As an alternative test, we compared the squared correlation between two latent
constructs to their average variance extracted (AVE) estimates (Fornell and Larcker,
1981). Based on the correlation coefficients given in Table I and the AVE values given
in Appendix 2, we can conclude that none of the squared correlations is higher than the
AVE for each individual construct. In fact, the highest squared correlation of 0.35
between communication and long-term relationship (with a correlation of 0.59) was
much lower than the AVE for the two constructs (0.56 and 0.57). These results
collectively provide strong evidence of discriminant validity among the theoretical
constructs.
Reliability was assessed using Cronbach’s alpha value (Cronbach, 1951; Nunnally
and Bernstein, 1994). Alternatively, following Bagozzi and Yi (1988), we computed
composite reliability (CR) scores to assess construct reliability. As reported in
Appendix 1, all factors have Cronbach’s alpha values and CRs greater than 0.70. In
addition, the AVE values for all constructs exceed 0.50. The Cronbach’s alpha values
for the outcome variables (customer service and financial performance) reported
in Appendix 2 show that these constructs exhibit adequate reliability as well.
Taken together, the results from the instrument development process show that the
theoretical constructs exhibit good psychometric properties.
3.5 Hypothesis testing and results
The hypothesized full structural model (Figure 1) was tested using LISREL, with
variance-covariance matrices for the latent variables and residuals used as input. The
summated-scores for the four first-order latent variables were used as indicators for the
second-order supply-chain relational capabilities construct. The summary statistics
and the correlation matrix for the constructs used in the model are presented in Table I.
The model parameters were estimated using the method of maximum likelihood.
The value for the model fit indices as given in Figure 2 shows that the model fits the
data very well. The hypothesized relationships were tested using their associated
t-statistics. Figure 2 presents the results of the hypothesized relationships among the
study variables. All the hypothesized relationships were found to be significant at the
0.01 level.