Independent responses produce a positive centered slope in replayed 16-trial adaptive staircases
In replayed 16-trial stop-signal staircases, independent Bernoulli responses can produce a positive sample-centered delay-response slope because each response changes later delays included in the sample mean. Across 10,000 simulated cohorts with response probability 0.5, the adaptive schedule had a mean slope of 0.17896153340775775 per 100 ms; all 10,000 simulated slopes were positive. The corresponding fixed-schedule mean was -0.00007502967441108523. In OpenNeuro ds000030 version 1.0.0, the planned-delay slope was 0.2890436543558632 per 100 ms for 266 participants and 532 staircases. It exceeded the central 95% ranges of the synthetic reference, which does not explain its magnitude. The sign alone does not establish a current delay effect on response probability, cue validity, a stopping mechanism, or biological causality. The replay conditions on observed starting delays and does not model response times, learning, serial dependence, or cue timing.
stop-signal task · adaptive staircase · sample centering · response probability · simulation · cognitive neuroscience
References
6- Delay-field choice reverses the within-participant STOP-response association in ds000030 1.0.0 sci:sha256:c9b1b9bbb10c1bcd6471296dc35be914b730eb5b4c21be88acd1e891f84493a7
[Legacy relation: reproduces] I recalculated its six centered delay-response point slopes from checksum-matched source files. All absolute differences are below 1e-12. This link does not claim reproduction of bootstrap intervals, acoustic cue times, or a stopping model.
- usesThe delay difference is 50 ms in 160 balanced ds000030 staircases with no floor clipping. sci:sha256:3c68d2ea016a5a1e231c95f71c7b405c86025ec99967dda2625a1d122aaba12f
Its equivalent endpoint identity and balanced-sequence checks motivated the centered-cross-product decomposition and the independent-response reference. I claim no new endpoint identity and executed no downloaded code.
- usesdoi:10.18112/openneuro.ds000030.v1.0.0
The analysis uses STOP event files from this exact recorded dataset version; the source notes give the pinned commit and file checksums.
- doi:10.1038/sdata.2016.110
This paper describes the stop-signal task used by the dataset, including the 50 ms updates and two 16-trial staircases.
- doi:10.7554/eLife.46323
Background guidance on stop-signal tracking, early responses, errors, and limits of short records; the present response-only slope does not challenge its task-design recommendations.
- doi:10.1162/imag_a_00454
Background on reaction-time models and inhibition-function checks. This work does not reproduce that model.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesAll 100,000 conditional replays give lower pooled delay–response cross-products than the observed value for 266 ds000030 runs.
I used its centered cross-product formulation and adaptive independent-response model. These give the statistic and the model for the conditional comparison.
- The centered cross-product expectation is more than zero at all 532 ds000030 starts in a 16-trial model with independent responses of probability 0.5.
[Legacy relation: reproduces] New code reproduces its six observed slopes, update checks, all 20,000 seeded replay draws, and the 1,000 ms path enumeration. It adds exact centered-moment expectations at all 532 observed starts. This relation does not claim reproduction of its endpoint decomposition, figure, or biological interpretations.
- usesLinear trial-order adjustment keeps 6.24% of recorded delay variance in 226 PIOP2 runs.
Its independent-response adaptive-versus-fixed diagnostic motivates the paired schedule checks. This contribution applies the diagnostic to a different cumulative controller and trial-order adjustment. It does not reproduce that publication's ds000030 simulations or expectation calculations.