Confirmatory Factor Analysis
Consequently, the new data were subject to a CFA in order to validate the hypothesized structure of the eWOM trust construct. The adopted model was specified so that each of the 24 items exclusively load on one of the five sub-dimensions which were identified in the earlier factor analysis. While the chi-square test was highly significant (x2 = 773.84, df = 247, p < .001), the absolute, incremental, and parsimonious fit indices suggest that the hypothesized construct is acceptable (absolute fit indices: GFI = .89, AGFI = .86, RMSEA = .065, RMR = .10, SRMR = .06; incremental fit indices: CFI = .95, NNFI = .94, NFI = .92; parsimonious fit indices: normed chi-square: 3.1, AIC = 898.40, PIF = .73).
The reliability statistics are provided in Table 28. All observable indicators were found to be significant (t-values ranging from 13.36 to 26.71 and, therefore, passing the critical value for .01 significance) and considerably related to their specified constructs. This is indicative of the existence of the posited relationship between the sub-dimensions and their indicators. Further evidence for the indicator’s reliability was gained by reviewing the items’ squared multiple correlations, which turned out to be mainly over or near to the limit of .50. Only one item (Be6) was found to possess a slightly lower R2 of .40, which corresponded to earlier findings that had indicated a problematic low explanation of the item’s variance by the benevolence construct. The squared multiple correlation of the rest of the individual items ranged from .52 to .77. All sub-dimensions exhibited a high level of internal consistency: The Cronbach’s alphas varied from .81 to .94, the average inter-item correlation from .52 to .74, and the construct reliabilities ranged from .82 to .94. Hence, minimum internal consistency levels (i.e., .70 for Cronbach’s alpha and construct reliability; .50 for inter-item correlation) were achieved. Also the AVE met the recommended .50 threshold (Diamantopoulos & Siguaw, 2000; Hair et al., 2010) for all trust components. Here the Benevolence dimension had with .52 the lowest AVE value. Nevertheless, in the validation samplem the majority of its variance is still explained by the higher-order construct. The corrected item-to-total correlations of the individual items were all above .50.
Evidence for reliability of the first-order constructs as indicators of the second-order construct was given by positive and significant loadings on the higher-order construct. Completely standardized loadings (y) ranged from .64 to .93. Similarly to the reliability samples, the squared multiple correlation of the benevolence dimension was slightly below the proposed cut-off value of .50 (R2 = .43), while the remaining constructs demonstrated adequate levels with .53 or better. An AVE of .91 and a construct reliability of .91 both supported the internal consistency of the second-order construct.