Past-stop counts give negative expected future-stop mean contrasts under independent Bernoulli stops with a common probability, both groups present, and at least three scored trials
We prove a bias in a future-stop score. With independent Bernoulli stop assignments and one common probability, a cumulative past-stop count has an expected future-stop mean contrast of -(T-2)/(2T). This result requires at least three scored trials and both next-stop groups. A score with a known stop probability has expectation zero under the stated assumptions when all runs are included. We also test the original adjusted regression in 220 AOMIC PIOP2 recorded designs. The past-only count signal gives a mean future-stop coefficient of -0.458397 units, with a 95% participant bootstrap interval of [-0.486629, -0.430889]. There are 218 negative coefficients. No signal parameter is fitted to human responses. The planned next-stop prediction gain is 0.000718 nats, with a refit interval that includes zero. The human future-stop association remains negative after added history controls. The proof and benchmark do not establish its cause. The package preserves source events, fixed retrieval URLs and hashes, plans, exact checks, all fitted results, uncertainty draws, code, figures, and reproduction instructions.
stop-signal task · cumulative history · future-event diagnostic · Bernoulli assignments · participant bootstrap · response time · prediction · regression
References
3- usesCorrect go responses were 65 ms slower after stops under time and face controls in 220 AOMIC PIOP2 runs sci:sha256:b9d2889e73ab3f41c94e9fd528758167f9d6815b5ffba6732f84c68f402ba57b
We use its source event subset, retrieval manifest and code, response model, and participant estimates. We reproduce its future-stop diagnostic coefficients. Its reported future-stop association set the initial history-test question.
- usesdoi:10.18112/openneuro.ds002790.v2.0.0
The analysis uses the stop-signal event tables from OpenNeuro ds002790 version 2.0.0; the package preserves selected source bytes, fixed commit, retrieval URLs, and hashes.
- doi:10.1038/s41597-021-00870-6
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesA lookup from 20 initial trials predicted 97.7% of later stop/go assignments in 113 AOMIC PIOP2 test records
Its restricted recent-stop forecast and assignment-law limits motivated a different test with complete training schedules and 20 initial test assignments. Our result has different inputs and scope. It does not contradict that restricted forecast result or test its theoretical proof.
- usesPredictable integrable signals give zero expected future-stop scores with rates from other independent Bernoulli runs, a common probability, fixed denominators, and retained runs
We used its proof of group mean bias and its known-probability score. We extend the score to rates from other independent runs with an unknown common probability.
- usesA past stop pair score rejected a model of independent stops with common trial rates in 226 AOMIC PIOP2 v2.0.0 runs
Its exact past-count contrast bias and known-probability score motivated a score that uses fixed pairs without a known probability. We read its full description before the local plan. The package gives a new conditional-symmetry proof, benchmark, and public assignment-model audit.
- usesA test martingale at threshold 20 bounds false rejection by 5% under a known conditional stop law and predictable response scores
This result gives the motivating past-stop count contrast and a known-probability score with zero expectation. We use these prior results to construct a product test with a bound for repeated inspection.