Regression analysis

The multiple regression analyzing is performed to review the effect of each independent variable: the responsiveness, the assurance, the tangibles, the reliability and the trust on the satisfaction of customers (dependent variable).

The model of multiple linear regression expresses the satisfaction as below:

Satisfaction = B₀ + B₁ * trust + B₂ * responsiveness + B₃ * assurance + B₄ * reliability + B₅ * tangibles.

Of which, B1, B2, B3, B4, and B5 are partial regression coefficients.

The trust, the responsiveness, the assurance, the tangibles and the reliability are independent variables and the satisfaction is dependent variable.

The regression analysis is implemented by the method of stepwise selection.

This method just in turn adds one independent variable to the model. Independent variables which have the strongest positive or negative correlation with the dependent variable will be applied to the equation firstly. If the variable does not meet the conditions, the process will be stopped and there will be no more independent variables in the equation. If it meets the input conditions, the next (second) independent variable will be added in. It will be the one which explains the most the changing level of the dependent variable in combination with the first variable. The analysis is just kept going on like that. After putting the first variable, the computer will consider if to delete it or not according to the exit standard. In the next step, variables which are not in the equation will be considered to input. After each step,

30

variables in the equation will be considered again to delete. Variables will be removed until there will be no more variables which can achieve the entry and exit standard.

(Hoang Trong and Chu Nguyen Mong Ngoc, 2005; Loan Le, 2000).

The result of the regression analysis by the Stepwise according to the entry standard with PIN=0.05 and the exit standard with POUT=0.10 shows that:

Independent variable of the assurance does not meet the entry standards then be removed from the research model. Four remaining variables which are the responsiveness, the tangibles, the reliability and the trust are qualified and are put into the model to be considered.

The multiple regression equation is estimated by the Stepwise (Table 4.5) shows that the model 4 including independent variables of the trust, the responsiveness, the reliability and the tangibles is the most appropriate to express the satisfaction.

Table 4.5: Summary of the model

Model Change Statistics R Square

.451 .044 .011 .012

32 (a) Independent variable: Trust

(b) Independent variable: Trust, Responsiveness

(c) Independent variable: Trust, Responsiveness, Reliability

(d) Independent variable: Trust, Responsiveness, Reliability, Tangibles (e) Dependent variable: Satisfaction

Adjusted R square = 0.508 (Table 4.5). It illustrates that 50.8% of variances of satisfaction is explained by the four above independent variables, and the rest is affected by other variables.

The best regression equation of the satisfaction of customers (Table 4.6):

Satisfaction = 3.46E-016 + 0.515 * trust + 0.254 * responsiveness + 0.120 * reliability + 0.117 * tangibles

The F-test resultswith the value sig. = 0.000 shows that the model of multiple linear regression is appropriate with the set of data and it can be used (Appendix 7).

The value of sig. (Table 3.8) of independent variables of the trust, the responsiveness, the reliability and the tangibles is less than 0.05. It is significant in the model.

Table 4.6: Statistical coefficient of each variable in the regression equation Coefficients^a

3

a. Dependent Variable: Satisfaction

The result of the regression model testing (Table 4.6) demonstrates that there is no multicollinearity phenomenon because the variance inflation factors – VIF of variables in the model are very low, from 1.0 to 1.38 and less than 10 (Hoang Trong and Chu Nguyen Mong Ngoc, 2005).

Checking whether the variance of error is unchanged which is infringed or not by correlative testing Spearman with the hypothesis Ho is: the ranking correlation coefficient of overall is 0. The regression equation has lots of explained variables then the ranking correlation coefficient could be calculated between the absolute value of the residual and each separate variable.

The inspection result shows that (appendix 8): value sig. of variables of the credibility, the responsiveness and the tangibles have the absolute value of the residual in order are 0.755, 0.856, 0.332 and 0.041. This means that we cannot reject the hypothesis Ho, meaning the variance of error is unchanged. Thus, the linear regression model above could be used (Hoang Trong and Chu Nguyen Mong Ngoc, 2005).

Inspecting the dispersed chart of the standardized residual and the standardized predicted value shows that the residual distributes randomly and does not form any specific shape (appendix 9). So, the linear association and equal variance are satisfied.

Checking the chart of frequencies (appendix 10) shows that the residual distribution is normal approximate (in average mean=0 and the standard deviation Std.

Dev.=0.988 is closing to 1). Thus, we can conclude that the standard distribution hypothesis is not infringed.

33

Therefore, the regression equation is performed as above is appropriate. The factor of trust with the regression coefficient of 0.515 is the most influential factor to the satisfaction of customers. Following factors that have influence levels on the satisfaction of customers such as the responsiveness with the regression coefficient of 0.254, the reliability with the regression coefficient of 0.120 and the tangibles with the regression coefficient of 0.117.

The assurance is removed from the equation; it does not mean that this variable does not contribute to satisfying customers. We can know that with the system of the hotel service of An Giang Tourimex, this basic content is accepted by customers. If we invest to improve this content, perhaps it will increase the satisfaction of customers.

However, the improving level will not be much as the other contents. Instead of that, we need to focus on upgrading the contents in the order of the importance as follows:

the trust, the responsiveness, the reliability, and the tangibles. It will work much more efficiently.