SEM method guides
Process flows, commonly used cut-offs, and key references for PLS-SEM, CB-SEM, regression, and multi-group analysis — aligned with how WAnalytic reports results.
Partial least squares SEM
Composite-based SEM for prediction and theory exploration when constructs are measured with multiple indicators. Suited to complex models, smaller samples (with care), and non-normal data relative to covariance SEM.
Process flow
- Specify measurement model — reflective or formative constructs, indicators, and higher-order structure if needed.
- Specify structural model — paths among constructs (mediation, moderation, controls).
- Estimate PLS algorithm — outer loadings/weights and path coefficients.
- Evaluate measurement model — reliability, convergent and discriminant validity.
- Evaluate structural model — collinearity, R², f², Q² / PLSpredict, significance via bootstrapping.
- Report and interpret — effect sizes, mediation, and managerial/theoretical implications.
Acceptable values (measurement)
| Criterion | Typical rule of thumb | Notes / references |
|---|---|---|
| Indicator loading (reflective) | ≥ 0.708 preferred; ≥ 0.40–0.70 may be retained if content-valid and AVE/CR hold | Hair et al. (2019, 2022) |
| Cronbach’s α | ≥ 0.70 (exploratory sometimes ≥ 0.60) | Nunnally & Bernstein (1994); Hair et al. |
| Composite reliability (ρc / CR) | ≈ 0.70–0.95 (very high values may signal redundancy) | Hair et al. (2019) |
| AVE | ≥ 0.50 | Fornell & Larcker (1981); Hair et al. |
| Fornell–Larcker | √AVE of construct > its correlations with other constructs | Fornell & Larcker (1981) |
| HTMT | < 0.85 (stricter) or < 0.90 (conceptually similar constructs) | Henseler et al. (2015); Hair et al. (2019) |
Acceptable values (structural)
| Criterion | Typical rule of thumb | Notes / references |
|---|---|---|
| Inner VIF | < 5 (often < 3.3 preferred) | Hair et al. (2019) |
| Path significance | Bootstrap CI excludes 0 (e.g. 5,000 samples); report p / t | Hair et al. (2022) |
| R² | Context-dependent; ~0.25 / 0.50 / 0.75 as weak / moderate / substantial (rough guide) | Hair et al. (2011, 2019) |
| f² effect size | ≈ 0.02 / 0.15 / 0.35 small / medium / large | Cohen (1988); Hair et al. |
| Q² / PLSpredict | Q² > 0 indicates predictive relevance; compare MAE/RMSE to naive benchmarks | Stone–Geisser; Shmueli et al. (2016, 2019) |
Key references
- 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.
- Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based SEM. Journal of the Academy of Marketing Science, 43, 115–135.
- Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50.
Covariance-based SEM
Factor-based SEM that reproduces the covariance matrix. Prefer when theory is confirmatory, measurement is reflective, multivariate normality is plausible (or estimators allow otherwise), and global model fit is required.
Process flow
- Model specification — CFA measurement model, then structural paths (and constraints).
- Identification check — degrees of freedom, scale setting (e.g. marker loading = 1).
- Estimation — ML (default), or robust / DWLS / other estimators as appropriate.
- Measurement evaluation — loadings, AVE/CR (when applicable), CFA fit.
- Structural evaluation — path estimates, mediation, nested model comparisons.
- Overall fit & modification — CFI, TLI, RMSEA, SRMR; MI only with theory.
Global fit (common cut-offs)
| Index | Often cited “good fit” | Notes / references |
|---|---|---|
| χ² / df | Often < 3 (sometimes < 5); χ² sensitive to N | Kline (2016); report with other indices |
| CFI / TLI | ≥ 0.95 close; ≥ 0.90 sometimes acceptable | Hu & Bentler (1999); Hair et al. (2019) |
| RMSEA | ≤ 0.06 close; ≤ 0.08 reasonable; report CI | Hu & Bentler (1999); Browne & Cudeck (1993) |
| SRMR | ≤ 0.08 | Hu & Bentler (1999) |
Measurement & assumptions
| Criterion | Typical guidance | References |
|---|---|---|
| Standardized loadings | Often ≥ 0.50 (ideally ≥ 0.70) with significance | Hair et al. (2019) |
| CR / AVE | CR ≥ 0.70; AVE ≥ 0.50 | Fornell & Larcker (1981); Hair et al. |
| Multivariate normality | Preferred for ML; use robust estimators or bootstrapping if violated | Kline (2016); Finney & DiStefano |
| Sample size | Rules of thumb (e.g. N/q, power); prefer a priori power analysis | Kline (2016); Wolf et al. (2013) |
Key references
- Kline, R. B. (2016). Principles and Practice of Structural Equation Modeling (4th ed.). Guilford.
- Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis. Structural Equation Modeling, 6(1), 1–55.
- Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). Multivariate Data Analysis (8th ed.). Cengage.
- Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In Bollen & Long (Eds.), Testing Structural Equation Models.
Regression analysis
Observed-variable models relating predictors to an outcome. Use when constructs are single-item or scores are already computed, and the research question does not require a full latent measurement model.
Process flow
- Define outcome and predictors — including controls and interactions if needed.
- Screen data — missingness, outliers, coding, multicollinearity.
- Estimate model — OLS (or GLM / logistic as appropriate).
- Check assumptions — linearity, residual normality/homoscedasticity, independence, VIF.
- Interpret — coefficients, R² / adjusted R², effect sizes, diagnostics.
Acceptable values & diagnostics
| Criterion | Typical guidance | References |
|---|---|---|
| VIF | < 5 (often < 10 as a loose upper bound; < 3 preferred) | Hair et al. (2019); O’Brien (2007) |
| Tolerance | > 0.20 (preferably > 0.40) | Inverse of VIF rules |
| R² / Adj. R² | Field-specific; report both and interpret substantively | Cohen (1988); Field (2018) |
| Standardized β | Compare relative predictor importance within the model | Cohen et al. (2003) |
| Residuals | Inspect plots; Durbin–Watson ≈ 1.5–2.5 for independence (time/order data) | Field (2018) |
Key references
- Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences (3rd ed.).
- Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). Sage.
- Hair, J. F., et al. (2019). Multivariate Data Analysis (8th ed.).
Multi-group analysis
Tests whether structural (and sometimes measurement) relationships differ across groups (e.g. gender, country, treatment). In PLS, pair with MICOM before interpreting path differences.
Process flow (PLS-oriented)
- Define groups — clear, mutually exclusive categories with adequate n per group.
- Estimate models by group — same specification in each group.
- MICOM (measurement invariance) — configural → compositional → equality of means (as needed).
- Compare paths — permutation MGA, parametric, or Welch–Satterthwaite approaches.
- Report — which paths differ, effect sizes, and invariance stage reached.
Process flow (CB-SEM-oriented)
- Multi-group CFA — configural, metric, scalar invariance (as required by the claim).
- Structural multi-group model — constrain/free paths across groups.
- Compare nested models — Δχ² / ΔCFI (e.g. ΔCFI ≤ 0.01 often cited for invariance).
- Interpret group differences only after adequate measurement invariance.
Acceptable values & guidance
| Topic | Guidance | References |
|---|---|---|
| MICOM (PLS) | Establish configural & compositional invariance before path MGA; equal means for latent mean comparisons | Henseler et al. (2016); Hair et al. (2022) |
| Permutation MGA | Significant path difference when permutation p < 0.05 (two-tailed as specified) | Chin & Dibbern; Hair et al. |
| CB invariance (ΔCFI) | ΔCFI ≤ 0.01 often used as support for invariance (with RMSEA / other indices) | Cheung & Rensvold (2002); Chen (2007) |
| Group sample size | Prefer balanced groups; underpowered groups inflate Type II error | Kline (2016); Hair et al. |
Key references
- Henseler, J., Ringle, C. M., & Sarstedt, M. (2016). Testing measurement invariance of composites using partial least squares. International Marketing Review, 33(3), 405–431.
- Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9(2), 233–255.
- Hair, J. F., et al. (2022). A Primer on PLS-SEM (3rd ed.) — multi-group chapters.
PLS-SEM vs CB-SEM
Both estimate relationships among constructs, but they optimize different criteria and answer slightly different research goals. Choose based on theory, data, and reporting needs — not only software convenience.
| Dimension | PLS-SEM | CB-SEM |
|---|---|---|
| Core idea | Composite / variance-based; maximizes explained variance | Factor / covariance-based; reproduces covariance structure |
| Typical use | Prediction, complex models, exploratory / extension of theory | Confirmatory theory testing, global model fit |
| Measurement | Reflective and formative composites | Primarily reflective factors (formative needs special care) |
| Distribution | Fewer normality demands; bootstrap inference | ML prefers MVN; robust options available |
| Sample size | Often workable with smaller N (still power-aware) | Usually needs larger N for stable fit/estimates |
| Model fit | Local criteria + predictive metrics (R², f², Q², PLSpredict); limited global fit | χ², CFI, TLI, RMSEA, SRMR central to evaluation |
| Discriminant validity | HTMT + Fornell–Larcker | AVE/shared variance, CFA nested tests, etc. |
| When to prefer | Prediction focus, formative measures, very complex path models | Strong theory, reflective CFA tradition, fit indices required by outlet |
Practical recommendation
- If the journal/theory tradition expects CFA fit indices and reflective latent variables → start with CB-SEM.
- If the goal is prediction, score composites, or include formative constructs in a complex path model → start with PLS-SEM.
- You can estimate both for triangulation when assumptions allow (WAnalytic comparison tools support this workflow).
Key references
- Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a silver bullet. Journal of Marketing Theory and Practice, 19(2), 139–152.
- Hair, J. F., et al. (2019). When to use and how to report the results of PLS-SEM. European Business Review.
- Jöreskog, K. G., & Sörbom, D. — classical CB-SEM / LISREL tradition; see also Kline (2016).
- Rigdon, E. E., Sarstedt, M., & Ringle, C. M. (2017). On comparing results from CB-SEM and PLS-SEM. Journal of Targeting, Measurement and Analysis for Marketing / related debates in Internet Research & SEM literature.
Cut-offs are widely used heuristics, not universal laws. Always justify choices for your discipline, sample, and model complexity, and follow the latest journal guidelines.