Data Analysis

Descriptive Statistics

In this study, the results of the descriptive statistics of the survey are presented in Table 3.


Table 3. Descriptive statistics.

n Minimum Maximum Mean Standard Deviation Variance
Statistic Standard Error
Q4 390 1 5 3.22 0.056 1.039 1.079
Q5 390 1 5 3.10 0.060 1.114 1.240
Q6 390 1 5 3.09 0.060 1.110 1.233
Q8 390 1 5 3.82 0.042 0.786 0.617
Q9 390 1 5 3.55 0.045 0.842 0.710
Q10 390 1 5 3.68 0.043 0.808 0.652
Q11 390 1 5 3.64 0.042 0.783 0.613
Q13 390 1 5 3.66 0.042 0.790 0.624
Q14 390 1 5 3.56 0.042 0.792 0.627
Q15 390 1 5 3.65 0.042 0.783 0.614
Q18 390 1 5 3.15 0.048 0.893 0.798
Q19 390 1 5 2.65 0.050 0.934 0.873
Q20 390 1 5 2.57 0.052 0.965 0.931
Q25 390 1 5 3.03 0.054 1.001 1.002
Q28 390 1 5 2.98 0.054 1.003 1.005
Q29 390 1 5 3.18 0.052 0.972 0.944
Q33 390 1 5 3.35 0.043 0.799 0.638

Table 4 presents the results that explain the characteristics of the respondents who participated in the survey. Male (49.5%) and female (50.5%) respondents participated in this survey in an equal ratio, and consumers who repurchased smartphones 2–5 times in the last two years accounted for more than 80% of all respondents.

Table 4. Analysis of survey respondents.

Gender
Frequency Percentage Valid Percentage Cumulative Percentage
Man 193 49.5 49.5 49.5
Woman 197 50.5 50.5 100.0
Total 390 100.0 100.0
Age
Frequency Percentage Valid Percentage Cumulative Percentage
20 s 112 28.7 28.7 28.7
30 s 177 45.4 45.4 74.1
40 s 52 13.3 13.3 87.4
50 s 49 12.6 12.6 100.0
Total 390 100.0 100.0
Number of Smartphone Repurchases
Number of smartphone repurchases Frequency Percentage Valid Percentage Cumulative Percentage
2 29 7.5 7.5 7.5
3 91 23.3 23.3 30.7
4 90 23.0 23.0 53.7
5 105 27.0 27.0 80.7
6 33 8.3 8.3 89.1
7 9 2.3 2.3 91.4
8 6 1.4 1.4 92.8
9 1 0.3 0.3 93.1
10 20 5.2 5.2 98.3
12 1 0.3 0.3 98.6
15 3 0.9 0.9 99.4
16 1 0.3 0.3 99.7
17 1 0.3 0.3 100.0
Total 390 100.0 100.0

Factor Analysis

In this study, factor analysis was based on the collected data. For factor analysis, maximum likelihood was used as the factor extraction method, and oblimin with Kaiser normalization was used as a factor rotation method. In addition, factor analysis indicated that 17 observed variables could be clustered into five latent variables. The results of the factor analysis in this study were validated through the Kaiser–Meyer–Olkin (KMO) test and Bartlett's test. The result of the KMO test was 0.889, which suggests that the factor analysis was appropriate (Table 5).

Table 5. Kaiser–Meyer–Olkin (KMO) and Bartlett's tests.

Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.889
Bartlett's Test of Sphericity Approximate chi-square 5406.133
df 528
Sig. 0.000

Based on consumer data, the results of factor analysis included survey results grouped into five factors, and the reliability of the elements that form each factor was determined to be excellent. The results appear in Table 6.

Table 6. Factor analysis.

Factor Cronbach's Alpha
1 2 3 4 6
Customer Satisfaction Q11 0.905 0.896
Q10 0.883
Q9 0.831
Q8 0.689
Social Influence Q6 0.871 0.784
Q5 0.716
Q4 0.618
Habit Q29 0.709 0.708
Q28 0.661
Q25 0.612
Emotional Loyalty Q20 0.817 0.8
Q19 0.745
Q18 0.670
Intention to Repurchase Q14 0.808 0.848
Q15 0.773
Q13 0.715
Q33 0.604

Correlation Analysis

This study analyzed the directionality of the factors through correlation analysis between the derived factors, as shown in Table 7.

Table 7. Correlation analysis.

Social Influence Emotional Loyalty Intention to Repurchase Customer Satisfaction
Social Influence Pearson correlation 1 0.327 ** 0.196 ** 0.182 **
Sig. (2-tailed) 0.000 0.002 0.001
n 390 390 390 390
Emotional Loyalty Pearson correlation 0.327 ** 1 0.515 ** 0.397 **
Sig. (2-tailed) 0.000 0.000 0.000
n 390 390 390 390
Intention to Repurchase Pearson correlation 0.169 ** 0.467 ** 1 0.728 **
Sig. (2-tailed) 0.002 0.000 0.000
n 390 390 390 390
Customer Satisfaction Pearson correlation 0.182 ** 0.397 ** 0.728 ** 1
Sig. (2-tailed) 0.001 0.000 0.000
n 390 390 390 390

Regression Analysis

Regression analysis was performed to analyze the linear causality between several independent and dependent variables in this study. The results are shown in Table 8, Table 9 and Table 10. The analysis was based on the stepwise method and was analyzed using SPSS 24.0.

Table 8. Regression analysis.

Model Variables Entered Variables Removed Method
1 Customer satisfaction Stepwise (criteria: probability of F to enter ≤ 0.050, probability of F to remove ≥ 0.100).
2 Emotional loyalty
Dependent variable: intention to repurchase.

Table 9. Model summary.

Model R R Square Adjusted R Square Standard Error of the Estimate Change Statistics Durbin–Watson
R Square Change F Change df1 df2 Sig. F Change
1 0.728 a 0.530 0.529 0.47442 0.530 390.500 1 346 0.000
2 0.753 b 0.568 0.565 0.45575 0.038 29.934 1 345 0.000 1.869
a Predictors: (constant), customer satisfaction; b predictors: (constant), customer satisfaction, emotional loyalty.

Table 10. Coefficients a.

Model Unstandardized Coefficients Standardized Coefficients t Sig.
B Standard Error Beta
1 (Constant) 0.996 0.135 7.360 0.000
Customer satisfaction 0.716 0.036 0.728 19.761 0.000
2 (Constant) 0.780 0.136 5.743 0.000
Customer satisfaction 0.634 0.038 0.645 16.715 0.000
Emotional loyalty 0.185 0.034 0.211 5.471 0.000
a Dependent variable: intention to repurchase.

According to the results in Table 9, the second research model was analyzed to be highly complete, and the regression analysis results of this model are presented in Table 9.

Table 10 shows the results of analyzing the coefficients of the regression analysis model variables.


Analysis of Research Model Using ANN (Relative 7:3, Number of Hidden Layers (One))

This study enhanced the analysis of the research model by using the ANN algorithm. To this end, with one hidden layer, two cases were analyzed separately. Table 11, Table 12, Table 13, Table 14 and Table 15 are the results of analyzing the research model assuming one hidden layer. With the ANN algorithm, this study advanced the analysis of the research model.

Table 11. Case processing summary.

n Percentage
Sample Training 283 72.75%
Testing 106 27.25%
Valid 389 100.0%
Excluded 1
Total 390

Table 12. Network information. MLP - multilayer perceptron.

Input Layer Factors 1 Customer satisfaction
2 Habit
3 Social influence
4 Emotional loyalty
Number of units 51
Hidden Layer(s) Number of hidden layers 1
Number of units in hidden layer 1a 8
Activation function Sigmoid
Output Layer(s) Dependent variables 1 Predicted value for MLP predicted value
Number of units 6
Activation function Softmax
Error function Cross-entropy

Table 13. Model summary a.

Training Cross-entropy error 13.233
Percentage incorrect predictions 0.0%
Stopping rule used 1 consecutive step(s) with no decrease in error a
Testing Cross-entropy error 50.596
Percentage incorrect predictions 7.4%
Dependent variable: predicted value for intention to repurchase; a error computations are based on the testing sample.

Table 14. Regression analysis.

Model Variables Entered Variables Removed Method
1 Customer satisfaction Stepwise (criteria: probability of F to enter ≤ 0.050, probability of F to remove ≥ 0.100).
2 Emotional loyalty
3 Social influence
Dependent variable: predicted value for intention to repurchase.

Table 15. Model summary.

Model R R Square Adjusted R Square Standard Error of the Estimate Change Statistics Durbin–Watson
R Square Change F Change df1 df2 Sig. F Change
1 0.814 a 0.663 0.663 0.34972 0.663 1429.476 1 726 0.000
2 0.831 b 0.690 0.690 0.33463 0.028 65.913 1 725 0.000
3 0.835 c 0.696 0.696 0.33155 0.006 14.557 1 724 0.000 1.869
a Predictors: (constant), customer satisfaction; b predictors: (constant), customer satisfaction, emotional loyalty; c predictors: (constant), customer satisfaction, emotional loyalty, social influence.

Table 12 presents material that explains the method of analyzing research models with ANN. In the study presented here, a research model consisting of four independent variables and one dependent variable was subjected to multiple regression analysis. In addition, the sigmoid function, which combines the node and weight of the hidden layer when transmitted from raw data to the hidden layer, was used as an active function in ANN. Beyond that, the softmax function was selected as an activation function for calculating the results from the hidden layer to the output layer.

In this research, when the number of hidden layers was set as one under the ANN algorithm, the third research model was found to be the most complete model.

Model 3 (Table 16) determined that consumer satisfaction (0.621), emotional loyalty (0.125), and social influence (0.063) influenced intention to repurchase, unlike the other models. In addition, it determined that consumer satisfaction had the greatest influence on intention to repurchase.

Table 16. Coefficients a.

Model Unstandardized Coefficients Standardized Coefficients t Sig. Variance Inflation Factors
(VIF)
B Standard Error Beta
1 (Constant) 1.128 0.070 16.190 0.000
Customer satisfaction 0.698 0.018 0.814 37.808 0.000 1.000
2 (Constant) 0.892 0.073 12.260 0.000
Customer satisfaction 0.636 0.019 0.742 32.982 0.000 1.188
Emotional loyalty 0.145 0.018 0.183 8.119 0.000 1.188
3 (Constant) 0.805 0.076 10.646 0.000
Customer satisfaction 0.621 0.020 0.724 31.832 0.000 1.238
Emotional loyalty 0.125 0.018 0.158 6.785 0.000 1.290
Social influence 0.063 0.017 0.086 3.815 0.000 1.211
a Dependent variable: predicted value for intention to repurchase.

Analysis of Research Model Using ANN (Relative 7:3, Number of Hidden Layers (Two))

Table 17, Table 18, Table 19, Table 20, Table 21 and Table 22 are the results of analyzing the research model by assuming two hidden layers in the ANN algorithm.

Table 17. Case processing summary.

n Percentage
Sample Training 283 72.75%
Testing 106 27.25%
Valid 369 389
Excluded 1 1
Total 370 390

Table 18. Network information. 

Input Layer Factors 1 Customer satisfaction
2 Habit
3 Social influence
4 Emotional loyalty
Number of units 48
Hidden Layer(s) Number of hidden layers 2
Number of units in hidden layer 1 a 9
Number of units in hidden layer 2 a 7
Activation function Sigmoid
Output Layer(s) Dependent variables 1 Predicted value for MLP predicted value
Number of units 5
Activation function Softmax
Error function Cross-entropy
a Excluding the bias unit.

Table 19. Model summary.

Training Cross-entropy error 13.272
Percentage incorrect predictions 0.0%
Stopping rule used 1 consecutive step(s) with no decrease in error a
Testing Cross-entropy error 30.578
Percentage incorrect predictions 6.8%
Dependent variable: predicted value for intention to repurchase; a error computations are based on the testing sample.

Table 20. Regression analysis.

Model Variables Entered Variables Removed Method
1 Customer satisfaction Stepwise (criteria: probability of F to enter ≤ 0.050, probability of F to remove ≥ 0.100).
2 Emotional loyalty
3 Social influence
Dependent variable: predicted value for intention to repurchase.

Table 21. Model summary.

Model R R Square Adjusted R Square Standard Error of the Estimate Change Statistics Durbin–Watson
R Square Change F Change df1 df2 Sig. F Change
1 0.843 a 0.710 0.710 0.32800 0.710 1754.999 1 726 0.000
2 0.855 b 0.731 0.731 0.31626 0.021 55.218 1 725 0.000
3 0.859 c 0.737 0.737 0.31310 0.006 15.515 1 724 0.000 1.869
a Predictors: (constant), customer satisfaction; b predictors: (constant), customer satisfaction, emotional loyalty; c predictors: (constant), satisfaction, emotional loyalty, social influence.

Table 22. Coefficients a.

Model Unstandardized Coefficients Standardized Coefficients t Sig.
B Standard Error Beta
1 (Constant) 0.836 0.071 11.801 0.000
Customer satisfaction 0.783 0.019 0.843 41.893 0.000
2 (Constant) 0.646 0.073 8.868 0.000
Customer satisfaction 0.724 0.020 0.779 36.741 0.000
Emotional loyalty 0.128 0.017 0.158 7.431 0.000
3 (Constant) 0.558 0.076 7.380 0.000
Customer satisfaction 0.710 0.020 0.764 35.834 0.000
Emotional loyalty 0.108 0.018 0.133 6.070 0.000
Social influence 0.062 0.016 0.083 3.939 0.000
a Dependent variable: predicted value for intention to repurchase.

The analysis results of Table 21 show that the third research model was the most complete when the number of hidden layers was set to two under the ANN algorithm.

Model 3 (Table 22) determined that consumer satisfaction (0.710), emotional loyalty (0.108), and social influence (0.062) influenced repurchase intention, unlike the other models. In addition, it determined that consumer satisfaction had the greatest influence on repurchase intention.