In the previous sections, we established the reliability and validity of the measures. In this section, we will provide evidence supporting the theoretical model as illustrated by the inner model (section 4.7.6). Since our model is based on formative measures, we do not provide for goodness of fit measures. “Models with low R-square and/or low factor loadings can still yield excellent goodness of fit” (Chin 1998b). We rather try to relate how well the endogenous latent variable is predicted. Chin (1998a) suggests the following criteria for assessing inner
models: (a) the
R
2 of endogenous latent variables, (b) the estimates for the path coefficients with bootstrapping to examine their significance, (c) the effect size2
f
, and (d) the prediction relevance (Q
2andq
2).The last criterion for evaluating the predictive relevance is the predictive
sample reuse technique, that is
Q
2 which is calculated based on the blindfolding procedure. However, “The blindfolding procedure is only applied to endogenouslatent variables that have a reflective measurement model operationalization” (Henseler, Ringle et al. 2009). Therefore, we are not using the values of Q2 and
2
q to assess the predictive relevance and the relative impact of each exogenous
latent variable to the latent variable under evaluation.
As discussed previously, the predictive power of the model is assessed by
the coefficient of determination value (R2) of the endogenous construct, “SB
purchase” as well as by the significance of all path coefficients. PLS R-square is
an index that helps us to relate how well SB purchase (the dependent variable) is
predicted by the overall model. Chin (1998a) suggests that
R
2 values of 0.67 can be considered as substantial, values of 0.33 as moderate, and values of 0.19as weak. Furthermore, a moderate
R
2 value can be accepted when a few exogenous latent variables are used to explain an endogenous latent variable.The results for this study indicate that the
R
2 values of the endogenous reflective construct “SB purchases” are moderately acceptable in respect to theoverall model. Specifically customer satisfaction, trust in SB and WOM explain approximately 24% of SB purchase variance. As far as the affect that customer satisfaction has on word-of-mouth, the path coefficient (0.389) as well as the t- value (10.88) indicate that customer satisfaction has a high impact upon word- of-mouth. In contrast, the path coefficient (0.007) as well as the t-value (1.72) indicate that customer satisfaction has a low impact upon trust in SB. Results are shown in Table 5.7 and a schematic representation of the results is provided in Figure 5.1.
Figure 5.1: Inner Model Results – H1, H3, H4, H5, H7
The second criterion for assessing the inner model is the values of the path coefficients that the PLS algorithm provides us with. These values should be evaluated in terms of sign and magnitude, and their significance is assessed through bootstrapping. The standardized path coefficients of the inner model indicate that “trust in SB” has the strongest relationship with the dependent
variable while the constructs of customer satisfaction and word-of-mouth have a very small affect on SB purchases. Furthermore, they indicate a strong relationship between customer satisfaction and word-of-mouth and a weak relationship between customer satisfaction and level of trust in SB.
The change in R-squares was explored to identify the impact each latent variable had on the dependent latent variable. The results indicate that the level of trust in SBs has the highest impact on SB purchases. Furthermore, we
calculate the effect size f2, in order to specify the effect of the predictor latent
variable at the structural level (Chin 1998a). The following equation depicts algebraically the procedure for calculating the effect size:
2 2 2 2
1
included excluded includedR
R
f
R
The
R
included2 is the value when customer satisfaction, trust in SB and WOM areused to predict SB purchases, that is 0.235. The
R
excluded2 values are provided in the table 5.7 below, that is 0.233 when customer satisfaction is omitted, 0.036 when trust in SB is omitted, and 0.230 when WOM is omitted. The effect sizerespectively is f2of 0.0026, 0.2601, and 0.0065.
In addition, the predictive power of the model is assessed for four out of the nine grocery retailers in order to identify any variations. Two foreign retailers (AB and Carrefour), one Greek retailer (Sklavenitis), and one hard discounter (Lidl) were selected. These four retailers were selected for their size, their heavy involvement in the development of SBs, as well as their overall strategy. Table 5.8 presents the inner model results for these four grocery chains. The constructs of customer satisfaction, trust in SB and WOM for the store explain SB purchase variance from 22% to 39%. Therefore, the total model results are
endogenous reflective construct “SB purchases” are also moderately acceptable
in respect to the overall model for the four grocery chains.
Table 5.7: Inner Model Results – H1, H3, H4, H5, H7
Constructs Inner Model Inner Model Excl. Customer Satisfaction Inner Model Excl. Trust in SB Inner Model Excl. WOM Dependent SB purchases (R2) .235 .233 .036 .230 Independent Customer Satisfaction .040 (1.25) -- .047 (1.21) .066 (1.85) Trust in SB .456 (14.49) .457 (14.85) -- .469 (14.89) Word-of-mouth .072 (2.17) .088 (2.83) .165 (4.69) -- 2 f = .0026 f 2= .2601 f 2= .0065 Customer Satisfaction to word-of-mouth .389 (10.88) Customer Satisfaction to trust in SB .007 (1,72)
Table 5.8: Inner Model Results for the four grocery chains
Constructs ALL RT AB Sklavenitis Carrefour Lidl Dependent SB purchases (R²) .235 .388 .218 .228 .290 Independent Customer Satisfaction .040 (1.25) -.051 (.31) -.069 (.39) .026 (.20) -.040 (.24) Trust in SB .456 (14.49) .587 (4.42) .391 (1.73) .451 (5.23) .550 (4.24) Word-of-mouth .072 (2.17) .143 (1.21) .226 (2.10) .088 (0.94) .015 (.16) Customer Satisfaction to word-of-mouth .389 (10.88) .515 (3.42) .519 (2.80) .298 (2.02) .307 (2.02) Customer Satisfaction .007 (1.72) .011 (.46) .009 (.50) .006 (.34) .009 (.50)
Overall these results suggest that H1, H3 and H7 cannot be supported. We cannot support that customer satisfaction (H1) and word-of-mouth (H7) have a strong impact on SB purchases, nor that customer satisfaction has a strong impact of the level of trust in SB (H3). Moreover, we find a significant positive impact of the level of trust in SB on SB purchases, and of customer satisfaction on word-of-mouth. Consequently, we find support for the affect of customer satisfaction on word-of-mouth (H4) and for the affect of the level of trust in SBs on SB purchases (H5).