Test of the measurement model
Table 5 presents descriptive statistics, reliability coefficients and the correlation matrix of the research variables.the first step in the data analysis consisted of simultaneously estimating the measurement and structural models using PLS. Psychometric properties of both
the formative and reflective construct indicators (measures) were thus assessed within the context of our research model.
Given that the usual reliability and validity criteria do not apply to a formative construct, it must first be verified that there is no multicollinearity among the indicators forming this construct [70]. This was verified with the variance inflation factor (VIF), based on the guideline that this statistic should not be greater than 3.3 for any formative indicator [71].d As shown in Table 5, this condition held for all six indicators.
Once the validity of the formative constructs has been assessed, the unidimensionality and reliability of the reflective constructs must then be evaluated. As depicted in Figure 2, the fact that all indicator loadings (λ) on these constructs were greater than the 0.7 threshold confirmed their unidimensionality. As shown in Table 6, composite reliability coefficient values above the 0.7 threshold also provide strong evidence of the reliability of the three reflective constructs. Evidence of the convergent validity of the reflective constructs was also found, as their average variance extracted (AVE) values are all above the 0.5 threshold.
The last property to be verified is discriminant validity, which shows the extent to which each construct in the research model is unique and different from the others. Here, the shared variance between a reflective construct and other constructs must be less than the AVE by a construct from its indicators. The results in Table 6 show this to be the case for all three reflective constructs included in the model. For the two formative constructs, discriminant validity is demonstrated by a correlation with any other construct that is significantly different from unity (at p < 0.001) [72].