Euclidean weight cosines and contrast gains have different signs in 35 of 910 tested ds000102 run-2 model pairs.
Using recorded event and motion designs from 26 participants in ds000102, we compared two disagreeing run-2 condition encodings under seven motion settings and a fixed AR(1) covariance grid (rho 0, 0.2, 0.4, 0.6, 0.8). Euclidean contrast-weight cosines and transfer gains had different signs in 35 of 910 repeated run-2 model pairs, all at positive rho. Five participant designs had negative gains under all settings; 13 of 728 positive-rho pairs changed gain sign relative to rho zero. The sign of conditional noise correlation matched both gains under the specified GLS model. These are conditional calculations on recorded designs, not independent observations or measured brain effects. We did not identify the true condition labels, analyze images, estimate actual noise, or make a population or causal claim. Results apply to the recorded designs, fixed response and sample assumptions, and tested covariance inputs.
functional MRI · contrast weights · generalized least squares · temporal covariance · motion correction · condition labels · noise correlation · ds000102
References
3- uses10 of 26 ds000102 run-2 contrast weight cosines are less than zero with six motion columns. sci:sha256:1c3328d21134072be0392427b7646a719f36b4065c758c2b7bfaef3fd44b14c9
We reused reviewed event-convolution and motion helpers, the seven fixed model settings, source hashes and versions, and all 364 baseline cosines. This enabled a fixed-covariance comparison with identical recorded designs.
- doi:10.1016/j.neuroimage.2010.10.046
- doi:10.52294/ApertureNeuro.2022.2.NXOR2026
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesThe examined ds000102 contrast gain analyses have no measured covariance bound.
We used the reviewed design and retrieval functions, source checksums, model settings, and gain comparison values. They supplied the recorded designs and reference calculations for the covariance thresholds.
- usesSerial correlation increases the median variance cost of label-robust common-amplitude weights
I used its five AR(1) inputs and unit marginal variance convention. This work calculates common-amplitude variance ratios for a different coefficient.
- A second calculation gives the same 35 sign differences in 910 ds000102 run-2 model pairs.
[Legacy relation: reproduces] Fresh checksum-matched source files and separately written gamma-integral, separate-condition dense GLS code reproduce all 1,820 primary rows, 780 reset rows, 53,144 mask rows, and all 35 run-2 sign differences. The inspected target code also gives nine unchanged CSV files on rerun.
- usesThree of 26 primary ds000102 two-run model pairs have negative noise correlations at a specified variance ratio of four.
The difference between Euclidean weight cosine and conditional noise correlation defined the metrics for the two-run variance tests.