Multi-group comparisons of grade for relationships among the variables
Based on the full model, M1 created unconstrained model M4 (morphological equivalence) for grade, and then created fully constrained model M5 (loadings equivalence) from M4 (see Table 5), revealing that these two models exhibited acceptable fit indices. Unlike the gender group, a significant difference in χ² was observed between M4 and M5 (χ = 23.097-13.774), with a difference in df = 8 and p = 0.013<0.05. This indicated that the correlation between the variables varied across grade levels, and the models significantly differed due to these grade distinctions.
Further analysis of specific paths through which differences are manifested with grade as the moderating variable, distinct patterns emerged across grade levels. Specifically, among K1 children, five path coefficients were significant, while three were not (see Figure 4). Notably, closeness TCR did not significantly predict SR and SE, and SE did not predict mathematics in this group. For K2 children, six path coefficients were significant, while two were not (see Figure 5). It was observed that closeness TCR had no significant predictive effect on SR, and SR on vocabulary was not significant. In the case of K3 children, four path coefficients were significant, while four were not (see Figure 6). Conflict TCR did not exhibit a significant predictive effect on SR and SE, while closeness TCR did not significantly predict SR, and SE on vocabulary was not significant.
Comparing the path coefficients across the three grades, it was evident that only the prediction of SR on mathematics remained significant across all grades. This effect was most pronounced for K1 (0.364 > 0.248, 0.255). However, significant grade differences were observed in the remaining seven paths. Specifically, conflict TCR significantly predicted SE and SR for K1 and K2. Closeness TCR had a significant predictive effect on SE for K2 and SR for K3 but was not significant for SE or SR in K1. SE exhibited a predictive effect on vocabulary for K1 and K2 and on mathematics for K2 and K3. Interestingly, SR did not significantly predict vocabulary for K2 but exhibited a strong predictive effect of up to 0.454 for K1.
<Insert Figure 4 HERE>
<Insert Figure 5 HERE>
<Insert Figure 6 HERE>