Consistent with the approach taken above, we first evaluated the correlation between these alternative models and the ROI divisions of OPA in each hemisphere separately before considering OPA as a whole. Similar to our previous analyses, the pattern across the ROI divisions of OPA was varied (
Figure 6). To quantify the relationship between maps, these values were submitted to a linear mixed model with ROI (LO1, LO2, V3A, V3B, V7, and OPA Other) and model (GIST, LGN, CNN, and navigational affordances [NA]) as factors. In the left hemisphere, we found only a significant main effect of model (
F(3, 356) = 12.33,
p = 1.07-7), driven primarily by higher correlations with the LGN and GIST models on average (ROI by model interaction
F(15, 356) = 0.91,
p = 0.54, main effect of ROI
F(5, 370) = 1.37,
p = 0.23). Pairwise comparisons (Bonferroni corrected) indicated several significant comparisons: LGN versus CNN,
t(354) = 4.12,
p = 0.0003; LGN versus NA,
t(354) = 5.4,
p < 0.0001; GIST versus CNN,
t(354) = 2.77,
p = 0.03; and GIST versus NA,
t(354) = 4.06,
p = 0.0004 (
p > 0.05 in all other cases). We also found a significant main effect of model in the right hemisphere (
F(3, 322) = 6.72,
p = 0.0002), but again no ROI by model interaction (
F(15, 322) = 0.50,
p = 0.93) or main effect of ROI (
F(5, 348) = 1.98,
p = 0.08). On average, correlations with the LGN model were highest, followed by GIST: LGN versus CNN,
t(332) = 3.09,
p = 0.01; LGN versus NA,
t(332) = 4.05,
p = 0.0004; GIST versus NA,
t(332) = 2.88,
p = 0.02 (
p > 0.05 in all other cases, Bonferroni corrected). As above, we considered the impact of different numbers of ROIs by rerunning our LMM with only those participants with complete ROIs. Importantly, the ROI × model interaction remained nonsignificant in both hemispheres (
p > 0.05), with the main effect of model reaching significance in both hemispheres and the main effect of ROI in the right hemisphere (see
Supplementary Table S4 for a full statistical breakdown).