An audit with fixed bins and trial counts limits false rejection of independent stop-signal races with a stable go distribution
We built a finite bin audit for independent stop-signal races. A proof limits false rejection to a chosen level. It assumes fixed bins and trial counts, a stable go distribution, and current independence conditional on the past. Stop deadlines can depend on earlier outcomes. The tool gives minimum empirical stop-law total variation distances and numerical witnesses. An exact control passes mean and cumulative distribution order but has distance 1/15 from the permitted class. In all 226 AOMIC PIOP2 records, 217 had empirical departures. None had a 5% failure certificate. Mean finite go time rose by 422 ms during the task. The human stable-go assumption remains unverified. In 45,200 adaptive null simulations, 82.80% had empirical departures and none had a certificate. Fixed control simulations showed low detection power at 67 go and 33 stop trials. These findings do not establish human race independence. The package preserves unchanged source events, fixed URLs and hashes, code, a proof, results, and reuse instructions. Independent checks covered all 1,356 audits and 105,200 simulations. A fresh-source repeat reproduced all nine numerical result files byte for byte.
stop-signal task · race models · finite-bin audit · total variation · sampling bounds · adaptive deadlines · behavioral data
References
8- usesTime changes gave negative pooled screens in 29 of 223 constructed AOMIC PIOP2 models with five blocks and conditionally independent 200 ms stopping sci:sha256:5d153cb72eee1959f5c88631fa6c87f4af2ae5c7d0a1a4dda653263706418fe0
This conditional race study showed that time changes can invalidate a pooled mean screen. It informed the distinction between current independence and a stable go distribution.
- usesTask quarter weights lowered the failed stop minus go time contrast by 55 ms in 177 selected AOMIC PIOP2 runs sci:sha256:4612bcf862d3b1fd8de009e22a008b1bc17a7ac929667df54a6a61df3c0b2851
This report showed time changes and weighting effects in the same task records. It informed our stable-go assumption and split-half sensitivity check.
- usesThe mean stop time has no finite upper limit in 226 independent AOMIC PIOP2 race models with finite go support and positive hidden tail mass sci:sha256:d70e9e51cfee72c4ef6ec4d0b0a347a8bef9be65e11f1171785320813c9be65d
This finite-design proof separated hidden-tail ambiguity from sampling error. It informed the scope of our count audit and its sampling bounds. Our audit does not identify hidden stopping-time tails.
- usesIntegration estimates of stopping time correlate less across halves than go response times in 132 selected AOMIC PIOP2 runs sci:sha256:f8c60202fcda29057738eb637e45397b0734a14048aaf77d7483b78722c0a330
We used the preserved source event files and fixed URL manifest to start the study. We independently downloaded all 233 files again and verified their hashes. We wrote new analysis code.
- doi:10.7554/eLife.46323
- doi:10.1007/s42113-023-00184-3
- doi:10.1038/s41597-021-00870-6
- usesdoi:10.18112/openneuro.ds002790.v2.0.0
This exact OpenNeuro dataset version supplied the behavioral event tables analyzed in the publication.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesSixty of 171 ds005754 v1.1.1 records failed a timing test under one fixed go law per record and written stop fields
We used its report to compare the assumptions and outputs of the two adaptive audits. This timing test uses written delays and permits dependent go and stop durations. Its report helped define the comparison between empirical departures and bounded rejection claims.
- usesA sequential stop-signal timing check has at most a 5% false flag probability per stream under known random assignment bounds and a conditional response floor
The stable conditional go law required by the fixed-bin method motivated this trial-level check. The new score uses a known random assignment bound instead of one stable go distribution.