Appendix - Statistical Details

---

Confirmatory Factor Analysis

The confirmatory factor analysis estimated paths between each measure of a construct (i.e., theoretical latent variable) to each of its questionnaire items. Each measure was exogenous in the model and scaled by setting its standard deviation equal to one. The measures of the latent constructs were also allowed to pair-wise correlate. The individual items in each measure were reflective and affected by a random disturbance term. Each disturbance term was free to vary with a path between it and the indicant set equal to one. The estimation method used was maximum likelihood in Calis (i.e., Covariance Analysis of Linear Structural Equations) in PC SAS version 6.12.

The overall fit of the model to the data was good. The goodness of fit index was 0.92 and adjusted for degrees of freedom it was 0.90. The root mean square residual was 0.04. The chi-square statistic was statistically significant at a 1% level and had a value of 363.66 with 194 degrees of freedom. The normed chi-square statistic was 1.87. Bentler's comparative fit index was 0.97. The incremental fit indexes (i.e., Bentler and Bonett's normed and non-normed indexes and Bollen's normed and non-normed indexes) ranged from 0.92 to 0.97. These results and the relatively large sample size, even with the undesirable significant chi-square statistic, imply a good fit between the model and the data (Hair, Anderson, Tatham & Black, 1987).

Estimation of the Model

As in the confirmatory factor analysis, the model was expressed so that all the questionnaire items measured only their own construct. All indicants were reflective of their measures. Each indicant was also impacted by a disturbance term that was free to vary with an associated path to the indicant set equal to one. The measures of past computer experience, computer staff support, and ease of system use were exogenous in the model. Each had its standard deviation set equal to one. For these measures, the associated path between each indicant and its measure was free to vary. The degree of system use, measured as the percentage of time spent using the system, was an observable, exogenous variable. The three remaining measures in the model were endogenous. For these three measures, the path from the measure to one of its indicants was set equal to one in order to scale the measure. The endogenous measures were also influenced by disturbance terms free to vary with associated paths to the measure set equal to one. The hypotheses were expressed as paths between appropriate measures.

The model was estimated using Calis in PC SAS version 6.12. The estimation method was maximum likelihood. The resulting goodness of fit index was 0.91 and when adjusted for the degrees of freedom in the model it was 0.88. The root mean square residual was 0.08. The chi-square statistic was 454.40 with 217 degrees of freedom. It was also statistically significant. The normed version of the chi-square statistic was 2.09. Bentler's comparative fit index was 0.96 and the incremental fit indexes (i.e., Bentler & Bonett's and Bollen's normed and non-normed indexes) ranged from 0.90 to 0.96. These results, even given the significant chi-square statistic, indicate a good fit between the model and the data, particularly when considering the size of the sample used (Hair, Anderson, Tatham & Black, 1992).

---