Ensuring Robustness in PLS-SEM and Regression: Evaluating Multivariate Assumptions in the Study of Innovative Work Behaviour among Malaysian Academics

Abstract

This study presents a methodological examination of multivariate assumptions, including normality, linearity, homoscedasticity, multicollinearity, and residual independence, within a behavioral research model focused on innovative work behavior (IWB) among academics in Malaysian higher education institutions (HEIs). Thus, the purpose of this study is to comprehensively assess whether important multivariate assumptions underlying multiple regression and PLS-SEM are satisfied when modelling IWB among academics. Using survey data collected from 389 permanent academic staff, IWB was modelled as the primary outcome variable, predicted by psychological empowerment, flexible work arrangements, perceived organizational support, knowledge sharing, transformational leadership and individual innovation capability. The model estimate was performed using composite construct scores obtained from verified Likert-scale measures, which were examined using SPSS Version 29 for assumption testing and SmartPLS 4.0 for structural modeling. Model estimation was performed using composite construct scores obtained from verified Likert-scale measures, examined using SPSS Version 29 for assumption testing and SmartPLS 4.0 for structural modelling. The findings indicate that the data met all essential multivariate assumptions, except for multivariate normality, which is a key assumption. However, the other assumptions, including linearity, homoscedasticity, multicollinearity, and residual independence, were met. By applying this approach, it improves the validity of the ensuing inferential analysis. Thus, the current study methodologically contributes by providing a structured, contextual framework for assumption testing in higher education research, serving as a reference for other studies in the behavioral and social sciences. It is, therefore, recommended that all researchers across behavioral and social sciences, management, and all relevant disciplines, adopt and adhere to this framework for testing assumptions for their data to be applicable for multivariate data analysis.