GSCA
Two SEM domains
What is GSCA?
How is GSCA different from PLS?
What is Integrated GSCA?
References
GSCA is a statistical method for componentbased structural equation modeling. Its major characteristics can be described in terms of model specification, estimation, and evaluation.

Model specification
To specify a structural equation model with components, GSCA involves three submodels — measurement, structural, and weighted relation. The measurement model expresses the relationships between components and their indicators, the structural model specifies the relationships between components, and the weighted relation model explicitly defines components as weighted sums of their indicators. GSCA then combines the three submodels into a single one, referred to as the GSCA model, which facilitates the definition of a global optimization criterion.

Model estimation
GSCA typically utilizes a least squares method, called the alternating least squares (ALS) algorithm, for estimating model parameters, without taking recourse to a distributional assumption such as indicators’ multivariate normality. The ALS algorithm minimizes a single optimization criterion, derived from the GSCA model, to estimate all parameters simultaneously. GSCA relies on resampling methods, such as bootstrapping, to obtain standard errors and confidence intervals, which can be used for testing the statistical significance of the parameter estimates.

Model evaluation
GSCA provides overall model fit measures, which evaluate how well a model fits the data. These measures evaluate a model’s explanatory power of components and indicators (e.g., FIT and AFIT) or quantify the discrepancies between the sample and modelimplied covariances (e.g., GFI and SRMR). Recent research has suggested rule of thumb cutoff criteria for GFI and SRMR [19]. GSCA also offers an outofsample prediction error measure (OPE) for comparing the overall predictive power of models [20].
In addition, GSCA provides local fit measures, which assess the explanatory or
predictive power of specific submodels (e.g., FIT_M, FIT_S, OPE_M, and OPE_S).