Regarding the discriminant validity, this study first checked whether the correlations among the latent constructs were significantly less than unity (see Table 8). A series of nested confirmatory models were then compared, in which the correlation coefficient between similar attributes/constructs should exceed that between different attributes/ constructs. The intercorrelation among GSCM CSFs and the BSC performances ranges from 0.40 to 0.82. Constraining this correlation to unity invariably significantly changed the goodness-of-fit statistic