The Advantages Of Covariance-Based Approaches

PLS has some advantages over covariance-based approaches. First, covariance-based approaches yield very unreliable results for theory building studies, called factor indeterminacy. Because, these approaches produce more than one solutions which are mathematically proper but without determining which of the several solutions relates well to the underlying hypothesis. Additionally, covariance-based approaches can support numbers of statistically equivalent models by the same data and thus, it leads a difficulty to justify causality in the models. Therefore, covariance-based approaches are appropriate for empirical validation in well-established theories. While, PLS composes constructs from factor scores and using these scores in the following

Nevertheless, covariance-based approaches produce severe modeling errors, which lead unreliable results when a model is in mixed constructs. While, PLS is capable of estimating both reflective and formative constructs under athe same model (Lowry & Gaskin, 2014). PLS is sensitive to moderator effects than most of the covariance-based approaches are. Moreover, PLS is better handling measurement errors, thus this characteristic of PLS helps to studies that have smaller sample sizes. Furthermore, in case of a complex model, covariance-based approaches require an enormous sample size for precise estimations. In addition, these approaches are not capable of achieving convergence as the number of factors and indicators increase (Chumney, 2012; Lowry & Gaskin, 2014). Although, PLS is known to be a robust method when above-mentioned conditions are satisfied, its main weakness is originally it does not provide the overall goodness of fit statistic for theory testing and confirmation. Therefore, a global measure of model fit is not possible to be given by PLS estimation (J. F. Hair Jr et al., 2010; Hair et al., 2011; Tenenhaus, Vinzi, Chatelin, & Lauro,

FirstFirst, patient satisfaction theory and formulation is underdeveloped and varies in the literature. Inconsistency of the theory preventscludes from a common understanding of patient satisfaction concept and its measurements worldwide. Hence, one of our objectives is to test a proposed hypothesees that derived from satisfaction literatures of different disciplines in Mongolia. For this reason, the proposed patient satisfaction models with either formative or reflective constructs were evaluated separately to assess which model satisfies the PLS requirements best, since no solid evidence on causality between indicators and exogenous latent variables. However, in both models, endogenous latent construct is reflective and in one model, measurement constructs are reflective and in another, formative. In this sense, our study is exploratory and mixed-model design. Therefore, PLS estimation is known to be the most convenient when identifying key constructs and variables, and causal directions of observed and unobserved variables. Second, all of the service quality variables are in 7-point Likert scales and demographic variables are in nominal scales. Also, dependent variables are in 7-point Likert scales. Owing to the nature of the ordered categorical scale, the obtained data is positively skewed as in