Extended Data Fig. 4: Variance partitioning analysis for group, individual and session-based variance.
From: Infants have rich visual categories in ventrotemporal cortex at 2 months of age
(a,b,c) Group average pairwise covariances across 12 categories’ beta estimates in EVC, VVC and LO respectively using a 12-category GLM that is collapsed across exemplars. Pairwise comparisons were calculated within subjects/within run (331 unique pairs in n = 101 2-month-olds, 128 pairs in n = 44 9-month-olds, 51 pairs in n = 17 adults), across subjects/runs (54125 unique pairs in 2-month-olds, 7980 pairs in 9-month-olds, 1224 pairs in adults) and within subject/across runs (490 unique pairs in 2-month-olds, 148 pairs in 9-month-olds, 51 pairs in adults) to partition the session, group and subject variances respectively. Variance estimates are not z-scored, and X and Y axes of the matrices are identical, ordered by stimulus category. (d) Overall estimate of within run (session), across subjects (group) and across run (individual) noise calculated using the trace of the RSMs in (a) and (b) for EVC, VVC and LO. All sources of noise varied with age. Error bars are the 95% confidence intervals, calculated by bootstrap resampling across subject/run pairs in the group average RSM. (e) The same noise estimates in (c) and (d), with only the group and individual level variance plot to illustrate developmental changes in shared and idiosyncratic noise. The solid line is the true value of the trace in the covariance similarity matrices; the error bands are shaded between the 95% CI values at each categorical age variable on the x axis.
