3.2.2. Regression Analysis

The most primitive form of regression analysis is the method of least squares, which was first used scientifically in 1885 [74]. Moreover, the method was developed as the method of least squares procedures [75]. Regression analysis is defined as an analytical method to understand the relationship between a dependent variable and independent variables [76]. In addition, it can be used to estimate and predict the expected value of a dependent variable in a conceptual model. After performing regression analysis with the Minitab 19 software, we obtain results including the beta coefficient (*β*), *p*-value, t-value, and R2. Next, we can use these to evaluate whether or not the hypotheses should be accepted or rejected. Adjusted R2 indicates how independent variables influence the dependent variable. In theory, if it is good (over 50%), the study is well justified [77]. In addition, *p*-value is the probability that the null hypothesis is supported. At large, a variable with a *p*-value not over 0.05 shows that the variable, within the model, is significant; while a *p*-value exceeding 0.05 implies that the variable should be removed [77]. Regarding the beta coefficient (*β*), which independent factor has the largest beta coefficient also has the greatest influence on the change in the dependent variable.

#### 3.2.3. Linear Structural Relations (LISREL)

LISREL is a statistics software for modeling structural equations [78] and was used as a research tool in 1972 [79]. For this research, LISREL 8.8 software was used. The reason for applying it is that we can test the hypotheses in the proposed model easily and flexibly. Moreover, we can use this software in various fields thanks to the ability of this software to easily adapt according to the situation, especially in constructing models for estimating relationships between independent and dependent variables. The third reason is providing crucial indicators such as χ2/df, P, RMSEA, AGFI, PGFI, GFI, NFI, CFI, IFI, and SRMR, LISREL also helps us determine the goodness of fit test [80]. Importantly, the results from running LISREL 8.8 also show the model's structural validity, and the *t*-test helps determine any significant relationship between the two variables. In theory, NFI should be over 0.9 [81], GFI and CFI higher than 0.9 [82].

#### **4. Results**

#### *4.1. Demographic Statistics*

Questionnaires were distributed over two months from 3 September 2021, to 19 November 2021, via an online survey—Google Forms in Vietnamese, Indonesian, Chinese, and English. According to previous research, monetary incentives are beneficial to enhance the response rate and completion rate [83]. Therefore, the authors used a lottery-style approach, which means we provided a USD 25 gift to the luckiest respondent by a random draw. Overall, we attained 616 responses. However, 52 participants were not qualified and 564 complied with the requirements of the survey. Table 1 depicts the demographic profile of the collected data. There are 186 Vietnamese, 189 Indonesian, and 189 Taiwanese respondents, which accounts for about 33% of each region. According to previous studies, urban vs rural differences are prone to obscure sophisticated influences [84]; measure of urban vs rural characteristics is considered in the research. The percentage of urban participants (78.01%; 440 samples) is four times higher than that for rural counterparts—21.99%; 124 samples. Moreover, 60.82% of the total participants are females while 39.18% are males. Only 11.88% are high school students, the rest completed higher education. Maritalwise, the number of single participants is dominant, with 436 samples (77.30%), while married individuals occupy 22.16% (125 samples). Regarding age, the proportion of individuals from 18 to 35 years old constitutes 82.62% (466 samples). The data show that respondents with the lowest income (less than USD 250 per month) take up the highest rate (37.23%; 210 samples), while those making the highest income (over USD 5000 per month) make up the smallest proportion (0.89%; 5 samples).


**Table 1.** Demographic statistics (*n* = 564).


**Table 1.** *Cont.*

Based on the database collected, Cronbach's alpha values were utilized to test the reliability of research constructs. Table 2 shows that all Cronbach's alpha indicators of the pooled data and each dataset surpass the generally agreed threshold of 0.8 [85]. Therefore, the research is reliable for each location and the cross locations.



#### *4.2. Regression Analysis and LISREL Testing*

#### 4.2.1. Regression Analysis

Based on the database collected, regression analysis was carried out via Minitab 19 software. According to the proposed models, we performed a regression analysis based on Equations (1)–(3) with four datasets. The results demonstrated in Table 3 show that, in three regression models for Equations (1)–(3) with four datasets, R<sup>2</sup> and adjusted R2, are over 0.05, which means all models are well justified. Moreover, we observed that indexes such as R2 and the adjusted R<sup>2</sup> of model 1 are higher than those of models 2 and 3 in the four datasets. Therefore, model 1 is more justified than models 2 and 3.




#### **Table 3.** *Cont.*

Note: \* indicates *p*-value ≤ 0.05.

Model 1 relating to Equation (1) is built as follows:


Model 2 relating to Equation (2) is built as follows:


Model 3 relating to Equation (3) is constructed as follows:


Table 4 shows the results of hypotheses testing. For the pooled data, all hypotheses are supported except H6 (*β* = −0.13 < 0). For each regional data, we find that H3, H7, and H9 are supported, while H6 is rejected in all three areas. Both Indonesian and Taiwanese data support H1, H2, and H8 while those hypotheses are rejected in Vietnam. However, regarding H4, Vietnamese data supports it (*β* = 0.62 > 0 and *p*-value ≤ 0.05) whereas both Indonesian and Taiwanese information rejected it. In contrast, H5 is supported with Indonesian data (*β* = 023 > 0 and *p*-value ≤ 0.05), while it is rejected in both Vietnam and Taiwan.

**Table 4.** Hypotheses testing—regression.



Note: \* indicates *p*-value ≤ 0.05. Supported: yes (*β* > 0 and *p*-value ≤ 0.05).

#### 4.2.2. LISREL Testing
