A0298
Title: A simple and direct procedure for data generation in PLS-SEM framework
Authors: Sophie Dominique - XLSTAT (France) [presenting]
Veronique Cariou - National college of veterinary medicine food science and engineering (France)
Mohamed Hanafi - ONIRIS (France)
Jean-Marc Ferrandi - ONIRIS (France)
Fabien Llobell - XLSTAT (France)
Abstract: Simulation studies usually require an initial step of generating datasets by varying the values of several pre-defined parameters. As with all statistical techniques, it is necessary to carry out simulation studies within the composite-based SEM framework to check the validity conditions, compare the performance and reliability of related methods, or identify the most efficient one under certain assumptions. Among the different composite-based SEM methods already proposed, we focus here on the PLS-SEM method, which remains undeniably the most widely used. While data generation procedures have been extensively studied in the context of covariance-based approaches, little attention has been paid to adapting these procedures to the context of composite-based approaches and, more specifically, to PLS-SEM. To fill the gap, we propose a direct data generation procedure that follows the logic of the PLS-SEM algorithm and involves the linear regressions derived from the structural model to estimate path coefficients. This procedure turns out to be simpler than the one recently proposed by Schlittgen. The main reason is that Schlittgen's procedure requires the calculation of the covariance matrix implied by the structural model, whereas such a calculation is unnecessary for our new data generation scheme. Finally, a comparison is made between the two different strategies evaluating the accuracy of the parameter values obtained with respect to those defined a priori.
