A comparative study of the predictive power of component-based approaches to structural equation modeling

This simulation study compares the predictive performance of GSCA and partial least squares path modeling for correctly specified and misspecified component-based models. It distinguishes prediction of indicators from component scores from prediction using the full set of parameter estimates. Across the studied conditions, reflective GSCA offered the strongest overall balance: it slightly outperformed Mode A partial least squares for indicator prediction and performed similarly for the second prediction task. Formative GSCA and Mode B partial least squares generally performed less well. The authors discuss implications for selecting a method when prediction is the primary goal.