4.1 Confirmatory Factor Analysis
In this study, confirmatory factor analysis is used to judge whether the hypothetical relationship between observed variables and latent variables is consistent with the data, that is, the unique validity of each construct. We used AMOS 23.0 to complete the confirmatory factor analysis (Figure
2).
The confirmatory factor analysis model results show that the p-value of each path coefficient is far less than the significant level (Table
3). Generally speaking, the model has statistical significance, and the designed theoretical indicators can effectively measure each part. In addition, there are 11 model adaptation indicators in part of the model, and only five of them are near to standard (AGFI = 0.893, NFI = 0.874, IFI = 0.894, TLI = 0.833, CFI = 0.892), among which the number of absolute fitness indicators is two (RMSEA = 0.094, GFI = 0.944), indicating that the model adaptation effect can be improved. After adjustment, we deleted one path and added some paths. As a result, the model was fitted. Table
4 shows the results.
In this study, we revised the index values according to the model parameters, combined with the design assumptions of the SEM model, established the covariant relationship among the errors, and revised the initial confirmatory factor analysis model. After it is modified, the χ²/df of the model is 3.122. The p-value of each path coefficient is far less than the significance level, which shows that the model has statistical significance. That is, the designed theoretical index can effectively measure each part. Most of the model adaptation indicators have reached the adaptation standard, including 11 model adaptation indicators, of which nine are up to standard, among which all the absolute adaptation indicators are up to standard, which is better than the initial model data, indicating that the overall model adaptation situation of confirmatory factor analysis is good.
4.2 Structural Equation Model
We can clearly see in Figure
3 left picture that the three control variables have no influence on digital trust, and their p-values are all greater than 0.05, which means that the hypothesis is not valid. It means that differences in gender, age, and educational background do not affect digital trust. In addition, the
user expectation's (UE) p-value is greater than 0.05, which indicates that it has no direct impact on
digital trust (DT).
The initial structural equation model results show that the p-values of most path coefficients are less than the significance level, which indicates that the model has statistical significance. However, some path coefficients are not significant, which does not conform to the theoretical assumptions. Moreover, most of the adaptation indicators of the model are not up to standard, and only three of them are up to standard, among which the number of appropriate indicators of value-added allocation is zero, indicating that the adaptation effect of the model needs to be improved. According to the model revision index, we revised the initial model and deleted some paths with a p-value larger than 0.05. Besides, we deleted some paths that do not conform to theoretical assumptions too. Table
5 also shows the factor load and p-value of each path of the revised structural equation model. It can be seen that the three control variables of gender, age, and educational background still have no influence on digital trust. User expectations have no direct impact on digital trust, and other paths have passed the test. In this study, according to the modified index value of model parameters, combined with the design assumption of the SEM model, we established the covariant relationship between errors and modified the initial equation structural model. P-values of all path coefficients are far less than the significance level, which indicates that the model has a high degree of interpretation. Most of the model adaptation indicators have reached the adaptation standard or critical value. Of a total of 11 model adaptation indicators, 9 are up to standard. The absolute fit indicators are up to standard, better than the initial model data, indicating that the revised model adaptation situation is good.
The above analysis results show that the H4 hypothesis (user expectation has a positive impact on digital trust) failed the test. The exciting discovery in this paper is that user expectation has no effect on digital trust, which is contrary to our hypothesis. However, Hypothesis
7 (User satisfaction plays an intermediary role between user expectation and digital trust) passed the test. This result demonstrated that user satisfaction really plays an intermediary role between user expectation and digital trust, and it is also a complete intermediary role. In addition, Hypothesis
3 also passed the test. User expectation has a positive impact on user satisfaction. A further novel finding is that user satisfaction plays an intermediary role between user perception and digital trust, which means Hypotheses
1 and
2 passed the test. In addition, the control variables (age, gender, and educational background) have no influence on digital trust.
According to the seven hypotheses of the model design, a total of six hypotheses are passed. The standardized direct effect of user perception on user satisfaction is 0.240. The results clearly show that when other conditions are unchanged, every time the “user perception” latent variable increases by 1 unit, the “user satisfaction” latent variable will increase by 0.240 units. This is consistent with what has been found in previous studies [
21,
22]. They found that user perception has a positive impact on user satisfaction. In fact, users perceive the characteristics of information security, service quality and efficiency of digital government in the digital environment, which will indeed improve their satisfaction. User perception is the cornerstone of user satisfaction. Without user perception, the measurement of user satisfaction is inaccurate. Since the standardized direct path coefficient of user perception to user satisfaction is 0.240, and the standardized direct path coefficient from user satisfaction to digital trust is 0.235, so the indirect effect of user perception to digital trust is 0.056. This result shows that when other conditions remain unchanged, the latent variable of user perception increases by 1 unit, the latent variable of digital trust will indirectly increase by 0.256 units. The standardized total effects of user perception to digital trust is 0.248. Because the standardized total effect refers to the sum of standardized direct effect and standardized indirect effect. This result indicates that Hypothesis
2 passes the test, that user perception does have a direct positive impact on digital trust. However, previous studies have shown that user perception has a negative impact on digital trust [
23]. This is mainly because the composition of user perception is different from his research. In this paper, user perception is composed of perceived security, quality and efficiency, these features are positive words. In previous studies, user perception generally refers to risk perception. Obviously, the greater the risk the user perceives, the smaller the trust is. In addition, user perception also affects digital trust through user satisfaction, which is a very interesting conclusion. Although it is somewhat different from previous studies, it brings infinite enlightenment to the development of digital government.
The standardized direct effect of user satisfaction on digital trust is 0.235. The results clearly show that when other conditions are unchanged, every time the “user satisfaction” latent variable increases by 1 unit, the “digital trust” latent variable will increase by 0.235 units. The standardized indirect effect is 0.00, and the standardize total effect is 0.235. This result proves that Hypothesis
5 passed the test. This finding is directly in line with previous findings [
27]. Yang and Feng found user satisfaction will affect user use intention and digital trust. Considering the reality, users feel satisfied with the use of digital government services, which will build his digital trust in digital government, so that he can use digital government services again and countless times. The standardized direct effect of user expectation on user satisfaction is 0.515. The results clearly show that when other conditions are unchanged, every time the “user expectation” latent variable increases by 1 unit, the “user satisfaction” latent variable will increase by 0.515 units. This result is consistent with the results of Wang and Wang [
25] and Huangfu [
26]. In fact, users' expectations for the ease of use and usefulness of digital services are indeed related to user satisfaction. If their expectations are met, their satisfaction will naturally increase. A further novel finding is that user expectations don't have a direct impact on digital trust, it can only indirectly affect digital trust through user satisfaction, and user satisfaction is the intermediary variable between user expectations and digital trust. This result overturns Hypothesis
4 (User expectation has a positive impact on digital trust) and confirms Hypothesis
7 (User satisfaction plays an intermediary role between user expectation and digital trust). From the practical point of view, it is difficult for user expectations to have a direct impact on digital trust. User expectations only affect user satisfaction, and digital trust is not directly linked.