19 NTU Management Review Vol. 31 No. 2 Aug. 2021 online customers. We model customers’ gratitude and trust by using data gathered at three points in time, and we employ the latent growth curve approach to assess how loyalty is impacted by both the static (level) and dynamic (velocity) components of the relationship state. We gathered our data from members of an online shopping platform. After three waves of the survey were completed, we obtained a total of 358 valid samples. We employ SPSS 22 and LISREL 8.80 software to perform confirmatory factor and latent growth curve analyses and to test for validity, reliability, common method variance and measurement invariance. All measures use a 7 point Likert scale. In accordance with Anderson and Gerbing (1988), we employ confirmatory factor analysis with nine constructs (relationship benefits = t1, gratitude and trust = t1-t3, attitudinal and behavioral loyalty = t3). The analysis results show that the measurement model fit the data acceptably. For all measures, the Cronbach’s α is between 0.685-0.962, the standardized factor load is between 0.5-0.93, the composite reliability is between 0.590.91, and the average variance extracted is between 0.41-0.63, all of which are within the range of acceptability. In accordance with Anderson and Gerbing’s proposed criteria for convergent validity (1988), the correlation coefficient between constructs and the confidence intervals do not include 1, indicating that each construct has good discriminant validity. We test and find that goodness-of-fit for the single factor model (χ2 = 6791.62, df = 902, CFI = 0.53, NNFI = 0.51, SRMR = 0.16, RMSEA = 0.192) is worse than that of the measurement model. Thus, the data collected via our questionnaires can be used to examine the relationship between variables, without the risk of common method variance. Before testing the hypotheses, we must confirm the longitudinal validity (gratitude and trust) by performing a measurement invariance test (Ployhart and Vandenberg, 2010). 3. Findings We use three items to assess gratitude regarding the relationship at one point in time (level), and the direction and rate of change of the relationship (velocity). We repeat the same process for the construct of trust. First, our analysis of the unconditional latent growth curve model shows that the gratitude velocity and trust velocity are statistically significant. The results reveal that the goodness-of-fit of the linear latent growth curve model is better than that of the non-linear latent growth curve model.
RkJQdWJsaXNoZXIy ODg3MDU=