Model misspecifications and bootstrap parameter recovery in PLS-SEM and CBSEM-based exploratory modeling

Abstract

Theories are uncertain and evolving in exploratory research. This uncertainty can manifest itself in SEM studies either at the measurement or structural level, or both, and result in model misspecifications. Researchers often favor the use of PLS-SEM over CBSEM in exploratory research due to its tractability, flexibility, and its ability to avoid factor indeterminacy. While these strengths of PLS-SEM are undoubtedly appealing, empirical support regarding the robustness of model parameters under conditions of model misspecifications is lacking. This Monte Carlo study evaluates the efficiency and accuracy of bootstrap parameter recovery by PLS-SEM, CBSEM, and the Bollen-Stine methods under various conditions of measurement and structural misspecification effect sizes, sample sizes, and data distributions. Results point to the favorability of PLS-SEM in exploratory modeling when structural parameters are of interest, while CBSEM and Bollen-Stine methods are appealing when the focus is at the measurement level. A two-pronged strategy is advisable when theoretical uncertainty exists both at the measurement and structural levels.