The models were estimated twice for every possible pair of constructs in the measurement model. In the first model the phi correlation between the constructs was set to vary (unconstrained model), and in the second the phi was constrained to 1.00 (constrained model; Anderson & Gerbing, 1988). The chi-square difference and the degrees of freedom were computed for both the constrained and unconstrained models. The results showed that all models in which the phi was set to unity displayed worse fit (all χ2 difference > 3.841, df = 1, and p = .05).