Mediators of the Customer Satisfaction-Loyalty Relationship

Exploratory Factor Analysis

The empirical analysis for this study started with exploratory factor analysis (EFA), which was used to reduce the number of scales assigned to each elaborated online behavior dimension. EFA was conducted in SPSS, using the Principal Components method, in order to extract the factors and the Schwartz's Bayesian Criterion (BIC) clustering criterion.

To establish the adequacy of the EFA, we used the Kaiser-Meyer-Olkin Measure of Sampling Adequacy (KMO) and we obtained high values (between 0.684 and 0.855) that indicate that the factor analysis is relevant for this data analysis. The psychometric properties of the measures developed in the questionnaire are presented in Table 2, and the results for the exploratory factor analysis are shown in Table 3.

Additionally, all the scales of the analysis were checked for internal consistency and reliability through Cronbach's alpha. Reliability is identified by Cronbach's alpha with a minimum of 0.70 (Cronbach, 1970). As shown in Table 3 all values were above the recommended level of 0.7, with values that vary from 0.774 to 0.862.

For this newly proposed model, one factor was extracted for all the dimensions studied. The criteria used to identify and interpret the factors was that each element should have a factor loading greater than 0.7 and Eigenvalues greater than 1. Also, the eligibility of the factors can also be observed in terms of the variance explained by each resulted factor, as the variation exceeds 70%. Therefore, all the factors are eligible and can be used in further analysis, namely regression analysis with mediators.

Regression Analysis with Mediation

In order to study the interrelationships between three variables, we use mediation; this involves a set of causal hypotheses regarding e-satisfaction and e-loyalty in an Internet retailing environment, while observing the mediating effect of various variables. In other words, mediation implies that an initial causal variable X (satisfaction) may influence an outcome variable Y (loyalty) through a mediating variable M. Mediation occurs if the effect of X (satisfaction) on Y (loyalty) is partly or entirely “transmitted” by M. A mediated causal model involves a causal sequence; first, X (satisfaction) causes or influences M; then, M causes or influences Y (loyalty). X (satisfaction) may have additional direct effects on Y (loyalty) that are not transmitted by M. A mediation hypothesis can be represented by a diagram of a causal model. Figure 5 displays the general mediation regression analysis used for the four conceptual proposed models.

To address the research questions of this paper, we examine the level of the mediating effect of each proposed mediator and whether it is a partial, or a complete, mediator. Several methods to test statistical significance of mediated models have been proposed. In this sense, Sobel test is such an example of method used in this analysis. The Sobel procedure was then used to statistically investigate the effect of the proposed mediator on the predictor–outcome relationship. The following z ratio for the Sobel test can be set up as follows:

$𝑧 = 𝑎𝑏/ \sqrt{𝑏^2𝑠_𝑎^2 + 𝑎^2𝑠_𝑏^2 }$

where

a and b are the raw (unstandardized) regression coefficients that represent the effect of X on M and the effect of M on Y, respectively;

$s_a$ is the standard error of the a regression coefficient;

$s_b$ is the standard error of the b regression coefficient.

For the mediating effect of trust on the satisfaction – loyalty relationship, the test statistic for the Sobel test is 2.51, with an associated p-value of 0.011. The fact that the observed p-value does fall below the established alpha level of .05 indicates that the association between the independent value (satisfaction in e- commerce environment) and the dependent value (in this case, the loyalty of e-customers) is increased significantly by the inclusion of the mediator (in this case, trust in Internet retailing) in the model; in other words, there is evidence of mediation, and thus hypothesis 2 is not supported. In this case, trust does not do a very good job in predicting loyalty in e-commerce setting.

The same line of analysis is accomplished for all other mediation variables considered: attitude (Table 6), hedonic value (Table 7), and utilitarian value (Table 8).

The results of the interrelationships when considering hedonic value as a mediator are presented in Table 6. The R2 of both equations are statistically significant, though their values are rather moderate (particularly in in the attitude - satisfaction equation).

Table 6. Regression model 2 for the mediating effect test of attitude

Variable	Model 2: Attitude
		E21: AT →SATIS	E22: AT + SATIS →LOY
Independent variable (Satisfaction)		0.548 (p < 0.001)
Mediator variable (Attitude)	0.480 (p < 0.001)	0.155 (p = 0.005)
R2	0.231	0.406
Std. Error of the Estimate	0.086	0.086
F	31.474	35.489
Sobel test for mediating effect	Test statistic	Std. Error
	1.715 (p=0.068)	0.043

The Sobel test performs a statistical test to see if the indirect path from the independent value to the dependent value is statistically significantly different from zero. This is the same idea as the test providing support for partial mediation. The test statistic is equal to 1.715, with standard error 0.043 (Table 6). The statistical significance is equal to 0.068. Assuming we had set our alpha at 0.05, technically, we would not reject the null hypothesis of no mediation. However, the 0.05 level is an arbitrary cut-off value, and 0.068 is very close to it, therefore in this case there is some evidence for partial mediation of attitude on the satisfaction – loyalty relationship in an e-setting. The product 0.480 × 0.155 is 0.074 and this value estimates the strength of the mediated or indirect effect of satisfaction on loyalty, that is, how much of the increase in customer loyalty occurs as satisfied people have a positive attitude towards online shopping. The 0.548 value estimates the strength of the direct (also called partial) effect of satisfaction on customer loyalty in Internet retailing, that is, any effect of satisfaction on loyalty that is not mediated by attitude. Further, the sum between 0.074 and 0.548 provide the total relationship between satisfaction and loyalty, considering the mediated effect of attitude.

The results of the interrelationships when considering hedonic value as a mediator are presented in Table 7. The R2 of both equations studied in these models are strong and statistically significant, denoting that these two variables (hedonic value and satisfaction) do a good job of predicting variance in customer loyalty in an e-setting.

Table 7. Regression model 3 for the mediating effect test of hedonic value

Variable	Model 3: Hedonic Value
	E31: HV →SATIS	E32: HV + SATIS → LOY
Independent variable (Satisfaction)		0.156 (p < 0.001)
Mediator variable (Hedonic Value)	0.675 (p < 0.001)	0.691 (p < 0.001)
R2	0.455	0.647
Std. Error of the Estimate	0.072	0.079
F	67.769	95.477
Sig.	0.000	0.000
Sobel test for mediating effect	Test statistic	Std. Error
	6.395 (p = 0.001)	0.072

For the mediating effect of hedonic value on the satisfaction – loyalty relationship, the test statistic for the Sobel test is 6.395, with an associated p-value of 0.001. These results indicate that the interrelationships are significant in the model and there is a relevant evidence of mediation. Hypothesis 4 is supported at a 0.005 level and hedonic value is a mediator worth considering when trying to influence the satisfaction and loyalty in Internet retailing

The results of the interrelationships when considering utilitarian value as a mediator are presented in Table 8. The R2 of the utilitarian value – satisfaction relation is rather small, and this coefficient increases when both variables display their impact on loyalty.

Table 8. Regression model 4 for the mediating effect test of utilitarian value

Variable	Model 4: Utilitarian Value
	E41: UV →SATIS	E42: UV + SATIS → LOY
Independent variable (Satisfaction)		0.536 (p < 0.001)
Mediator variable (Hedonic Value)	0.515 (p < 0.001)	0.167 (p = 0.006)
R2	0.265	0.408
Std. Error of the Estimate	0.084	0.088
F	37.914	35.787
Sobel test for mediating effect	Test statistic	Std. Error
	1.812 (p = 0.070)	0.047

Regarding the mediating effect of utilitarian value on satisfaction and loyalty, the test statistic for the Sobel test is 1.812, with an associated p-value of 0.070 (Table 8). In this case, utilitarian value is not a strong mediator for this proposed model and thus hypothesis 5 is not supported.

Nonetheless, the ANOVA values for each model report a significant F statistic, indicating that using the models is better than guessing the mean.