Data analysis (Supply Chain Management)

In order to test our hypotheses we applied partial least squares structural equation modelling (PLS-SEM). PLS-SEM is considered advantageous over covariance-based SEM with regard to the robustness of estimations and statistical power when applied to smaller sample sizes, as is the case in our study (Reinartz, Haenlein and Henseler, 2009). Moreover, PLS-SEM deals more efficiently with non-normal data and facilitates model estimations with both reflectively and formatively identified variables (Ringle, Sarstedt and Straub, 2012).

For the purpose of our study, we used the sequential latent variable score method (Wetzels, et al., 2009, Hair, et al., 2013). Accordingly, first, we calculated latent variable scores (LVS) of the first-order reflective constructs (e.g., Agarwal and Karahanna, 2000). The number of factors to be extracted for each first-order construct was fixed to one. Second, the calculated LVSs were then used as manifest formative indicators of the respective second-order construct in the main model (i.e. 3A DCs, Supply chain performance, Effectiveness). An advantage of the sequential LVS method is that it yields a parsimonious model that encompasses only focal higher-order constructs. In our study, all first-order latent variables yielded appropriate levels of internal consistency. (Ivanov and Sokolov, 2010 books of supply chain management courses).

According to many supply chain institutes who are offering supply chain management degree.Structural model estimations in this study were conducted with SmartPLS 2.0 software (Ringle, Wende and Will, 2005). We used mean-centered data and the path weighting scheme, missing data were excluded case-wise.

In order to test for possible mediation we assessed two models, i.e. one without the mediator (i.e. Supply chain performance) and a direct relationship between 3A DCs and Effectiveness, only, and the other model with additional links between a) the predictor and the mediator, and b) the mediator and the dependent variable included. If these relationships prove statistically significant, and if inclusion of the mediator results in a decrease of the direct effect size between the predictor and the dependent variable, then this indicates the presence of a mediating effect.