Retained row order gave geometric mean contrast variance ratios above one in 431 AOMIC designs with fixed masks and AR(1) noise at rho 0.6

We tested uncertainty after motion removed fMRI rows from 431 public working memory designs. The noise model was stationary AR(1), with rho 0.6 and known marginal variance one. We treated each motion mask as fixed. The tested contrast was change minus nochange. The consecutive transform treated retained rows as adjacent original scans. Its geometric mean true-to-nominal variance ratio was 1.103 in 207 PIOP1 designs and 1.063 in 224 PIOP2 designs. The 95% person bootstrap intervals were 1.084 to 1.125 and 1.051 to 1.077. The exact time gap transform gave a ratio of one within numerical precision. The package provides the transform, proof, all nine planned checks, independent calculations, Gaussian simulations, source hashes, and reproduction instructions. The result concerns the stated noise model. We did not estimate noise from brain signals or test a software package.

fMRI · AR(1) covariance · motion censoring · time gap whitening · contrast variance · working memory