Improving PLS-SEM with MARS: An Innovative Approach for Nonlinear Interactions Between Latent Variables

Partial least squares structural equation modeling (PLS-SEM) is a popular method for examining the interactions among latent variables. However, this approach has limitations in capturing the complexity of nonlinear relationships between latent variables, which are often ignored in PLS-SEM models. To get around these problems, this study combines the PLS-SEM method with multivariate adaptive regression splines (MARS), a flexible nonlinear method that can find more complex connections between hidden variables. This method modifies the PLS algorithm to identify and model nonlinear relationships more flexibly and accurately without requiring the assumption of a linear distribution. The simulation study results showed that MARS-PLS showed good accuracy with increasing amounts of data. The results of empirical studies using the modified unified theory of acceptance and use of technology (M-UTAUT) model for behavior intention to use M-banking show that PLS-SEM with MARS can predict the relationship between latent variables that are nonlinear with interaction models between latent variable scores. By providing new capabilities for capturing more intricate relationships between latent variables, this approach not only improves the validity and reliability of the model but also enriches the interpretation of results. For complicated modeling across multiple disciplines, this model can offer more accurate and adaptable solutions.