Figure	χ2 (df), p value	CFI (TLI)	RMSEA (SRMR)	nBIC	Model fit comparison
Figure 4a+	452.82 (200), < .001	.94 (.93)	.06 (.09)	45277.69
Figure 4b+	418.50 (199), < .001	.95 (.94)	.06 (.06)	45246.05	Figure 4a+ vs. Figure 4b: Δχ2 (Δdf = 1) = 34.32, p < .001
Figure 4c+	374.10 (199), < .001	.96 (.95)	.05 (.05)	45201.65	Figure 4a+ vs. Figure 4c: Δχ2 (Δdf = 1) = 78.72, p < .001 Figure 4a+ vs. Figure 4c: ΔnBIC = 44.4

The Figure 4c+ model was the best fitting model according to chi-square values and nBIC values. To compare the Figure 4c model here with that of Figure 5a, nBIC values were compared because these models are not nested. The nBIC value of the Figure 5a model is smaller by 6.10 than that of the Figure 4c+, indicating superiority of the Figure 5a model according to the criteria by Raftery (1995). That is, the reading comprehension-to-writing quality model shown in Figure 5a fits the data better than the writing quality-to-reading comprehension model (Figure 4c+ model).