Statistical analyses

The required statistical analyses were done using structural equation modeling (SEM) software such as IBM’s SPSS Amos ver. 22 and IBM’s SPSS Amos ver. 21 and replicated with the PROCESS macro module developed by Hayes,36 which includes the Johnson–Neyman technique37 for probing regions of significance. One reason why we performed SEM is that it allows testing all influence pathways simultaneously as opposed to multiple regression, which has a sequential approach.

Another reason for using both IBM’s Amos and SmartPLS was the novelty of the research. IBM Amos is a covariance-based modeling software, whereas SmartPLS relies on partial least squares. Both programs employ methods that have complementary strength, which cover sensitive areas of our research. According to Hair et al,38 partial least squares structural equation modeling (PLS-SEM) is best suited for predicting key target constructs and exploration or extension of existing theory, whereas for theory testing, theory confirmation, or comparison of alternative theories, covariance-based SEM is best suited.38

Moreover, because the adequate sample size is a controversial issue in SEM, we doubled the use of IBM SPSS Amos with that of SmartPLS, because the latter can well handle small samples and non-normal distributed data.39,40 However, because we needed to probe a special case of moderator–mediator, Amos was required as well in order to estimate the significance of the pathways and not only the effects of the interaction between terms. No cases of aberrant or missing data were encountered.