PLS-SEM

Partial Least Squares Structural Equation Modelling estimates a path model of latent constructs by maximising the explained variance in the dependent constructs. It is variance-based and prediction-oriented, in contrast to covariance-based SEM in AMOS or LISREL, which tests how well a theorised covariance structure fits the observed data.

PLS-SEM is favoured for smaller samples, non-normal data, formatively measured constructs and complex models with many constructs and indicators — which is why it dominates management, marketing and information-systems research. SmartPLS is the most widely used software for it.

Evaluation runs in two stages. The measurement model is assessed through indicator loadings, composite reliability, AVE and discriminant validity via HTMT. The structural model is then assessed through R squared, f squared, Q squared predictive relevance and path coefficients whose significance comes from bootstrapping, since PLS-SEM produces no parametric significance tests of its own.