← Statistics Blog

PLS-SEM Explained: When to Use It, How to Run It and How to Report Results in APA 7

StatFas · Updated 20 November 2026

Partial least squares structural equation modeling (PLS-SEM) estimates constructs as composites and maximizes the explained variance of the endogenous constructs. It is a good choice when your goal is prediction or explanation of a target construct, when the model contains formative constructs, or when the model is complex. Covariance-based SEM (CB-SEM) instead reproduces the covariance matrix and suits theory confirmation and global model fit.

PLS-SEM or CB-SEM?

A small sample or non-normal data is not enough on its own to justify PLS-SEM. Link the choice to your research objective, and ask whether the indicator residual variances could carry meaning and whether measurement error is unlikely to be large. Reflective measurement does not force CB-SEM: use consistent PLS (PLSc) when you want composite-based estimation that mimics common-factor results.

How to run PLS-SEM, step by step

  1. Specify the model. Mark each construct as reflective (Mode A) or formative (Mode B) and draw the structural paths.
  2. Check the data. Look at skewness, kurtosis and outliers, and confirm the sample is large enough, for example with the inverse square root method.
  3. Run the algorithm (path weighting scheme) and confirm that it converged.
  4. Assess reflective measurement: loadings ≥ .708, Cronbach’s α, ρA and composite reliability between .70 and .95, and AVE ≥ .50.
  5. Assess discriminant validity with HTMT and its bootstrap confidence interval (below .85, or at most .90).
  6. Assess formative measurement: redundancy analysis, outer VIF below 3, and the significance and relevance of the weights.
  7. Assess the structural model: collinearity (inner VIF), bootstrapped path coefficients with confidence intervals, R², f² and total effects.
  8. Assess prediction with PLSpredict and report SRMR as an approximate fit measure.

The free PLS-SEM tool runs these steps in your browser, draws the path diagram in SmartPLS 4 or Amos style, and writes the results in APA 7 format. The matching PLS-SEM tutorial walks through an example.

How to report PLS-SEM results in APA 7

Name the method and the settings, report measurement quality first, then the structural model. Replace the values in brackets with your own output.

PLS-SEM was estimated with the path weighting scheme (N = [n]). Reflective loadings ranged from [min] to [max], composite reliability from [min] to [max], and AVE from [min] to [max]. HTMT values were below [value]. [Predictor] significantly predicted [outcome], β = [value], SE = [value], 95% CI [[lower], [upper]], p < .001, f² = [value]. The model explained [value]% of the variance in [outcome] (R² = [value]). PLSpredict indicated [high/medium/low] predictive power.

Avoid the old PLS goodness-of-fit index, which does not reliably separate valid from invalid models, and avoid relying on the Fornell–Larcker criterion alone for discriminant validity.

Frequently asked questions

When should I use PLS-SEM instead of CB-SEM?

Choose PLS-SEM for prediction, key-driver explanation, formative measurement and complex models. Choose CB-SEM for theory confirmation and global model fit.

How do I report PLS-SEM results?

Name the method and settings, report reflective loadings, composite reliability, AVE and HTMT first, then bootstrapped path coefficients with confidence intervals, R², f² and PLSpredict results.

Which discriminant validity criterion should I report?

Report HTMT with its bootstrap confidence interval rather than relying on the Fornell–Larcker criterion alone.

Related guides