Table 4.12: Bank type
Bank type
Frequency | Rate (%) | |
A 100% electronic bank (operating solely on | 126 | 24.7 |
on Website and no branches) | ||
A traditional bank operates based on | 385 | 75.3 |
branch network and also provides | ||
electronic banking services | ||
Total | 511 | 100.0 |
Maybe you are interested!
-
Completing the personal credit rating model at Saigon Commercial Joint Stock Bank - 12 -
Building a Scale and Research Model of Factors Affecting Customers' Decision to Choose a Bank to Deposit Savings at -
Research Model and Research Methodology to Analyze the Impact of Bad Debt on Bank Efficiency -
Anova Test of Regression Model Fit -
Correlation Coefficient Results and Significance Level of Correlation Coefficient Test for Joint Stock Commercial Bank Group
Source: Calculated from survey data
4.3. Model testing results
4.3.1. Measurement model
The measurement model aims to assess the reliability of the scale. To assess the reliability of the scale, the study uses the Cronchbach's Alpha coefficient, the most common indicator to assess the reliability of scales. The Cronchbach's Alpha coefficient (α) is calculated as follows (Cronbach, 1951)[35]:
a S 2
N 2 COV
item
item
COV
In which: N is the number of observed variables, Cov: correlation between variables, s 2 : variance of variables. Cov with a dash symbol above is the average Cov value. The symbol ∑ reflects the summation.
The scale is considered reliable when the Cronchbach's Alpha coefficient is greater than or equal to 0.6 but preferably greater than 0.7 (Nunnally and Burnstein, 1994)[129].
In addition, to assess the reliability of the scale (observed variables), the composite reliability coefficient (ρc) of each latent variable can be calculated. To calculate this coefficient value, information about the indicator loading and error variance is used (Diamantopoulos and Siguaw, 2007)[63] according to the formula:
ρ c = (∑λ) 2 / [(∑λ) 2 + ∑(θ)]
In which: ρ c = composite reliability coefficient
λ = indicator loadings
θ = indicator error variances
If the value of ρ c > 0.6 we can conclude that the observed variables provide a scale.
reliable measure of the latent variable (Diamantopoulos and Siguaw, 2007)[63].
Another measure used to assess the reliability of a scale is the average variance extracted (AVE - ρ v ). This measure reflects the amount of variance captured by the latent variable relative to the variance due to scale error. ρ v
<0.5 indicates that the scale error accounts for a larger amount of variance in the indicators than in the latent variables (and thus calls into question the validity of the indicators/or latent variables) (Diamantopoulos and Siguaw, 2007)[63].
ρ v = (∑λ 2 ) / [∑λ 2 + ∑(θ)]
Where λ, θ, are defined above.
The table below summarizes information on the composite reliability coefficient (ρ c ), Average extracted variance (ρ v ) and Cronbach's alpha coefficient (α) of all 8 latent variables calculated using Mplus 6.11 and SPSS 18 software.
Through the calculated data, it shows that the Cronbach's alpha coefficient of all variables is greater than 0.8, so the scale is good, ensuring consistency (Nunnally and Burnstein, 1994)[129].
The composite reliability coefficients (ρ c ) were all greater than 0.8, reflecting that the scales were reliable (Fornell and Larcker, 1981[75]; Diamantopoulos and Siguaw, 2007[63]).
The average variance extracted (AVE - ρ v ) of the latent variables are all larger than
0.5 (Diamantopoulos and Siguaw, 2007)[63].
Thus, we can conclude that, through the measurement model, the observed variables are reliable enough and represent the latent variables well. Therefore, these variables can be used for analysis in the structural model.
Table 4.13: Statistical indicators of the measurement model
Latent variable
Observation variable | Load factor | Variance of error of measurement measure | Average variance (ρ v ) | Cronbach's alpha coefficient (α) | Composite reliability coefficient (ρ c ) | |
EB Service Quality (F1) | Ttb | 0.717 | 0.486 | 0.631 | 0.870 | 0.872 |
Rtb | 0.819 | 0.329 | ||||
Restb | 0.788 | 0.379 | ||||
Etb | 0.848 | 0.281 | ||||
Information system quality online (F2) | EUTB | 0.847 | 0.283 | 0.662 | 0.854 | 0.854 |
Atb | 0.791 | 0.374 | ||||
Stb | 0.801 | 0.357 | ||||
Quality of banking products and services (F3) | BSP_1 | 0.807 | 0.348 | 0.634 | 0.891 | 0.896 |
BSP_2 | 0.829 | 0.313 | ||||
BSP_3 | 0.838 | 0.297 | ||||
BSP_4 | 0.819 | 0.330 | ||||
BSP_5 | 0.677 | 0.542 | ||||
Quality EB overall (F4) | O_1 | 0.847 | 0.282 | 0.749 | 0.856 | 0.856 |
O_2 | 0.883 | 0.221 | ||||
Customer satisfaction (F5) | CS_1 | 0.812 | 0.341 | 0.669 | 0.908 | 0.890 |
CS_2 | 0.804 | 0.354 | ||||
CS_3 | 0.848 | 0.281 | ||||
CS_4 | 0.806 | 0.350 | ||||
Loyalty Customer (F6) | L_1 | 0.831 | 0.309 | 0.706 | 0.845 | 0.878 |
L_2 | 0.844 | 0.287 | ||||
L_3 | 0.846 | 0.285 | ||||
Shipping costs change (F7) | SC_1 | 0.767 | 0.411 | 0.511 | 0.879 | 0.837 |
SC_2 | 0.786 | 0.382 | ||||
SC_3 | 0.788 | 0.379 | ||||
SC_4 | 0.630 | 0.603 | ||||
SC_5 | 0.573 | 0.672 | ||||
Customer Trust (F8) | Tr_1 | 0.89 | 0.208 | 0.678 | 0.861 | 0.862 |
Tr_2 | 0.878 | 0.229 | ||||
Tr_3 | 0.687 | 0.528 |
Source: Calculated from survey data
4.3.2. Structural Equation Model (SEM)
4.3.2.1. Choosing the right model
After running the measurement model, the results showed that the scales were reliable enough to run the structural model with the goal of testing hypotheses about the relationships. Based on the hypotheses and research models built in the overview, to test the hypotheses of the model, the author built 3 structural models using the variable addition method:
Model 1 (Basic Model):
Customer Service Quality
online
γ1
Quality of information system
online
γ2
Overall Service Quality
Electronic Bank
γ4
The satisfaction of
client
γ5
Loyalty
of the customer
γ3
Quality
SPDVNH
Model 1 consists of 6 latent variables: e-banking service quality (F1), online information system quality (F2), banking product and service quality (F3), overall e-banking service quality (F4), customer satisfaction (F5) and customer loyalty (F6). Model 1 aims to test hypotheses H1, H2, H3, H4, H5.
Figure 4.1: Structural model 1
Model 2 (Adding the mediating variable Switching Cost)
In addition to the 6 latent variables in model 1, model 2 adds the latent variable switching cost (F6) to further test the hypothesis:
H6: Switching costs have an impact on the relationship between customer satisfaction and loyalty.
Customer Service Quality
online
γ1
Quality of information system
online
γ2
Overall Service Quality
Electronic Bank
γ4
The satisfaction of
client
γ5
Loyalty
of the customer
γ6
γ3
Quality
SPDVNH
Conversion costs
Figure 4.2: Structural model 2
Model 3 (Adding 2 mediating variables Switching Cost and Customer Trust)
Model 3, in addition to the 6 latent variables mentioned in model 1, also adds 2 more latent variables: switching costs (F6) and customer trust (F7) to test 2 more hypotheses:
H6: Switching costs have an impact on the relationship between customer satisfaction and loyalty.
H7: Customer trust has an impact on the relationship between customer satisfaction and customer loyalty.
Customer Service Quality
online
The trust of
client
γ1
γ6
Quality of information system
online
γ2
Quality
total electronic banking services
γ4
The satisfaction of
client
γ5
Loyalty
of the customer
γ7
γ3
Quality
SPDVNH
Conversion costs
Figure 4.3: Structural model 3
After running the above 3 structural models, synthesizing the results, the selection of the most suitable model will be carried out.
Model testing results 1
After running model 1 using Mplus 6.11 software, the results obtained are as follows:
Number of observed variables
511 | |
Number of dependent variables | 21 |
Number of independent variables | 0 |
Number of latent variables | 6 |
Table 4.14: Summary of model fit 1
Number of degrees of freedom
71 | ||
Neighboring values in Logarithmic units | H0 value H1 value | -13228.985 -12919.946 |
Values of AIC, BIC and adjusted BIC coefficients | AIC coefficient BIC coefficient Adjusted BIC coefficient | 26599.971 26900.753 26675.389 |
The Chi-square value assesses the goodness of fit of the model. image | Degrees of Freedom Value p-value | 618,078 181 0.0000 |
RMSEA coefficient | Estimate | 0.069 |
90% confidence interval | 0.063 0.075 | |
RMSEA coefficient | 0.000 | |
CFI and TLI coefficients | CFI coefficient | 0.948 |
TLI coefficient | 0.940 | |
Value When Squared Hit | Value | 8689.220 |
price of model suitability | Degrees of freedom | 210 |
software background image | p-value | 0.0000 |
build | ||
SRMR coefficient | Value | 0.037 |
Source: Calculated from survey data
The results of running model 1 show that this model has degrees of freedom df (Degrees of Freedom) = 181>0.
The RAMSEA (Root Mean Square Error Of Approximation) coefficient is 0.069 <0.08, indicating that the model fit is good (Taylor, Sharland, Cronin and Bullard, 1993[138]; Diamantopoulos and Siguaw, 2007[63]).
SRMR (Standardized Root Mean Square Residual) coefficient = 0.037<0.05 shows that the model fits well (Taylor, Sharland, Cronin and Bullard, 1993)[157].
The CFI coefficients = 0.948>0.9; TLI = 0.940>0.9 show that the model fits well (Segar & Grover, 1993[131]; Chin & Todd, 1995[51]).
The Chi-squared/Degrees of Freedom ratio (χ 2 /df) = 618.078/181 = 3.4 <5 shows that the model
The model fits well because the sample size is 511 >200 (Kettingger & Lee,1995[103])
For model 2 and model 3, because in these 2 models there is a test for the interaction
the dynamics of the mediating variables (switching costs (F7) and trust (F8)) so the author
Use Mplus software because it is the simplest and most effective software among the software that can run SEM.
The results from running two alternative models, model 2 and model 3, are summarized in the table below (with comparison with model 1).
Through the summary table of information on the level of fit and correlation of the 3 models, it is easy to see that model 1 is the best model. Model 1 has coefficients AIC = 26599.971, BIC = 26900.753, ABIC = 26675.389 while model 2 has AIC = 34351.168, BIC = 34732.44, ABIC = 34446.768, model 3 has AIC =
38650.728, BIC = 39091.311, ABIC = 38761.199. So model 1 is the best because it has the smallest AIC, BIC and ABIC, then model 2 and then model 3.
Furthermore, when considering the correlation coefficient (γ), model 1 also has the most statistically significant correlations (equal to model 2 and more than model 3).
Table 4.15: Model Fit and Correlation Coefficients
Information on suitability
Model 1 | Model 2 | Model 3 | |
Number of degrees of freedom | 71 | 90 | 104 |
Logarithmic Proximity Value H0 Value H1 value | -13228.985 -12919.946 | -17085.584 1,503 | -19221.364 1,492 |
AIC coefficient BIC coefficient Adjusted BIC coefficient | 26599.971 26900.753 26675.389 | 34351.168 34732.441 34446.768 | 38650.728 39091.311 38761.199 |
F1 (γ1) | -0.030 | -0.047 | -3.423 |
F2 (γ2) | 0.537* | 0.669* | 4,742 |
F3 (γ3) | 0.469** | 0.436** | -0.159 |
F4 (γ4) | 0.929** | 0.956** | 1,009** |
F5 (γ5) | 0.879** | 0.822** | 0.899** |
F7xF5 (γ6) | -0.044 | 0.022 | |
F8xF5 (γ7) | -0.042 | ||
Note:F1: Quality of KHĐT services; F2: Quality of online information system; F3: Quality of NH products and services; F4: Overall quality of NHĐT services; F5: Customer satisfaction; F6: Customer loyalty. γ is the coefficient of relationship between latent variables. * Significant at p value < 0.05 ** Significant at p value < 0.01 | |||
Source: Calculated from survey data