3.8.3. Construct reliability
We used the concepts composite reliability (CR) and average variance extracted (AVE) to determine quality at the construct
level (Fornell and Larcker, 1981). Here, CR should be higher than 0.7 and AVE higher than 0.5. As the Appendix A (Table
6), both criteria are fulfilled within our research model.
3.8.4. Discriminant validity
Discriminant validity describes the extent to which measurement items differ from one another (Campell and Fiske,
1959). Therefore, the square root of AVE is contained on the diagonal of latent variable correlation (see Table 6 in the
Appendix A). As these square root values are greater than the corresponding construct correlations (Fornell and Larcker,
1981; Hulland, 1999), we can state that this requirement has been fulfilled and the measurement model is valid.
3.9. Structural model
We used the coefficient of determination (R2) and significance levels of each path coefficient to evaluate the structural
model (Chin, 1998b). Fig. 2 indicates that the perceptual beliefs and attitude of TAM explain 23% of the variance of an employee’s
job satisfaction. In addition to that the two attitudes – toward IT as well as job – explain about 48% of the variance of
turnover intention. Besides, the R2 of perceived usefulness and attitude are both 58% (Fig. 2). Concerning the path coefficients,
we could state that solely one hypothesized path is not significant. This non-significant path within the research model
is the relation between attitude and turnover intention.