| Summary: | Partial Least Squares Path Modeling (PLS-PM) is a Variance-Based Structural Equation Modeling (VB-SEM) that is widely applied in management and social sciences. Therefore it was promoted as a method of choice for various analysis situation, despite the serious implications of the method has being declared in the research method journal since its inception. The current lack of methodological evidences for PLS-PM is adjusted to provide a correction for estimates when PLS is applied to reflective constructs. So, this novel approach called Consistent Partial Least Squares, denoted by Consistent PLS or PLSc, can explains different types of modeling (confirmatory and exploratory). However, the full potential of Consistent PLS is still remain vague due to its lack of methodological justification. This study is aimed to compare the Consistent PLS with established method of Covariance Based Structural Equation Modeling (CB¬SEM) using proactive Monte Carlo approach when all constructs are modeled as common factors. This study could clears up the possible ambiguity regarding the usefulness and appropriateness of Consistent PLS in the confirmatory modeling. A proactive Monte Carlo simulation (N = 50, 100,200 and 500) with different population models such as Theory of Reason Action (TRA), Theory of Customer Loyalty, and Unified Theory of Acceptance and Use of Technology (UTAUT) are analyzed for the simulation purpose. The data are generated from these population models with hypothesized parameter values using mass and psych package of R statistical programming. The mvrnorm package is employed to ensure the distribution of the data to be normal condition. Those population models are now be assessed by two statistical programs such as Analysis Moment of Structures (AMOS version 21.0) for CB-SEM and Advanced of Composite Model (ADANCO version 2.0) for Consistent PLS. The outcome of a Monte Carlo simulation reveals that Consistent PLS does not adjust for other limitations of PLS-PM in confirmatory modeling, namely bias in estimates of regression weight due to capitalization chance; overestimation of convergent validity and composite reliability due to the proportional of factor loadings; overestimation of construct correlations; discriminant validity is less efficient under Fornell & Larcker criterion; low effects of statistical power and squared correlation; and improper solution for small samples. Moreover, the outcomes shows that Consistent PLS has no advantage when using normal distributed data. The Consistent PLS is less efficient than of CB¬SEM when confirmatory modeling is adapted. Therefore, the CB-SEM is still the best method for confirmatory modeling and normal distributed data. Finally, several suggestions are given, primarily to the Consistent PLS for improvements.
|
|---|