Moreover, we assess the discriminates validity of the measures in two ways. First, we calculate the shared variance between all possible pairs of constructs and compare it with the average variance extracted (AVE) of each construct. The results show that for each construct, the AVE is much higher than the highest shared variance, indicating discriminate validity among our constructs. Second, we assess discriminated validity using two-factor CFA models involving each possible pair of constructs, with the correlation between the two constructs first fixed as 1 and then freely estimated. All the χ2 values of the freely estimated model were significantly lower than the restricted model, in support of discriminant validity (Anderson & Gerbing, 1988). The composite reliabilities of all constructs range from0.71 to 0.91 (see Table 1).