Residual covariance terms for face-house contrasts had opposite signs before spatial centering in 7 ECoG patients and 6 fMRI participants

We tested how cross-site covariance affects a learned face versus house contrast. Public studies supplied separate groups of seven ECoG patients and six fMRI participants. Earlier observations set each site preference. Later observations set within-category residual covariance. The two signed covariance components passed the fixed shared tests in both modalities. Both component tests had Holm p=0.046875. Within-preference contributions, divided by the marginal-variance reference, were 6.665 (1.458 to 12.284) in ECoG and 33.607 (26.967 to 40.237) in fMRI. Opposite-preference cancellation was 1.814 (0.836 to 2.928) and 15.211 (9.519 to 24.164), respectively. Intervals are pointwise 95% person bootstrap intervals. The net-addition shared test failed. Spatial centering changed mean cancellation to -0.066 (-0.168 to 0.024) in ECoG and -3.049 (-3.988 to -2.061) in fMRI. These controls limit a neural mechanism claim. Category residuals include image, task, drift, and measurement effects. The studies differ in people, spatial sampling, and response duration. We did not match people or sites. The package preserves the local fixed plan, code, source hashes and terms, all person results, controls, figures, and repeat steps. This is a measurement-level test of a fixed reader. It does not establish a limit on all decoders or a causal mechanism.

human neuroscience · ECoG · fMRI · noise covariance · visual categories · separate cohorts