4.2. Hypothesis Testing
Relationships between variables were tested by SEM. Based on Hu and Bentler [33
], the maximum likelihood estimation results indicated an adequate data fit (Comparative Fit Index, CFI = 0.952, Root Mean Square Error of Approximation, RMSEA = 0.056). Figure 2
also shows that hypothesis testing results demonstrated that moral norm was positively related to household storing and cooking routines, with a standardized coefficient of 0.22 *** (p
< 0.001), which supported H1. Perceived behavioral control was not positively related to household storing and cooking routines, with a standardized coefficient of 0.05 (p
= 0.097), which rejected H2. Food choices was also positively related to household storing and cooking routines, with a standardized coefficient of 0.54 *** (p
< 0.001), which supported H3. The hypothesis testing results further demonstrated that household storing and cooking routines has a negative effect on food waste, with a standardized coefficient of −0.12 * (p
< 0.05), supporting H4. In summary, H1, H3, and H4 were supported.
To examine mediating effects, following Taylor et al. [34
], this study used percentile bootstrapping and bias corrected percentile bootstrapping under a 99% confidence interval and 10,000 bootstrap samples. To test the indirect effects, this study calculated the confidence interval of the lower and upper bounds [35
]. Table 4
shows that, according to the bootstrap test results, the indirect effect (standardized indirect effect = −0.026 *, p
< 0.05) of moral norms on food waste was significant and the direct effect (standardized direct effect = 0.000, p
= 0.968) was not significant, indicating that household storing and cooking routines serve a full mediating role for the relationships between moral norms and food waste (H5a). Additionally, the total effect (standardized total effect = −0.271 **, p
< 0.01), indirect effect (standardized indirect effect = −0.065 *, p
< 0.05), and direct effect (standardized direct effect = −0.206 **, p
< 0.01) of food choices on food waste were all significant, which indicated that household storing and cooking routines plays partial mediation for the relationships between food choices on food waste (H5c). Thus, both H5a and H5c were supported. However, the indirect effect (standardized indirect effect = −0.006, p
= 0.065) of perceived behavioral control on food waste was insignificant. That is, household storing and cooking routines have no mediating effect on the relationship between perceived behavioral control and food waste. Therefore, H5b was rejected.
According to Liu et al. [36
], the moderating effect of unplanned events was examined by running invariance tests of both the measurement model and the structural model. Prior to the metric invariance test, K-means cluster analysis was used to divide the samples into high unplanned events (n = 623; high score group) and low unplanned events (n = 331; low score group). The equality between the factor loadings of both the high score and low score groups (measurement invariance) was then calculated. First, both the unconstrained model (CFA for both groups without factor loadings) and full-metric invariance model (CFA for both groups with full factor loadings) were created. Table 5
shows the results of comparing these two models. The fit indices of the unconstrained model (RMSEA = 0.043, CFI = 0.938, Tacker-Lewis Index, TLI = 0.925) and the full metric invariance model (RMSEA = 0.043, CFI = 0.937, TLI = 0.927) showed that both models had a good fit to the collected data. Although the Chi-square difference between both models (χ² (17) = 29.886) was significant (p
< 0.05), Chi-square is sensitive to large sample sizes [37
]. Therefore, following Jöreskog and Sörbom [38
], this study considered invariance when Chi-square was significant under p
-value less than 0.005. Based on the thresholds (ΔTLI ≤ 0.02, ΔCFI ≤ 0.01) from Wang and Wang [39
], such differences between these two models (ΔCFI = −0.001, ΔTLI = 0.002) can be neglected. The above results support the follow-up step of testing invariance of the two-group structural model.
To examine invariance of the structural model, following Liu et al. [36
], a Chi-square difference test was conducted to analyze the difference between the unconstrained model (baseline model) and the constrained model (nested model). The path coefficients from household storing and cooking routines to food waste were set to equal in this study, and other path coefficients of both the high score group and low score group were estimated freely. To examine H6, the Chi-square difference between the unconstrained model and constrained model was calculated. A significant Chi-square difference indicates that unplanned events have a moderating effect. Table 6
shows the invariance test results, which demonstrate that the Chi-square difference was significant (Δχ² (1) = 8.231 **, p
= 0.004), indicating that unplanned events have a moderating effect on the relationship between household storing and cooking routines and food waste. The invariance test results also show that the coefficient estimated from household storing and cooking routines to food waste was −0.354 *** (p
< 0.001) in the low unplanned events group and 0.001 (p
= 0.984) in the high unplanned events group. As expected, when household storing and cooking routines have a higher level of unplanned events, household storing and cooking routines do not have a negative relationship to food waste. Therefore, H6 was supported.