Table 3.

Critical comparison

	LANG (p)	Interaction with network (p)
DLT-VCM over neither	0.0001***	0.0001***
DLT-S over neither	0.0033***	0.0013**
DLT-VCM over DLT-S	0.0001***	0.0001***
DLT-S over DLT-VCM	0.3301	0.0007***

The p values that are significant under eight-way Bonferroni's correction (because eight comparisons are tested) are shown in bold. For the LANG network [LANG (p) column], integration cost (DLT-VCM) significantly improves network generalization r_diff both alone and over DLT-S, whereas DLT-S only contributes significantly to generalization r_diff in the absence of the DLT-VCM predictor (significant over “neither” but not over DLT-VCM). For the combined models [interaction with network (p) column], the interaction of each variable with network significantly contributes to generalization r_diff in all comparisons, supporting a significantly larger effect of both variables in the language network than in the MD network.