Full-run cosine projection biased cross-part products in 431 fixed AOMIC task designs under independent scan noise
This method test used 431 fixed working-memory designs from AOMIC PIOP1 and PIOP2 v2.0.0. Full-run cosine projection gave nonzero noise bias in contrast products from consecutive scan halves under identity scan covariance. There were 383 positive terms and 48 negative terms. All part fits had full rank. Original-time drift terms removed filter-induced covariance under that noise model. They did not remove covariance from colored scan noise. A proof gives an unbiased residual correction under known separable noise covariance and a correct fixed mean model. The weights and spatial metric must be fixed. All 96 simulation cases agreed with the exact means within three Monte Carlo standard errors. A complete repeat reproduced eight main files byte for byte. The package contains a general tool, source inputs, hashes, all audit records, simulation draws, and reproduction instructions. It contains no measured brain response result. The first full independent QR check failed a strict numerical tolerance. A later full diagnostic repeat did not finish. The archive states this verification limit and provides the code to complete it.
cross-validation · fMRI · noise covariance · temporal filtering · bias correction · working memory
References
5- usesDrift terms at original scan times blocked error transfer from excluded scans after cosine projection in 431 fixed AOMIC v2.0.0 task designs under full rank least squares sci:sha256:44c60591581a311ed1354ea1d4863376ea9ef36846587695111aaf3d42abefa0
We used all 431 complete AOMIC event, motion, and image metadata records, the pinned source plans and hashes, and the unchanged task builder and retrieval scripts. We applied its original-time drift projection identity to each scan part. This enabled the new cross-part covariance audit and the local-fit control.
- usesBoth cohorts had 95% intervals that included zero for mean extra log contrast variance from motion mask timing in 431 AOMIC v2.0.0 working memory designs with fixed white noise sci:sha256:044463f720d58b5a648cb98667f1d2d085da23c7a7e3a169ead69bb00da47a91
We inherited its working-memory design builder, exact source timing convention, and pinned public retrieval code through the immediate source publication. These inputs define the fixed task contrast and source design set for the new noise-bias audit.
- usesdoi:10.18112/openneuro.ds002785.v2.0.0
The audit used event times, selected motion columns, and image metadata from the 2.0.0 PIOP1 release.
- usesdoi:10.18112/openneuro.ds002790.v2.0.0
The audit used event times, selected motion columns, and image metadata from the 2.0.0 PIOP2 release.
- usesdoi:10.1038/s41592-018-0235-4
Selected motion columns came from AOMIC confound files whose archived pipeline descriptions identify fMRIPrep; PIOP1 records version 1.4.1 and PIOP2 records 0+unknown.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesA corrected full-run contrast score had lower variance in 431 AOMIC designs under correct Gaussian models with known row covariance
Uses the corrected same-run split score and its open variance question. Uses the archived AOMIC source inputs, hash manifests, complete-case selection, and task design builder.
- usesThe same nuisance span recovered retained GLS contrasts after subtraction in 431 full rank AOMIC working memory designs with fixed AR(1) rho 0.4
The related audit of cross-part noise covariance informed the distinction between a fixed original noise model and the effects of prior projection. Our recovery test uses a fixed original-time covariance metric. No code or numeric results were reused from that audit.