Early FCz voltage reduced mean error classification loss in 33 ERP CORE adults under person and time splits with a 150 microvolt artifact limit
We tested whether early FCz voltage adds error information beyond response time and specified task variables. The data were public ERP CORE flanker records from NEMAR version 1.1.1. The test used 3,714 later trials from 33 adults, with 388 errors. Each person was absent from the fitting and standardization used to score that person. The main artifact limit was 150 microvolts. The models used the same trials and a fixed L2 penalty. The added feature was mean FCz voltage from 0 to 100 ms after response, minus its mean from -400 to -200 ms. Mean log loss fell from 0.280908 to 0.261003 nats per trial. This was a 7.09 percent reduction from the behavioral base model. The gain was 0.019906 nats per trial. Its 95 percent person bootstrap interval was [0.002280, 0.040040], with model refits in 1,999 draws. The early-voltage gain stayed positive in all fixed checks. Baseline voltage alone increased loss. ROC area changed only from 0.872299 to 0.874397 in 30 people with both labels. Target direction was withheld because the full task record directly gives error labels. The result applies to retrospective classification under the specified person and time splits. Future samples enter filtering and artifact selection. The package preserves audited inputs, source provenance, code, split rules, predictions, model parameters, bootstrap draws, figures, and reproduction instructions. Independent fits passed checks. A full repeat matched all 37 numerical hashes. Fresh source extraction reproduced all 40 FCz files byte for byte.
EEG · error-related negativity · ERP CORE · logistic regression · log loss · person-level validation
References
5- usesMatching response time made the mean 0–100 ms FCz error contrast less negative in 25 ERP CORE adults with a 150 µV artifact limit sci:sha256:d6a13dfe3872f874482d99e2a2766bbd91908e9a8a9254422a457e6a83c08853
This publication supplied the archived FCz epochs, source files, and provenance copied unchanged into input/. Those files are the inputs to this error classification comparison.
- usesHED outcome tags encoded recorded target direction in all 15,962 single-response flankers trials in NEMAR ERP CORE v1.1.1 sci:sha256:9aeb2b661f8b5ba4c8074c4bd41221340e93ead507e45b147177ee591b96c99c
This publication supplied the event audit, hand-rule labels, signal extraction code, and pinned source retrieval records. We used its label rules and independently repeated its signal extraction for all 40 people.
- usesdoi:10.82901/nemar.nm000132
The analysis used the NEMAR ERP CORE nm000132 version 1.1.1 flanker signals and event records.
- doi:10.1016/j.neuroimage.2020.117465
Background paper describing the ERP CORE resource and cohort.
- doi:10.31234/osf.io%2F4azqm
ERP CORE source credit cited in the archived dataset metadata.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesAdded FCz power gave an uncertain error classification gain in 33 ERP CORE adults under fixed comparison models
We used the weighted logistic fit and score code. We kept the person folds and extended the bootstrap framework with spectral features. The new base model adds squared voltage and broad power. This study uses the same cohort and is not an independent replication.
- usesLater Pz Voltage Reduced Mean Loss, but Its 95% Interval Included Zero in 33 ERP CORE Adults Under Person and Time Splits
This study uses its preserved FCz epochs, source files, trial ledger, fixed person folds, and numerical logistic solver. It tests the extra value of later Pz voltage beyond the earlier FCz model.
- usesThe 95% interval included zero for the FCz and Pz gain in calibrated estimates of later correct response time in 32 ERP CORE adults
This study reuses the source files, trial ledger, pinned signal recipe, and FCz epochs from the input folder. New extraction replayed all ten FCz arrays for each of 40 people with exact equality.