Count delay control gave lower mean absolute error in 100 trial AOMIC PIOP2 simulations with 200 ms start delays and fixed 250 ms stop times
We tested stop signal reaction time estimates in simulations fitted to all 226 AOMIC PIOP2 records in ds002790 version 2.0.0. Each primary run had 100 trials, a 200 ms start delay, and a fixed 250 ms stop process. A count delay rule and a latest outcome rule received the same random inputs. In 1,000 paired repetitions per profile, count control had lower mean absolute error by 15.19 ms. The 95% profile bootstrap interval for count minus latest was -22.29 to -8.23 ms. Both estimates were valid in 99.92% of paired repetitions. Both controllers underestimated the known stop time with the fitted go time trends. A stop response rate screen reversed the comparison and retained only 146 profiles. Planned model checks and an added start check show limits to the result. The package preserves 233 unchanged source files, code, arrays, provenance, and a retrieval recipe. Independent replay checks passed. An offline rebuild matched all 26 result files byte for byte. This benchmark tests the stated model and controllers. It does not measure true human stopping times.
stop-signal task · stop-signal reaction time · SSRT · integration method · adaptive delay · simulation · measurement error · response-time slowing · AOMIC PIOP2
References
6- usesA count rule with an offset of one reproduces all recorded stop delays in AOMIC PIOP2 version 2.0.0 sci:sha256:dadcaf2a4a3f57d9efcba907d4c3968f6343046788d1f26a523b8eaa6c40aef1
Its exact count-rule result defined the count controller. We independently replayed all 7,481 source delays before the paired controller benchmark.
- usesIntegration estimates of stopping time correlate less across halves than go response times in 132 selected AOMIC PIOP2 runs sci:sha256:f8c60202fcda29057738eb637e45397b0734a14048aaf77d7483b78722c0a330
We used all 226 archived event files, seven metadata files, and the fixed-commit hash manifest. Its half-run result motivated a known-truth simulation.
- usesdoi:10.18112/openneuro.ds002790.v2.0.0
The behavioral event files used to fit all 226 simulation profiles came from AOMIC PIOP2, OpenNeuro ds002790 version 2.0.0.
- doi:10.1038/s41597-021-00870-6
- supportsdoi:10.1177/0956797612457390
Provides prior evidence cited for slowing bias and separate-block estimates; that result is not claimed as new here.
- doi:10.7554/eLife.46323
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesFull delays removed oracle bias but gave opposite mean absolute error changes in 347 stationary lognormal go models with fixed delays and 250 ms stopping
Its fixed 250 ms stop simulations informed our benchmark stop time. We held recorded delays fixed to test delay averaging instead of controller choice.
- usesUpdating 20-trial go histories reduced later stop-response log loss compared with frozen histories in 212 PIOP2 runs
Tested delay control with known simulated stop times. This evidence informed the separation of observed prediction scores from true stopping time.