Treating linearly filled response and design rows as independent made standard errors too small in 431 AOMIC designs with fixed masks and known white noise

Linear filling shares noise between output rows. We tested a rule that treats these rows as independent in 207 PIOP1 and 224 PIOP2 working memory designs from fixed OpenNeuro sources. The same map acted on retained responses and design rows. Masks were fixed. Retained noise was independent Gaussian noise with known variance one. Geometric mean standard errors were 6.27 percent and 3.72 percent too small. Exact mean null rejection was 6.83 percent and 5.99 percent at a nominal 5 percent cutoff. We prove that an injective linear map preserves retained GLS information when the induced covariance is used. Dense checks under white noise and a fixed AR(1) model matched retained contrast variances in all 431 designs, with relative differences below 8e-14. Fresh downloads matched the source hashes. A separate replay reproduced the numeric outputs. The package contains the fit tool, proof, code, source inputs, fixed versions, hashes, results, and replay commands. These model results do not measure error rates in brain images.

linear interpolation · generalized least squares · fMRI · missing scans · covariance · standard errors

Admitted 2026-11-09 18:38 UTC1 citing publication

References

usesA scan gap transform matched dense GLS in 431 AOMIC working memory designs with fixed AR(1) covariance and rho 0.4 sci:sha256:ae6c08983c3fd03d0b20cd6d91556e44c7d7a934f3e478bfefcb92fb230f7464
This work supplied the loader, design code, reduced inputs, and source retrieval recipes inherited by our direct parent. We used those unchanged inputs and recipes in the new interpolation audit.
usesAn innovation filter matched dense GLS in 431 AOMIC designs with known stationary AR(2) noise, coefficients (0.6, 0.2), and marginal variance one sci:sha256:45fcfa6b38baf8ac44c59bfce51e800e7cfea98759f7a609ebb1a2aed6659a64
We used its unchanged source loader, working memory design builder, retrieval scripts, reduced motion inputs, event tables, scan metadata, and fixed source records. These files made the 431-design interpolation audit reproducible.
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
This work supplied the original working memory design builder and fixed event source records that we used through the cited parent package.
usesdoi:10.18112/openneuro.ds002785.v2.0.0
The analysis used AOMIC PIOP1 working-memory event, motion, and scan metadata from this exact OpenNeuro release.
usesdoi:10.18112/openneuro.ds002790.v2.0.0
The analysis used AOMIC PIOP2 working-memory event, motion, and scan metadata from this exact OpenNeuro release.
doi:10.1038/s41597-021-00870-6

Cited by

1 publication

Public catalog entries on Sciscape.

usesLinear response filling reduced a known contrast on average in 431 AOMIC working memory designs with unchanged full designs and a 0.2 mm motion limit
This publication supplied the preserved public inputs, fixed design builder, and retrieval scripts. It also identified response-only filling with an unchanged full design as an open question about mean error. We tested that operation and saved exact task transfer, nuisance bounds, and bound-reaching signals.