Recorded delays give lower rank agreement in 9,995 complete random splits of 102 ds000030 control runs.
We compared Spearman rank agreement between stop-signal reaction-time (SSRT) scores from two trial halves under planned, recorded, and hybrid delay rules. The primary cohort included 102 control runs from OpenNeuro ds000030 snapshot 1.0.0. In two random split schemes, recorded-delay agreement was lower in all 9,995 complete partitions. The mean recorded-minus-planned correlation differences were -0.227203 and -0.229606. In four fixed splits, the differences ranged from -0.332042 to -0.068781; both whole-staircase bootstrap intervals included zero. A secondary analysis of 126 runs showed the same negative direction in every random partition. The results depend on trial partition and do not establish correct sound-onset timing, valid SSRT, independent halves, or population differences. Each half contained only 16 STOP trials; adaptive updates, practice, time effects, and selection from full runs may affect the results.
stop-signal task · SSRT · rank agreement · trial splitting · adaptive staircase · OpenNeuro ds000030
References
8- SSRT rank agreement is lower with recorded delays in two splits of 102 ds000030 controls sci:sha256:24a9839d28305476eb83ac828e48e86be3f108a54649b4d9aeec053b1c975882
[Legacy relation: reproduces] A new calculation gives the same full-run and half-run table fields and all 12 reported Spearman point values. It uses the same 102-person screen and original split rules. It passes 12,738 reference checks at tolerance 1e-8. It does not reproduce the original bootstrap random stream. New whole staircase and random splits extend the tested scope.
- usesA second calculation gives the reported delay results for 126 ds000030 control runs. sci:sha256:2848d4857a730d5b854ac6adfa3d0a967d68b9ff7e4977f4409bbe91aebe05ab
We used its pinned source URLs, expected checksums, and 126 control event tables as initial inputs. This information enabled fresh reads of all 130 public source files and comparison with the split report files. We read but did not execute its code.
- usesdoi:10.18112/openneuro.ds000030.v1.0.0
The analysis uses event files from OpenNeuro ds000030 snapshot 1.0.0.
- doi:10.3758/s13423-021-01948-3
- doi:10.3389/fpsyg.2012.00037
- doi:10.7554/eLife.46323
- doi:10.1038/sdata.2016.110
- doi:10.7554/eLife.88652
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesThe zero-entry contribution increases mean squared SSRT differences in four splits of 102 ds000030 controls.
Used its two whole-staircase split rules and reference half-score table. They enabled the fixed-split cross-term comparison, including a reversal of GO assignment. Its random splits and bootstrap streams were not repeated.
- usesIn 102 ds000030 controls, three of four task splits have maximum rank correlations more than the two zero-delay endpoint values.
This calculation uses the 102-control screen, four fixed splits, source hashes, and two reference score tables from this report.
- usesA selected delay coefficient gives lower mean test rank agreement in four splits of 102 ds000030 controls.
The full-run sample screens and endpoint definitions fixed the 102-participant sample and the reference calculation.