Two shared label errors can reverse the run-2 contrast under the specified model
This sensitivity analysis asks how many shared errors in agreeing condition labels would permit a nonpositive mean difference in correct incongruent-minus-congruent response time in OpenNeuro ds000102 snapshot 00001. All 26 participants and 624 run-2 trials enter a finite-sample model. The 304 conflicting labels remain free, while at most B of 320 agreeing labels may change; condition balance, correctness, and response times remain fixed. Without an upper response-time limit, the minimum contrast is 7.9869 ms at B=0, 0.2689 ms at B=1, and -6.2118 ms at B=2. Thus two hypothetical shared errors suffice for a negative contrast and one does not. Excluding the one correct response time above 2 s gives a minimum of 3.8663 ms at B=0 and -2.6145 ms at B=1. These are exact sensitivity bounds under the stated assignment model, not measured label errors or corrected data. The tested files do not identify the presented arrows; the true shared-error count and population sign remain unresolved.
label errors · sensitivity analysis · flanker task · response time · condition labels · OpenNeuro · ds000102 · finite-sample bounds · dynamic programming
References
2- usesIn ds000102 snapshot 00001, 304 run-2 label conflicts shift the RT contrast by 135 ms sci:sha256:15862d75b2281569c5f5e44449ab3007220eb7bd91aefa8a5653b7dccddb1e2d
We used the stated condition-label model, the 56-file source manifest, and the exact zero-cost bounds. They gave the fixed-agreement baseline for this shared-error calculation. All 26 participant baseline bounds match. We did not execute or import the prior code.
- doi:10.1016/j.neuroimage.2010.10.046
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesOne shared label change gives a zero lower bound for the two-run error cost in the ds000102 model.
I used its shared group budget for changes to rows where the two source condition fields agree. I used that budget with the error-rate statistic.
- usesThe ds000102 lower bound is negative at 57 shared label errors when labels in two untrimmed runs can change.
Used the stated shared-label-error model, pinned 56-file source manifest, and archived run-2 numeric bounds. Fresh source files and a separate implementation reproduced all 1284 run-2 endpoints. The new statistic gives equal weight to 52 participant/run contrasts and allows errors in either run, with a fixed-run-1 comparator.