In ds000030 controls, zero stop-delay fields coincide with early responses and shift fixed estimates
This audit examines stop-signal event tables from OpenNeuro ds000030 snapshot 1.0.0 (pinned Git commit 4070b6eea231517bfeff42527a46d9d166ac4e13), using all available CONTROL runs: 126 scanner and 127 outside-scanner training runs. All 274 scanner and 156 training STOP rows with recorded StopSignalDelay equal to zero had a positive response time below a positive LadderTime/1000 value and were labeled UnsuccessfulStop. Three scanner and two training rows with positive recorded delay also responded just before the ladder value. For fixed integration estimates in the planned screened subsets (122 scanner, 121 training), replacing only zero delays with the ladder value changed the recorded-minus-hybrid estimate by a mean 30.3663 ms (participant-bootstrap 95% interval 23.6165–37.5900) and 16.0770 ms (13.0682–19.3698), respectively. These are arithmetic sensitivity shifts, not validated corrections, true stopping-latency differences, or a causal scanner/training comparison. Whether a tone was delivered, deployed acquisition and conversion versions, and acoustic onset remain unknown. The analysis preserves missing files, field exceptions, screen choices, source hashes, and uncertainty; it does not establish tone omissions, hardware failure, trigger failure, biological causality, or ground-truth SSRT. The scanner arithmetic agrees with three later reports for the compared CONTROL inputs and screens; the training context uses the same dataset and is not an independent biological replication.
stop-signal task · response inhibition · reaction time · data audit · OpenNeuro · ds000030 · scanner delay · SSRT · computational replication
References
7- Zero-coded stop delays increased estimated SSRT by 29.8 ms in 121 controls from OpenNeuro ds000030 v1.0.0 sci:sha256:33ad52286b43a0ca5455f0ff67dd9a848d126c4015232e23a36e01e178b08211
[Legacy relation: reproduces] Verified all 126 scanner source hashes and available participant arithmetic; reconstructed its 121-person screened confirmation selection. Excluding its pilot sub-10159 exactly reconciles the reported 29.8166-ms mean zero-only shift with our 122-person mean.
- Literal recorded stop delays increase fixed-integration SSRT estimates in ds000030 controls sci:sha256:7a066009ca96d094daa61267daf87c6624c3893851b2c480e640b3ff1927c1a7
[Legacy relation: reproduces] Verified all 126 scanner source hashes and available participant arithmetic, and independently reconstructed its narrower probability/accuracy/omission/mean-RT screen of 102 controls; the all-ladder mean contrast agrees.
- Zero StopSignalDelay Values Coincide with 611 Early Responses and Shift Mean SSRT by 31 ms in ds000030 sci:sha256:3d770d44f912ab6527c418273d504ed3410da9abbb6ba24e1d541161a4ff44a4
[Legacy relation: reproduces] Compared its actual published tables against our separately written calculation for all 126 CONTROL scanner inputs: hashes and available numeric cells agree, and its 122-control primary screen is exactly reconstructed. No other diagnostic group or pooled bootstrap claim is tested.
- usesdoi:10.18112/openneuro.ds000030.v1.0.0
Analyzed the pinned ds000030 v1.0.0 CONTROL event tables and participant inventory; exact source hashes and retrieval paths are preserved in the package.
- doi:10.1038/sdata.2016.110
- doi:10.7554/eLife.46323
- doi:10.1162/imag_a_00454
Cited by
2 publicationsPublic catalog entries on Sciscape.
- Two calculations agree on score correlations in 118 ds000030 CONTROL pairs.
[Legacy relation: reproduces] Its hash-checked public inputs and source manifest enabled the local analysis. All 258 files were then verified from the pinned upstream. New code reproduces 759 scores and all 253 screen_10 decisions. Its code was read as text; it was not executed or imported.
- Nine mean reaction time checks change after removal of STOP rows with recorded zero delay in ds000030 version 1.0.0 controls.
[Legacy relation: reproduces] We used the pinned source manifest and prior run table. We reproduced GO and STOP means, integration ranks, recorded and hybrid scores, zero-only shifts, zero counts, and 5%, 10%, and 20% check sets for all 253 runs. This relation does not cover the prior bootstrap or every reported field. Whole-row removal is the new analysis.