3.3.4. Structural Equation Model Analysis
The basic idea of structural equation modeling (SEM) to establish a pattern of interconnections among the factors of interest based on previous concepts and existing understanding. This pattern is derived and assumed, followed by measurement, which approves the coefficients of the exogenous variables and constructs a covariance matrix known as the pattern matrix. Testing the correctness of the equation model involves assessing the ability of the constructed hypothetical model and data matrix to integrate with each other. If the proposed hypothetical approach can integrate the actual sampling effect, it indicates the correctness of the modeling [
40].
The main purpose of structural equation modeling is to construct specific outcome models to study the linkages between variables and changes during the hypothesis period. The model is usually described by a system of linear equations and consists of two parts: an econometric model and a structural model. The econometric model describes the relationship between the latent and observed variables, through which the latent variables can be defined based on the observed factors, whereas the constructive model describes the association between the latent variables. The relationship between the econometric model and the constructive model can be represented by a matrix equation as follows [
41]:
Structural equation modeling considers certain phenomena that cannot be directly observed, but rather when one wants explore, in depth, the latent variables. One then uses certain variables (indicators) that can be directly observed in order to express such latent variables and to then use them to determine the structural links among the latent variables. It is a type of statistical instrument that studies the macroeconomic change patterns of micro-individuals. Regarding the association between latent variables, it can typically be written as the following structural equation:
where the place η is the endogenous latent variable, B is the relationship between the endogenous latent variables, Γ is the impact of the exogenous latent variable on the endogenous latent variable, ξ is the exogenous latent variable, and ζ is the residual time period of the structural equation [
42,
43]. The relationship between the goal and latent variables with appreciation to every difference is frequently written as per the following dimension equation [
44,
45]:
Structural equation modeling has the blessings of permitting the computational blunders contained in the unbiased and established variables, as well as allows one to deal with multiple independent variables at the same time when compared to other models. Therefore, structural equation modeling is a wide ranging mathematical model that can deal with many problems in higher education, business management, market economy, tourism, psychology, socialism, and other professions [
45].
The above table provides an assessment of the impact of the potential variables and an assessment of the impact results (
Table 7). In terms of impact results, the following conclusions can be drawn from the above table. The relationship between perceived usefulness and perceived ease of use on usage attitudes was above 0.05, with systematic coefficients of 0.287 and 0.701, respectively. This suggests that perceived ease usability has a high effect on usage attitudes [
46]. The attitude toward use variable has a strong influence on human loyalty. Meanwhile, the “Summary Table of Model Regression Coefficients” shows all the measured relationships, which can be visualized in the above table, where “-” means that the item is a reference item and is not an output [
46]. The table above indicates the size relationship if the size relationship is good, where the normalized load coefficient cost is essentially higher than 0.6.
The table above indicates the impact of the mannequin fitting (
Table 8). There were many healthy mannequin indicators, and SPSSAU, alone, lists all of them. Few fashions can make all the match symptoms meet the standard; however, the most used in shape symptoms are endorsed to be inside the ideal range. There are many structural equation modeling SEM match metrics. However, there are no constant necessities for which metrics to use. Most research will solely use a few of these metrics for the purposes of document presentation.
In this experiment, the chi-squared levels of the freedom ratio was 2.219, which is much less than 3, thereby indicating the clear and fantastic shape of the model. Additionally, the RMSEA fee is 0.050, which is equal to 0.05, whilst the RMR fee is 0.034, which is much less than 0.05. Meanwhile, the GFI, CFI, and AGFI values were all larger than 0.9, and the NFI fee was 0.955, which is higher than 0.9 but drawing near 0.9, whilst all the different parameters were additionally inside the regular range. For that reason, this indicated that the mannequin was properly mounted and that the mannequin information was credible.
The above table shows the results of the covariance relationship MI indicator, and this time the parameter is set to MI > 10, before output (
Table 9). If the model fits the indicator poorly, one should then consider the need to establish the covariance correlation between B1 and D1 after analyzing again. This table contains the relationships between several variables. The interaction between these variables is measured by the mutual information value. The higher the mutual information value, the stronger the correlation between the two variables. Each relationship is also given a ‘parameter change’, which indicates how the mutual information value will change when the parameter of a variable is changed. These relationships include the interactions between PE0U2 and UB3, PE0U1 and UB2, A2 and PU3, A1 and BI3, A1 and BI1, UB6 and UB3, UB6 and PU2, UB6 and BI1, UB6 and A2, and UB5 and PE0U2.
The table above displays the ‘Influence Relationship MI Metric’, which indicates that the current parameter setting requires an MI value that is greater than 10 in order to generate output (
Table 10). The absence of data in the table implies that all MI values in the influence relationships are greater than 10. However, researchers can reset the threshold to MI > 3 for the purposes of the output MI data, although such values in the single digits are usually considered small. If there is a large MI value (e.g., greater than 20), it may be necessary to rebuild the influence relationship model. The relationship in the first row is “Effect of Usage Behavior on Attitude to Use”, which has a mutual information value of 10.266, and if one of the parameters of Usage Behavior is changed, this results in an increase in the mutual information value of 10.937. The relationship in the second row, “Effect of Usage Attitude on Perceived Usefulness”, has a mutual information value of 12.495, with a change in one of the parameters of Usage Attitude resulting in a decrease in the mutual information value. A change in one of the parameters of Usage Attitude would result in a decrease in the mutual information value of 4.514.
Whether to incorporate the influence relationship established in the table into the model (i.e., in order to rebuild the model) depends on two factors: (1) whether professional knowledge allows it, and (2) whether there are changes in the cost and model evaluation of the MI index. If the expert’s understanding does not permit it at present, the model cannot be modified, even if the cost of MI is significant. It is not advisable to arbitrarily adjust the model for the sake of meeting standards. The absence of information in the MI values in the table implies that all of the MI values were far less than 10, and that there was no need to adjust the model any further (as it will not improve the results). Therefore, the current model is the final result for the influence relationships.
The above matrix displays the fitted results for the R-squared values, and the significance of this indicator lies in the strength of its interpretation by the influenced term, i.e., Y, which is influenced by the influenced term X (
Table 11). The table shows a series of Independent variables and their respective R-squared values in relation to the dependent variable. In this table, ‘perceived ease of use’ is an independent variable that has a relatively high R-squared value (0.991) with the dependent variable. This means that there is a strong correlation between perceived ease of use and the dependent variable in the data sample, and that perceived ease of use can be used to predict the dependent variable. In other words, this R-squared value suggests that perceived ease of use has a high explanatory power for explaining changes in the dependent variable in this particular study, while other independent variables may have a lower explanatory power for the dependent variable. This suggests that perceived ease of use is more important to the dependent variable than the other independent variables in this study.
After summarizing the above, it can be concluded that the final coefficients of each result are organized and summarized inside a graph, which facilitates the display of the results and thus establishes the user for plotting in SPSSAU, specifically of the following type.
The figure above is drawn for SPSSAU and shows the ‘measurement relationship’ and ‘impact relationship’ (i.e., the standardized regression coefficients in the summary table of model regression coefficients) as well as the covariance relationship (i.e., the standardized estimated coefficients in the table of covariance relationships). The equation model contains a number of variables and their interrelationships, each represented by a box. The arrows indicate the paths in the equation, i.e., the relationships between the variables. The number on the arrow indicates the path coefficient or standardised regression coefficient, which indicates the strength and direction of the interaction between the two variables. As the above figure shows, perceived usefulness, perceived ease of use, attitude toward use, and behavioral intention all influence the final usage behavior, but it can also be seen that perceived usefulness and perceived ease of use are the first to influence users’ intention to use.
Users’ perceived usefulness of the DingTalk platform was positively correlated with their willingness to use it. It can be seen from the questionnaire that the perceived usefulness was directly related to the users’ willingness to use the Internet product, as is shown in
Figure 2. Furthermore, the DingTalk platform needs to make users aware that it can provide the information and services needed to help solve problems and can meet their usage needs, and only then will users be willing to use it; otherwise, users may choose traditional Internet products, such as WeChat and QQ.
DingTalk users’ perceived ease of use and willingness to use are closely related, as is shown in
Figure 2 from the collected questionnaires. This shows that perceived ease of use is closely related to the users’ willingness to use. Users who feel comfortable using the DingTalk platform are more likely to accept the benefits of the service and are therefore more willing to use it. As the DingTalk platform is easy to use, users rated the ease of use higher [
47].