The results of data analysis are also shown in Figure 22.9. First, as with the case of the measurement model. the proposed model was found to fit the data satisfactorily as the fit values were well within acceptable ranges [X2(24) = 43.32, P < 0.01, CFl = 0.99, GFl = 0.96, and RMSEA = 0.057]. According to TAM, perceived usefulness and perceived ease-of-use are the significant predictors of intention to use. As shown in Figure 22.9, perceived usefulness is significant in determining intention to use (path estimate = O.46,p < 0.(01). Similarly, perceived ease-of-use is found to have a significant effect on intention to use (path estimate = 0.28, p < 0.001). The squared multiple correlation (SMC) coefficient for intention to use is 0.40, which indicates that the two predictors. i.e., perceived usefulness and perceived ease-of-use, together explained 40 percent of the variance in intention to use
Conclusions and Recommendations
Overall, the results of SEM indicate that TAM is a reasonable representation of individuals’ reactions to a Web portal in an educational setting. Thus, in order to increase the use of the portal by the students, this university should enhance the perceived usefulness and perceived ease-of-use of the Web site. Perceived usefulness could be increased by adding features such as checking e-mail, weather, current events on the campus, schedule of classes, etc., that the students find useful and check frequently. Perceived ease-of-use could be enhanced by making the Web site easy to navigate.
Application of SEM: Second-Order Factor Model
We give an illustrative application of SEM in the context of banking services. The data were collected through personal interviews, and the sample size selected for this analysis was comprised of 250 respondents. The data set used in this analysis can be downloaded from the Web site for this book. In the following, we illustrate how the various steps involved in structural equation modeling presented in Figure 22.2 were carried out
Define the Individual Constructs
The purpose of this study was to predict bank patronage intent based on service quality. The theory developed was based on past research. Service quality was conceptualized as having five dimensions.P The theory postulated that service quality would influence service attitude and service satisfaction, and the latter two constructs would then influence patronage intent. Thus, in our data set, we have 8 constructs and 30 indicators, each construct having multiple indicator variables.
These are the five service quality dimensions, namely, Tangibility (4 indicators), Reliability (4 indicators), Responsiveness and Patronage Intention (3 indicators). These indie ors are contained in Table 22.3. Each indicator of the dimensions was measured using a For example, indicator one was “When it comes to modem equipment (tangibles), my pcception of my bank’s service performance is Low
1-2 -3-4-5 – 6-7-8-9 High. To measure global attitude, we used four indicators on a 7-point scale: favorable-unfavorable, good-bad, positive-negative, and pleasant-unpleasant. All attitude indicators were reverse coded for analysis. To measure overall satisfaction, we used the following four indicators using a 9-point scale: I believe I am satisfied with my bank’s services (strongly disagree-strongly agree), overall, I am pleased wilb my bank’s services (strongly disagree-strongly agree), using services from my bank is usually a satisfying experience (strongly disagree-strongly agree), and my feelings toward my bank’s services can best be characterized as (very dissatisfied-very satisfied). We used three indicators to measure patronage intention using a 9-point scale: The next time my friend needs the services of a bank I will recommend my bank (strongly disagree-strongly agree), I have no regrets for having patronized my bank in the past (strongly disagree-strongly agree), and I will continue to patronize the services of my bank in the future (strongly disagree-strongly agree
We first test the measurement model in order to validate the psychometric properties of our measures. In our example, because prior research has already established the reliability and validity of the five-component service quality construct, we test for the measurement properties in our sample using a confirmatory mode. In testing for the measurement model, we freely correlate the eight constructs and fix the factor loading of one indicator per construct to a value of unity. All measured indicators are . allowed to load on only one construct each, and the error terms are not allowed to correlate with each The mell Measurement model is described in Figure 11.\
which equals 0.064. Based on our previous discussion of model fit criteria, these fit indices collectively indicate that overall fit of the measurement model is acceptable and that the researcher needs to now test for reliability and validity.
Convergent validity is father’ established if all item loading are equal to or above the recommended cutoff level of 0.70. In our sample, of a total of 30 items in the measurement model, 8 items had loading ~O.90, 16 items With loading in the range ~0.80 to <0.90, and 6 items with loading in the range ~0.70 to ‘<0.80 (see Table 223). All of the factor loading were statistically significant at the p < 0.05 level. Thus, the data in our study supports convergent validity of the model. Discriminant validity is achieved if the square root of the average variance extracted is larger than correlation coefficients. In our study, we found all of the correlation estimates met the criterion
except in 4 out of the 28 cases. Given that the five dimensions measure different aspects of service quality, some degree of inter correlation is expected. However, given the size of the correlation matrix (i.e., 28 estimates), some violations can occur through chance. Test for discriminant validity is shown in Table 225. Values on the diagonal of the correlation matrix represent the square root of the AVE. For example, to test for discriminant validity between tangibility and responsiveness, we compare the correlation between responsiveness with their respective square root of the average variance extracted. So, the square root of the average variance extracted for tangible and responsiveness dimensions lie 0.75 and 0.81, and both are greater than their correlation of 0.65. III
summary, overall the scale items were both reliable and valid for testing the structural model
Specify the Structural Model
Based on theoretical considerations, we hypothesized perceived service quality as a-onler construct consisting of the five dimensions of ‘¥llgibility (TANG), reliability (REL), resPonsiveness
(RESP), assurance (ASSU), and empathy (EMP). Specifically, we model service quality as a second-order model with first-order dimensions of TANG, REL, RESP, ASSU, and EMP. In other words, these five dimensions are indicators of service quality and therefore the arrows flow out from service quality to the five dimensions (see Figure 22.11). On the right-hand side of the Figure 22.11, we link second-order service quality with attitude toward service (ATT) and satisfaction with service (SAT). The latter two constructs are linked to patronage intention (PAT). The entire structural model (i.e., 8 constructs) is tested simultaneously, as shown in Figure 22.11. In testing for the structural model, we free the structural linkages and fix the factor loading of . A one indicator per construct to a value of unity. All measured items are allowed to load on only one construct each, and the error terms are not allowed to correlate each other. We also fix the second-order loading of one dimension (i.e., tangibility) to unity for scaling purpose. While the measurement model tests for reliability and validity of the measures, the structural model tests for the structural relations in the model
Assess Structural Model Validity
NNFI are “,0.90 and RMSEA and SRMR are ••0.08. Collectively, these fit indices indicate that the structvral model is acceptable. That is, the second-order perceived service quality model is robust and theoretically explains the constructs of attitude, satisfaction, and patronage intention. The structural coefficients linking the five dimensions with second-order service quality (i.e., the second-order loading) are all significant and in the expected direction. Table 22.7 contains the structural coefficients with corresponding t values. For example, the loading for tangibility is 0.82, which indicates that service quality explains 67 (=(0.82)2) percent of the variance in tangibility.
Similarly, the loading for reliability is 0.85, which indicates that service quality explains 72 (=(0.85)2) percent of the variance in reliability. On the right-hand side of the figure, the coefficients for SQ-ATT is 0.60, for SQ-+SAT is 0.45, for ATT-+SAT is 0.47, for ATT-+PAT is 0.03. and SAT-+PAT is 0.88. All these links are significant at p < O.ODS.exceptfor the ATT-+PAT.
Draw Conclusions and Make Recommendations
The magnitude and significance of the loading estimates indicate that all of the five dimensions of service quality are relevant in predicting service attitude and service satisfaction. Moreover, service quality has a significant impact on both service attitude and service satisfaction as the structural coefficients for these paths are significant. Service satisfaction, in turn, has a significant impact on patronage intent. Service attitude does not directly impact patronage intent; rather, it has only an indirect effect through service satisfaction.