This chapter tests the third chapter of the research hypothesis and conceptual framework proposed in this article, as well as the research methodology. This chapter is divided into eight parts. The first part introduces the relationship between the research methods used in this research for multi-dimensional dynamic data collection and satisfaction. The second part introduces the sample group of this study (target population, sampling unit, sample size and sampling procedure). The third part defines the research tools of this study, which describe the different parts of the questionnaire. The fourth part describes the validity, consistency and reliability of the internal research device. The fifth part classifies the pilot tests. The sixth part describes the data collection process. The seventh part is the maintenance of statistical data. Descriptive analysis is applied to the demographic analysis influencing factors of the interviewees, and by modeling the structure, the reasoning analysis is used to test hypotheses and statistically maintained equations. The last part of this chapter is to summarize the statistical tools used in hypothesis testing.
Research Methods Used
This study adopted a quantitative method using questionnaires as data collection tools. Participants must complete a separate questionnaire individually. They can fill out on-site or visit online. Questionnaires distributed to students of three universities through Questionnaire Star [19]. The survey contains many questions about college students’ satisfaction with campus facilities.
Besides, before the researcher distributed the questionnaire to the target sample, the questionnaire had been pre-tested on a small number of students and is considered reliable and convenient to use. The researchers also verify the reliability of each structure using a modified research tool. And after collecting the quantitative data, the researcher analyzed all the data, using structural equation modeling to process the results of this research. It stated that the questionnaire is a research tool that consists of a series of questions, aiming to collect information from the respondents. It can also be thought of as a written interview, which can be done face-to-face by phone, computer or post. Questionnaire surveys provide a relatively cheap, fast and effective way to obtain a lot of information from a large number of people. Questionnaire surveys are cheaper and faster than other methods to measure the behaviors, attitudes, preferences, opinions and intentions of a relatively large number of subjects.
Respondents and Sampling Procedure
It claimed that sampling survey is a method used to collect data from or about members of the population so that inferences about the entire population can be obtained from a subset or sample of the members of the population (Williams & Onsman, 2010). Properly conducted sample surveys will support inferences drawn from the sample about the scientific basis for the population as a whole. The population must be determined before selecting a sampling technique. In fact, since structural equation modeling (SEM) is a statistical processing method in this research, it is also important to determine the appropriate sample size. Therefore, this section describes the target population, sampling unit, sample size, and sampling procedures of this study.
Target population
Article concluded the target population is an informal term mainly used in epidemiology. A general definition refers to a group or group of elements to know more about. Most of the time, "target population" and "population" are synonymous. However, adding the word “target” emphasizes that sometimes we ignore the mark during sampling and do not always achieve the goal: the sample may not represent the population we initially wanted to proceed with the sample. For example, we may want to investigate all hospitalized adults (target population) in the United States, but the budget limits your survey to four cities in the United States. The sample population and the target population in this case may be very different.
In some aspects, such as regression analysis in epidemiology, it is particularly important to determine the target population. And data analysis in science always includes the right units. This ignores specific information about the population. For example, suppose the target population is people living in the United States: do the target population include only citizens? residents of foreigners? Number of refugees? If data analysis run a return involving people without specifying the target population, it can lead to "hopeless ambiguity " (Lavrakas, 2008).
The target population of this study was at three universities, Southwest University (SWU), Chongqing Normal University (CQNU), and Chongqing Electronic Engineering Vocational College (CQVC-EE), and a business school student in Chongqing. The three higher educations in Chongqing are determined by the number of students in the undergraduate year in Chongqing's business sector. The results are summarized in Table 1 and Table 2.
Table 1
The list of students from the three universities in Chongqing was used as the sampling unit.
Source: Illustrated by YanLi Chen
This study used multistage sampling with probabilistic and non-probability sampling components as the sampling procedure to determine appropriate sampling units and identified 500 students as the final sample for this paper.
Sampling Units
The researcher divided the target population into eight sampling units according to their professional direction, educational background, and previous academic achievements. The researcher will personally release the questionnaire of this study to the target university and complete the survey with the assistance of the staff of the target secondary schools of Southwest University.
According to the research (Clark-Carter, 2010) the selection of sampling methods is related to the research object, educational background, time, and accuracy. It deduced that sampling is a reasonable definition by selecting a limited number of things or fragments from the total population to provide clear evidence for the total population (Srinivasan, 1985). According to Wang et al. (2014), more sampling procedures should be determined. In representative data collection, appropriate sampling techniques should be selected to determine the universality and quality of the survey. The survey results should be provided to the study’s target population, not only to the sample; otherwise, there will be no research and target value (Lavrakas, 2008).
In this study, the accurate target population was obtained by judgment sampling. Consequently, in this empirical academic research, the researcher distributed the in-person questionnaires among the undergraduates and postgraduates majoring in physical, art design, history, culture, and English subjects with the assistance from the staff who were responsible for the student management in each of the selected academies from the Southwest University of China. According to the screening criterion of “sufficient experience with online learning”, 372 undergraduate students and 114 postgraduates were filtered out for their inadequate online learning experience. A total of 4978 undergraduates and 1000 postgraduates were suitable for the criteria of this dissertation and have been selected as the first stage samples.
This research adopts multiple sampling methods, divided into three stages: judgmental sampling for the initial, quota sampling for additional, and judgmental sampling again for the final. Therefore, the researcher divided the first period of samples into eight sub-units or sampling units according to their basic characteristics and education level. With the help of the staff from the secondary colleges, the researcher will arrange an appropriate number of undergraduates and postgraduates according to the proportion of the quota sampling scheme. Finally, 492 undergraduates and 481 postgraduates would have been selected as the final stage samples.
Table 2
The student list by year of three universities in Chongqing, China were selected as sample units.
Data Collection and Analysis
In this section, the researchers conducted a cross-sectional comparative study of the two data sets by Descriptive Statistics of Demographic Characteristics. First, the researchers conducted a descriptive analysis of 492 undergraduate respondents' personal information, such as gender, university information, years of study, major orientation, and other demographic characteristics at Southwest University in Chongqing. Among the respondents, 182 (37%) were female, and 310 (63%) were male. In terms of professional distribution, 92 (18.7%) were history and culture, 188 (38.3%) were English, 110 (22.4%) were physics, and 101 (20.6%) were art and design. The results are summarized in Table 3.
Table 3
Interview Analysis of Undergraduate Students
Next, the researcher conducted a descriptive analysis of the personal information of 481 graduate student respondents from Southwest University in Chongqing, such as gender, university information, years of study, professional orientation, and other demographic characteristics. Among the survey respondents, 287 (59.7%) were female, and 194 (40.3%) were male. In terms of professional distribution, 134 (27.8%) were history and culture, 154 (32.1%) were English, 77 (16.1%) were physics, and 115 (24%) were art and design. The means and standard deviations are shown in Table 4.
Table 4
Interview analysis of graduate students
Descriptive Analysis of Observed Variables
The descriptive statistics of the researcher post-measurement scale were analyzed. This included analysis of centralized trends and variability in the data set. The mean (X̅) was used to measure the centralized trend or center of the data set, and the standard deviation (SD) was used to measure the dispersion or variability within the data set. The mean is the average number of items in the data set, calculated by dividing the sum of all quantities by the number of items in the data set (Kenney & Keeping, 1962). The standard deviation is a statistical measure of dispersion. The standard deviation determines the extent to which each result in a data collection deviates from its mean (Underhill & Bradfield, 1998).
In this study, the researcher analyzed data of undergraduate respondents, using a 5-point Likert scale to measure the latent variables: perceived ease of use, usefulness, system quality, service quality, self-efficacy, information quality, and satisfaction. The mean range for all observed variables was 3.13 ~ 3.87, and the standard deviation range was 0.991 ~ 1.193. The means and standard deviations are shown in Table 5.
Table 5
Statistics of graduate measurement scales
At the same time, the researcher conducted the same analysis on the data from graduate student respondents, using a 5-point Likert scale to measure the latent variables: perceived ease of use, perceived usefulness, system quality, service quality, self-efficacy, information quality, and satisfaction. The mean range for all observed variables was 3.19 ~ 3.86, and the standard deviation range was 0.972 ~ 1.215. The means and standard deviations are shown in Table 5
Table 6
Descriptive Statistics of Undergraduates Measurement Scales
Convergent Validity
Convergent validity affirms the consistency of relationships between constructs regardless of the methods used for measurement (Churchill, 1979). The methods used to measure convergent validity are Cronbach’s Alpha reliability, factor loading, composite or construct reliability, and average variance extracted. The results are summarized in Table 6.
Cronbach’s alpha (CA) is a statistical test used to assess the internal consistency of items within a construct (Killingsworth et al., 2016). The higher the value of Cronbach’s alpha, the higher the reliability of the item. The value of Cronbach’s alpha ranges from 0 to 1, with values between 0.7 and 0.8 acceptable or very good. Values between 0.8 and 0.9 are considered very good, while values above 0.9 are considered excellent (Hair et al., 2003). The questionnaire was distributed to 500 students with similar characteristics to the target respondents for the reliability or pilot testing. The Cronbach’s alpha values for all the constructs listed in Table 8 were greater than 0.7. Therefore, the internal consistency of these items was confirmed for questionnaire distribution. The internal consistency of perceived ease of use, perceived usefulness, perceived satisfaction, social influence, performance expectancy, facilitation, and behavioral intention was very good, with Cronbach’s alpha values of 0.869, 0.866, 0.873, 0.864, 0.853, 0.881, and 0.887, respectively.
Factor loadings measure the coefficients between the groups of conformations (O’Rourke & Hatcher, 2013). The higher the value of the factor loadings, the higher the reliability of the items (Hair et al., 2010). An acceptable factor loading threshold is 0.5 or higher (Hair et al., 1998). In this study, all individual items had factor loadings greater than 0.50, and most were above 0.70, ranging from 0.561 to 0.915, as shown in Table 8.
Composite or construct reliability (CR) and average variance extraction (AVE) are alternative measures of scale item reliability and consistency (Peterson & Kim, 2013). According to Fornell and Larcker (1981), values of 0.7 or above and 0.4 or above for CR and AVE, respectively, are acceptable. The results of CR in this study were above the threshold values. The composite reliability values ranged from 0.859 to 0.887. The mean was also greater than 0.4 and ranged from 0.576 to 0.726.
Table 7
Confirmatory Factor Analysis Result, Composite Reliability (CR) and Average Variance Extracted (AVE) undergraduate
Based on the data illustrated in Table 7, the entire value of the factor loading, average variance extracted (AVE), and composite reliability (CR) were at an acceptable level, which indicated that the convergent validity was well established.
Similarly, the second set of data in this study had factor loadings greater than 0.50 for all individual items, and most were above 0.70, ranging from 0.561 to 0.915, as shown in Table 8. Composite or construct reliability (CR) and average variance extraction (AVE) are alternative measures of scale item reliability and consistency (Peterson & Kim, 2013). According to Fornell and Larcker (1981), values of 0.7 or above and 0.4 or above for CR and AVE, respectively, are acceptable. The results of CR in this study were above the threshold values. The composite reliability values ranged from 0.837 to 0.900. The mean was also greater than 0.4 and ranged from 0.526 to 0.746. The results are summarized in Table 7.
Table 8
Confirmatory Factor Analysis Result, Composite Reliability (CR) and Average Variance Extracted (AVE) postgraduate
Note
SE = Standard Error, ***=p < 0.001; **=p < 0.01; *=p < 0.05.
Source Constructed by the author
Discriminant validity is confirmed when the AVE's square root is larger than any intercorrelated construct coefficient (Fornell & Larcker, 1981). As illustrated in Table 8, the square root of AVE for all constructs at the diagonal line was greater than the inter-scale correlations. Hence, the discriminant validity was guaranteed.
Table 9
Discriminant Validity undergraduate
Discriminant validity is confirmed when the AVE's square root is larger than any inter-correlated construct coefficient (Fornell & Larcker, 1981). As illustrated in Table 9, the square root of AVE for all constructs at the diagonal line was greater than the inter-scale correlations. Hence, the discriminant validity was guaranteed.
Table 10
Discriminant Validity postgraduate
Source Constructed by the author
Research Framework
The structural model was evaluated through structural equation modeling to confirm the applicability of the model, the causal relationships between variables, and the influencing factors affecting online education at Chongqing Southwestern University. The structural model shows the path or relationship between potential variables, which can be direct or indirect (Byrne, 2010).
Fitness of Structural Model
The fit of the structural model was evaluated by using a fit index. The fit indices chosen are the same as for the CFA. They include the chi-square statistic CMIN/DF), the fit index (GFI), the adjusted fit index (AGFI), the normative fit index (NFI), the comparative fit index (CFI), the Tucker-Lewis index (TLI), and the root mean square error of approximation (RMSEA). These indices will assess seven variables; it is perceived ease of use, perceived usefulness, self-efficacy, service quality, system quality, information quality, and satisfaction.
The model's fit was evaluated by comparing the statistical values of the indices with the acceptable fit values in Table 10. The statistical values of the indices were CMIN/DF = 2.735, GFI = 0. 866, AGFI = 0. 843, NFI = 0. 879, CFI = 0. 919, TLI = 0. 911, RMSEA = 0. 059. numerically, GFI, AGFI, NFI, and TLI indices were unacceptable. Therefore, the structural model was modified, and the fit was recalculated. The results are summarized in Table 11.
Table 11
Goodness-of-Fit for Structural Model before Adjustment(undergraduate)
Remark
CMIN/DF = The ratio of the chi-square value to degree of freedom, GFI = Goodness-of-fit index, AGFI = Adjusted goodness-of-fit index, NFI = Normed fit index, CFI = Comparative fit index, TLI = Tucker-Lewis index, and RMSEA = Root mean square error of approximation
Figure 1: Structural Model Before Adjustment
Source: Constructed by the author
The modifications to the structural model were performed by correlating the measurement errors between the items in the constructs. Based on the modified structural model, the goodness-of-fit indices were recalculated in Table 12. The results of the statistical values were CMIN/DF = 1.910, GFI = 0.908, AGFI = 0.908, NFI = 0.916, CFI = 0.958, TLI = 0.954, and RMSEA = 0.043. The applicability of the structural model was confirmed.
Table 12
Goodness-of-Fit for Structural Model after Adjustment(undergraduate)
Remark
CMIN/DF = The ratio of the chi-square value to degree of freedom, GFI = Goodness-of-fit index, AGFI = Adjusted goodness-of-fit index, NFI = Normed fit index, CFI = Comparative fit index, TLI = Tucker- Lewis index, and RMSEA = Root mean square error of approximation
Figure 2
Structural Model After Adjustment
The model's fit was evaluated by comparing the statistical values of the indices with the acceptable fit values in Table 11. The statistical values of the indices were CMIN/DF = 3.163, GFI = 0.843, AGFI = 0.818, NFI = 0.843, CFI = 0.886, TLI = 0.876, and RMSEA = 0.066. numerically, GFI, AGFI, NFI, and TLI indices were unacceptable. Therefore, the structural model was modified, and the fit was recalculated. The details were recalculated in Table 13.
Table 13
Goodness-of-Fit for Structural Model before Adjustment(postgraduate)
Remark
CMIN/DF = The ratio of the chi-square value to degree of freedom, GFI = Goodness-of-fit index, AGFI = Adjusted goodness-of-fit index, NFI = Normed fit index, CFI = Comparative fit index, TLI = Tucker-Lewis index, and RMSEA = Root mean square error of approximation
Source: Constructed by the author
Figure3
Structural Model Before Adjustment
The modifications to the structural model were performed by correlating the measurement errors between the items in the constructs. Based on the modified structural model, the goodness-of-fit indices were recalculated in Table 14. The results of the statistical values were CMIN/DF = 2.843, GFI = 0.854, AGFI = 0.827, NFI = 0.878, CFI = 0.917, TLI = 0.908, and RMSEA = 0.061. The applicability of the structural model was confirmed.
Table 14
Goodness-of-Fit for Structural Model after Adjustment (postgraduate)
Remark
CMIN/DF = The ratio of the chi-square value to degree of freedom, GFI = Goodness-of-fit index, AGFI = Adjusted goodness-of-fit index, NFI = Normed fit index, CFI = Comparative fit index, TLI = Tucker-Lewis index, and RMSEA = Root mean square error of approximation
Figure 4
Structural Model After Adjustment
Research Hypothesis Testing
The degree of correlation between the independent and dependent variables proposed in the hypothesis is measured by regression coefficients or standardized path coefficients. As shown in Table 14 and Fig. 4, the seven hypotheses proposed were supported by the study's data in Chengdu. Ioannidou and Erduran (2020) are convinced that quantitative research aims to test the proposed hypotheses by considering various independent and dependent variables. Therefore, the seven alternative hypotheses are described in the following.
The alternative hypotheses are tested and justified in this section after modifying the matrix of structural equations for which the fit meets the relevant criteria. Table 16 summarizes the details of the results of all alternative hypothesis tests.
Hypothesis Testing Result of the Structural Model
Note
SE = Standard Error, ***=p < 0.001; **=p < 0.01; *=p < 0.05.
Source
Constructed by the author
Explanation of the Results for Hypotheses Examining
According to the data in Table 15, all assumptions are supported, as detailed below.
Hypothesis 1
The significant causal correlation between information quality has a significant influence on satisfaction.
Based on the results of the above study, the researchers derived the following extended results.
Hypothesis 1
confirms that satisfaction is one of the important factors of information quality, with a standardized path coefficient value of 0.120*** and a t-value of 3.025 in the structural method. An information system is an important subject in the management information system field; some researchers took the performance method of information systems as an important dependency factor in evaluating the institute information system (DeLone & McLean, 1992).
Hypothesis 2
The significant causal correlation between system quality has a significant influence on satisfaction.
Similarly, hypothesis 2 confirms that satisfaction is one of the second most important factors of system quality, with a standardized path coefficient value of 0.298*** and a t-value of 6.788 in the structural method. Yuce et al. (2019) also pointed out that the system quality plays a great role in the final impression of users and evaluators that with the improvement of the quality of online teaching, the online teaching system quality can better meet the needs of learners and complete teaching tasks more effectively, so as to improve users' satisfaction with learning.
Hypothesis 3
The significant causal correlation between service quality has a significant influence on satisfaction.
At the same time, hypothesis 3 confirms that satisfaction is one of the important factors of service quality, with a standardized path coefficient value of 0.105*** and a t-value of 2.499 in the structural method. Yuce et al. (2019) also demonstrated that service quality plays a great role in the final impression of users' and evaluators' satisfaction.
Hypothesis 4
The significant causal correlation between perceived usefulness has a significant influence on satisfaction.
Similarly, hypothesis 4 confirms that satisfaction is one of the important factors of perceived usefulness, with a standardized path coefficient value of 0.230*** and a t-value of 4.882 in the structural method. Perceived usefulness is regarded as the main influencing factor of representative satisfaction (Yuan et al., 2016).
Hypothesis 5
The significant causal correlation between perceived ease of use has a significant influence on satisfaction.
Next, hypothesis 5 confirms that satisfaction is one of the factors of perceived ease of use importance, with a standardized path coefficient value of 0.356*** and a t-value of 7.588 in the structural method. Yu et al. (2012) and Roca et al. (2006) considered that perceived ease of use of online public systems is associated with satisfaction.
Hypothesis 6
The significant causal correlation between self-efficacy has a significant influence on perceived usefulness.
Finally, hypothesis 6 confirms that satisfaction is one of the important factors of perceived ease of use, with a standardized path coefficient value of 0.221*** and a t-value of 5.520 in the structural method. Thompson and Cope (2002) also showed that when users conduct online searches, the specially set network self-efficacy, an individual's tendency to a specific function, will significantly affect the search results and perceived usefulness.
Based on the data in Table 16, all of the six hypotheses proposed by the researcher for this study were supported, as follows.
Table 16
Hypothesis Testing Result of the Structural Model
Note
SE = Standard Error, ***=p < 0.001; **=p < 0.01; *=p < 0.05.
Source
Constructed by the author
Hypothesis 1
The significant causal correlation between information quality has a significant influence on satisfaction.
For the second set of data in this paper's study, the researchers came up with the following extended results. Hypothesis 1 confirms that satisfaction is one of the important factors of information quality, with a standardized path coefficient value of 0.383*** and a t-value of 7.626 in the structural method. Information quality is a major factor in satisfaction and has a large impact on satisfaction (Aparicio et al., 2017).
Hypothesis 2
The significant causal correlation between system quality has a significant influence on satisfaction.
Similarly, hypothesis 2 confirms that satisfaction is one of the second most important factors of system quality, with a standardized path coefficient value of 0.309*** and a t-value of 6.231 in the structural method. According to Aldholey (2018), the correlation between overall quality and system quality is a prerequisite for the happiness of online learning users, which is a closely significant impact on learning satisfaction.
Hypothesis 3
The significant causal correlation between service quality has a significant influence on satisfaction.
Meanwhile, hypothesis 3 confirms that satisfaction is one of the important factors of service quality, with a standardized path coefficient value of 0.122*** and a t-value of 2.649 in the structural method. Spreng et al. (1996) considered that products and services could meet people's expectations, so system quality will affect satisfaction with the service to some extent.
Hypothesis 4
The significant causal correlation between perceived usefulness has a significant influence on satisfaction.
Similarly, hypothesis 4 confirms that satisfaction is one of the important factors of perceived usefulness, with a standardized path coefficient value of 0.095*** and a t-value of 2.003 in the structural method. Limayem and Cheung (2008) found that perceived usefulness affects students' satisfaction and willingness to continue using online education.
Hypothesis 5
The significant causal correlation between perceived ease of use has a significant influence on satisfaction.
Next, hypothesis 5 confirms that satisfaction is one of the factors of perceived ease of use importance, with a standardized path coefficient value of 0.368*** and a t-value of 7.234 in the structural method. Masrek and Gaskin (2016) and Cheng (2020) also suggested that there are subjective cognitive differences in information systems, and perceived ease of use is one of the important factors that can determine whether users would adopt information systems and their satisfaction.
Hypothesis 6
The significant causal correlation between self-efficacy has a significant influence on perceived usefulness.
Finally, hypothesis 6 confirms that satisfaction is one of the important factors of perceived ease of use, with a standardized path coefficient value of 0.223*** and a t-value of 4.180 in the structural method. It has been found that online learning self-efficacy has a positive impact on online learning satisfaction (Hamdan et al., 2021).
Direct, Indirect, and Total Effects of Relationship
In the study of this paper, the relationship between the variables can have both direct and indirect effects, and the AMOS software can calculate and determine these effects. Direct effect (DE) relationships mean that the two variables are related, and there is no effect of the moderating variable. Conversely, these indirect effects (IE) relationships mean that the variables are correlated through at least one moderating variable only. The total effect (TE) is the sum of the direct and indirect effects of the path of the relationship (Raykov & Marcoulides, 2000). R-squared (R2) values indicate the proportion of change in the dependent variable (Fornell & Larcker, 1981). In other words, the r-squared value determines the variance in the variable that can be explained by other variables (Henseler & Sarstedt, 2012). An acceptable level of R2 is at least 0.1 (Falk & Miller, 1992).
In this study, there are five independent variables, including information quality, system quality, service quality, perceived ease of use, and self-efficacy, with perceived usefulness as the intermediate variable and satisfaction as the dependent variable, and Table 17 summarizes the statistical results of the direct, indirect, and overall effects. Corresponding path analysis data are shown in Fig. 3.
Table 17
Direct, Indirect, and Total Effects of Relationships
Note
IV = Independent Variable, DV = Dependent Variable, DE = Direct Effect, IE = Indirect
Effect, TE = Total Effect; ***=p < 0.001; **=p < 0.01; *=p < 0.05.
Perceived Usefulness
Behavioral intention was the dependent variable in this study, with an R2 of 0.08, indicating that 8% of the variance of perceived usefulness can be determined based on self-efficacy. The variance performance was 0.283***, which indicates that self-efficacy has an effect on perceived usefulness.
Satisfaction
Satisfaction was the dependent variable in this study, with an R2 of 0.337, indicating that 33.7% of satisfaction variance could be determined based on information quality, system quality, service quality, perceived usefulness, perceived ease of use, and self-efficacy.
Figure 5
Result of the Structural Model
Note
Solid line reports the Standardized Coefficient with * as p < 0.05, and t-value in Parentheses; Dash line reports Not Significant
The second set of data came from graduate students and again had five independent variables, including information quality, system quality, service quality, perceived usefulness, and perceived ease of use. Perceived usefulness is the intermediate variable while satisfaction is the dependent variable. Table 17 summarizes the statistical results for direct impact, indirect impact, and overall impact. Corresponding path analysis data are shown in Fig. 5.
Table 18: Direct, Indirect, and Total Effects of Relationships
Note
IV = Independent Variable, DV = Dependent Variable, DE = Direct Effect, IE = Indirect
Effect, TE = Total Effect; ***=p < 0.001; **=p < 0.01; *=p < 0.05.
Perceived Usefulness
Behavioral intention was the dependent variable in this study, with an R2 of 0.05, indicating that 5% of the variance of perceived usefulness can be determined based on self-efficacy. The variance performance was 0.223***, which indicates that self-efficacy influences perceived usefulness.
Satisfaction
Satisfaction was the dependent variable in this study, with an R2 of 0.402, indicating that 40.2% of satisfaction variance could be determined based on information quality, system quality, service quality, perceived usefulness, perceived ease of use, and self-efficacy.
This has a potential variable in the open. Namely, self-efficacy has a significant effect on satisfaction. Its influence point is 0.021**. Among them, information quality, system quality, service quality, perceived usefulness, and perceived ease of use directly affect satisfaction with effective points of 0.383***, 0.309***, 0.122***,0.095*, and 0.368***, respectively. Corresponding path analysis data are shown in Table 18 and Fig. 6.
Figure 6
Result of the Structural Model
Note
IV = Independent Variable, DV = Dependent Variable, DE = Direct Effect, IE = Indirect Effect, TE = Total Effect; ***=p < 0.001; **=p < 0.01; *=p < 0.05.