4.2. Model Evaluation
The purpose of this investigation was to test the relationship model of the influence of latent variables (e-money behavior and e-money attitude), as assessed through social factors, effort expectancy, and facilitating conditions. These variables were taken from several models, namely, TPB, TAM, and UTAUT, and each factor was measured by valid indicators. The outer model is a formative measurement of latent variable models of the first order. The measurement model needs to be assessed for the reliability and validity of each latent variable. The validity can be assessed by using convergent validity, which describes the level of confidence in the goodness of the measurement of each indicator. Furthermore, the model needs to be assessed by using discriminant validity, which illustrates the differences or discrepancies between indicators in latent variables. Convergent validity was assessed by Average Variance Extracted (AVE) and Composite Reliability (CR). AVE measures the level of construct variation compared with the level of measurement error. AVE values above 0.70 indicate excellent measurements, and AVE values that can be accepted are at least 0.50. CR is a measure of reliability whose value is lower than Cronbach’s alpha; an acceptable CR value is at least 0.70 [
41].
Table 2 shows the AVE, CR, Rho-A, and Cronbach’s alpha for the latent variables (e-money attitude, e-money behavior), effort expectancy, facilitating conditions, and social factors. All latent variables were found to be constructively valid according to Cronbach’s alpha, Rho-A, and composite reliability values, with values above the critical value (0.70). Likewise, with the AVE value, all latent variables were greater than 0.50. Additionally, discriminant validity was also measured using Heterotrait–Monotrait Ratio (HTMT) criteria. Many authors have suggested that a latent variable construct is valid if the HTMT value is below 0.90, and some authors even recommend that it be below 0.85. A HTMT value of 1 indicates that the variable is invalid [
42,
43]. In
Table 3, we can see the HTMT value of e-money attitude with e-money behavior is 0.877, while others are below 0.85, and that of the facilitating conditions variable with social factors is 0.570. If the cut-off is 0.90, it can be said that the latent construct in the model meets the requirements.
Cross-loading was used to detect discriminant validity. An indicator has a higher correlation with itself compared with other variables.
Table 4 shows the cross-loading values. All indicators that use a latent variable indicator in the cross-loading value model of each indicator are greater than the latent variable itself (bolded numbers) compared with other variables (smaller and non-bolded numbers). For example, in the first row, A-Att1 is an indicator measuring the variable e-money attitude, and written in the second column is 0.822, which is greater than the values in the other columns (0.589, 0.686, 0.625, and 0.465). This indicates that A-Att1 is a valid indicator as a measure of the e-money attitude variable compared with its effectiveness as a measure of other variables. For indicators of other variables, the value is greater than the variable itself compared with other variables. This also indicates that the indicators measuring these latent variables are valid.
Table 5 shows the values of R-squared and adjusted R-squared, which describe the ability of the social factors, facilitating conditions, and effort expectancy to explain the e-money attitude and e-money behavior variables. For the e-money attitude, R-squared is 0.603, which means that the e-money attitude variable is explained by the two independent variables at 60.3%, and the rest (39.7%) is influenced by other variables. The R-squared of e-money behavior variables is 0.611, which means that the influence of social factors and effort expectancy in e-money attitude on e-money behavior is 61.1%, while 39.9% is influenced by other variables.
Multicollinearity occurs when two or more independent variables in a model correlate, resulting in redundant information and responses. Multicollinearity is measured by variance inflation factors (VIFs) and tolerance. If the VIF value exceeds 4.0, or if it has a tolerance of less than 0.2, this indicates that there is a multicollinearity problem in the model.
Table 6 shows that the values of VIFs for the independent variables on e-money attitude and e-money behavior are smaller than 0.4. This indicates that the tested model is free of multicollinearity problems [
41].
Figure 3 and
Figure 4 show the structural and measurement models. The measurement model shows the validity of the construct of latent variables composed of valid indicators, where the T-statistic value is greater than the critical value (1.96), and the loading value is greater than 0.60, indicating that all the construct indicators are valid. In the structural model (
Table 7), which describes the path of the relationship between the latent variables, the T-statistic values from 2.591 to 4.758 are greater than T-critical (1.96) at a significance of 5%, except for the social factor pathway to e-money behavior, for which the T-statistic is only 1.861, significant at the 10% level. The coefficients of all paths in the inner model (original sample) range from 0.144 to 0.483, with a standard deviation from 0.078 to 0.101. The coefficient indicates the magnitude of the effect of latent variables on other latent variables. All coefficients are positive, which means that the relationship between these variables is unidirectional. If the independent latent variable changes, then the dependent latent variable will increase. For example, the e-money attitude to e-money behavior path coefficient is 0.483, reflecting the magnitude of the change that will occur if the e-money attitude changes. The interpretation of the meaning of changes in variables depends on the measurement and scale used. Not every change can be interpreted quantitatively.
The basic framework of the model includes one or several TPB, TAM, and UTAUT domains. The focus of the model is the integration of the three models lying in the domain that is positioned at the far right of the model, namely, e-money behavioral intention. Within various associated study contexts, the model has been widely applied, either portraying intention as a mediator or without mediation. The results show that behavior towards an object is consistently predicted by attitude. Likewise, the results of data analysis show that the intentions and behavior regarding the use of e-money are significantly predicted by attitudes towards e-money. These results support previous research applied in a variety of contexts. For example, in the case of online shopping, it was stated that shopping behavior online or through the internet was influenced by a person’s attitude towards the shopping system [
44,
45] in the context of mobile banking adoption [
46], use of non-cash systems [
27], and adoption of smart home technology [
47]. More specifically, it supports research carried out in the context of the use of electronic payments, mobile payments, the use of e-money, and similar topics [
48,
49].
More specifically, the e-money behavioral intention domain is a construct that is measured through the following indicators: application installation, continuity of use, plan to use in the short term going forward, familiarization, the user’s intentions to not reinstall, immediately top up, and recommend that others use e-money for payment transactions, as well as whether it is the main means of payment when making transactions. The composite intention and behavior are explained by the subjects’ attitudes towards e-money, as measured by the following indicators: good ideas, fun, knowledgeable, modern, and top class. This means that customers intend to use and actually use e-money in transactions because of their attitude about it. For example, customers will continue to use and get used to e-money for payment transactions because they feel that using e-money is a good idea, modern, not old-fashioned, and fun. This indicates that the attitude towards a particular object is a prediction of a person’s behavior related to the object, consistent with the TPB framework and the previous research that forms the basis of this paper.
Besides being influenced by attitude, one’s behavior in using e-money is significantly influenced by effort expectancy and social factors. The two predictors are domains commensurate with social influences taken from the UTAUT framework [
50] and subjective norms in the TPB framework [
51]. This indicates that one or several domains from an established and widely applied basic framework, in this case, TPB and UTAUT, still provide consistent results. Analysis in the context of the use of e-money as a means of payment transactions consistently supports previous research results, i.e., that social factors and effort expectancy are predictors of behavior [
26,
27,
44,
47,
48,
52]. This means that customers will continue to use e-money, make it the first choice of payment, and keep the installed e-money application on smart devices because of external persuasion such as shops that they visit and close friends or family as social factors.
Analysis based on available data shows statistically significant results, indicating that the behavior and intention to use e-money are positively influenced by the existence of outsiders who provide useful and practical assessments of their attributes in a meaningful way. The significance of these findings supports previous research that applied TPB and UTAUT as a whole in the context of non-cash transaction payments [
26,
27,
46,
53]. Attitudes towards e-money are the closest explanatory variable in the hypothesis model tested in this paper. The attitude domain is taken from the TPB framework [
51,
54,
55] as an explanation of intentions of behavior. As an explanation, the formation of attitude is influenced by external factors, namely, factors external to the performer. In this study, the attitude that is being optimized is influenced by social factors and facilitating conditions. External factors are parties, close people, or sellers that exist externally to the customer, while facilitating conditions are more focused on the available infrastructure that allows customers to make transactions using e-money. This concept is taken from the UTAUT framework, whereas the TPB framework is more directed towards behavior control. Some valid measures of these variables include the availability of facilities at merchants, internet connection support, adequate smartphone, user skills, financial institution support, and the possibility of payment default. The analysis shows that customer attitudes towards e-money are significantly predicted by these two domains. That is, positive customer attitudes related to the measures used are caused by the condition of infrastructure as a support and also by the persuasion of social factors. The existence of internet connection, support by adequate devices, and user skills, coupled with the encouragement of external parties such as shops, financial institutions, and the people closest to the individuals, will make users of e-money have a positive attitude. The positive attitude is reflected by feeling happy, feeling unworried about personal data being misused, and feeling up to date. This finding does not contradict and instead strengthens previous studies that tried to take part of the domain or apply the TPB and UTAUT approaches completely in various contexts, for example, the context of payment transactions that do not use cash [
50,
55,
56,
57,
58].