Errors preceded higher mean recorded accuracy and slower correct responses across three ds004883 v1.0.1 flanker tasks in 121 selected people

We tested accuracy and correct response time after recorded errors in OpenNeuro ds004883 version 1.0.1. The primary selection retained 121 people with all three flanker tasks. The model controlled previous and current congruency, recording segment, and five bins defined on all trials. Pairs required adjacent stimuli and positive onset gaps of at most three seconds within the same recording segment. Mean adjusted accuracy increased by 1.16 percentage points after errors. Correct responses were 10.06 ms slower. The 97.5% primary intervals were 0.41 to 1.90 points and 7.35 to 12.66 ms. These two intervals use nominal 95% joint coverage under the stated bootstrap assumptions. The accuracy result does not establish an increase in each task. All eight later exploratory models retained positive estimates, with accuracy sizes from 1.16 to 2.72 points. These models held people fixed and changed time bins, cell terms, and the pause limit. A related AutoSci report uses the same cohort. The package preserves sources, code, plans, results, bootstrap draws, and checks. A complete fresh source retrieval and two repeated runs matched all 19 numerical output hashes. The association does not show that errors caused better control.

post-error accuracy · post-error slowing · flanker task · response time · trial-level analysis · OpenNeuro ds004883