Page 153 - 33-3

NTU Management Review Vol. 33 No. 3 Dec. 2023

step is to check the structural model. We provide a detailed description of the two-step
method as follows.

5.1 Analysis of the Measurement Model

This study uses confirmatory factor analysis to test the measurement model and
study its reliability and validity. In PLS, it is generally recommended to use Cronbach’s
α, composite reliability (CR), and average variance extracted (AVE) to verify its

reliability (Hair, Black, Babin, and Anderson, 2009). For the value of Cronbach’s α, it
is recommended that the test value should be above 0.7, indicating that the facet has
good reliability (Nunnally, 1978). As for the value of CR, the recommended detection
value should be above 0.7, indicating that its facet achieves internal consistency (Chin
and Newsted, 1999; Chin, Marcolin, and Newsted, 2003). Besides, the AVE of each

dimension should be greater than 0.5, indicating that each facet has sufficient convergence
effectiveness (Fornell and Larcker, 1981). Table 3 shows that the reliability index results
of all these dimensions meet the recommended standards.

In addition, in order to verify validity, we measure content validity and structural
validity as well. Since the items of this study are derived from previous research, the
content validity can be ensured. On the other hand, construct validity is often used to test
discriminant validity. The PLS test discriminates the validity by placing the square root of
the AVE of the individual dimension in the diagonal of the correlation coefficient matrix.

In order to pass the test of discriminant validity, the square root of AVE should be greater
than the correlation coefficient of the dimension and other dimension in the model (Chin,
1998; Chin et al., 2003). Table 4 shows the discriminant validity of all constructs, which

fulfill the standards. The numbers on the diagonal are larger than the numbers under the
diagonal, so it can be explained that all constructs have discriminant validity.
If the correlation coefficient is greater than 0.74, multiple-collinearity of the two
variables is suspected. However, Table 4 shows that one of the correlation coefficients is
higher than 0.74; thereby, the multiple-collinearity problem may exist, so we calculate

the variance inflation factor (VIF), which can be used to determine whether there is
a collinearity problem. We then use regression analysis to find the VIF value of each
variable. VIF values range from 1.0 to 2.0, well below the recommended maximum level

of 3.3-10.0 (Diamantopoulos and Siguaw, 2006; Hair et al., 2009). There are no VIF values

145