M. (2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2–24. https://doi.org/10.1108/EBR-11-2018-0203
- Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135. https://doi.org/10.1007/s11747-014-0403-8