Directly to the table:
| Dimension | PLS-SEM (Variance/Prediction-Oriented) | CB-SEM (Covariance/Validation-Oriented) |
|---|---|---|
| Core Objective | Maximize R², explain and predict dependent variables | Test overall model fit, validate theory |
| Sample Size | Robust with small samples (30–100 is sufficient) | Requires larger samples, generally ≥ 200 |
| Data Distribution | Non-parametric, does not assume multivariate normality | Parametric method, requires approximate multivariate normality |
| Measurement Philosophy | Composite model: indicators can be interchangeable | Common-factor model: indicators should be highly correlated |
| Measurement Types | Reflective, formative, and single-item are all acceptable | Mainly reflective; formative requires special settings |
| Model Complexity | Can handle 50+ constructs and a large number of indicators | Fit may fail with many indicators, needs simplification |
| Fit Indices | Traditional χ², CFI are not applicable; relies on SRMR, NFI, HTMT and other “approximate” indices, still evolving | Provides mature fit indices: χ², CFI, TLI, RMSEA, SRMR, etc., can be tested globally |
| Structural Path Coefficients | Usually slightly lower than CB; high statistical power, easy to achieve significance | Absolute values of coefficients are larger, but power decreases with small samples |
| Predictive Power Assessment | Built-in PLSpredict, CVPAT, can report out-of-sample prediction errors (RMSE/MAE) | Traditional CB lacks built-in predictive indices, requires additional k-fold, etc. |
| Theoretical Stage | Exploratory/Development Theory (early-stage) preferred | Validation/Comparative Theory (mature-stage) preferred |
| Publishability of Results | Recently accepted by mainstream journals, but must explain “why use PLS” | Still regarded as the “gold standard”, especially for factor models |
| Common Misconceptions | “Small sample + non-normality” is not the only reason; must emphasize prediction or theoretical development purpose | Fit indices may be distorted with insufficient samples or severe non-normality |
In summary:
If the research is in the theoretical development stage, has limited samples, or includes formative constructs, PLS-SEM should be prioritized; if the research aims to rigorously test the fit of mature theories, then CB-SEM is more appropriate.