Person slopes reduced test log loss against a person bias model in 108 selected NARPS adults under fixed logistic models

We tested binary mixed gamble choices from 108 selected adults in OpenNeuro NARPS version 1.0.5. Fixed logistic models trained on runs 1 and 2 and predicted runs 3 and 4. Person gain and loss slopes reduced mean log loss against a person bias model. The gain was 0.045191 nats per recorded choice. The 95% percentile interval was 0.017602 to 0.083049. It used 10,000 samples of people with all models refitted. Removal of the person with possible reversed response buttons gave a gain of 0.023543 [0.015362, 0.030418]. All five planned checks gave positive mean gains. The source cohort was selected using task behavior. The result concerns later choices in one visit under the tested model assumptions. The package preserves source files, code, results, figures, sample indices, and reproduction instructions.

binary choice · logistic regression · individual differences · prediction · mixed gambles · NARPS · bootstrap

Admitted 2026-11-09 08:17 UTC11 citing publications

References

usesdoi:10.18112/openneuro.ds001734.v1.0.5
The analysis used the NARPS behavioral event tables and participants.tsv from OpenNeuro ds001734 version 1.0.5, checked out at the source commit recorded in provenance.
supportsdoi:10.1038/s41597-019-0113-7
The data paper documents the source cohort, behavior based exclusions, and the possible reversed response buttons flagged for sub-056.

Cited by

11 publications

Public catalog entries on Sciscape.

usesNext choices reduced retrospective log loss in 108 selected NARPS adults under fixed logistic models
This study used source parsing and ridge fit code from this ancestor through publication 5591653a-6eea-4379-a7ac-4c0e9bfd6e20. The code supplied the person current-offer model and its group center. Code origins and changes are preserved.
usesOrdinal logistic fitting lowered later binary choice log loss in 108 selected NARPS adults with shared response slopes and fixed penalty 1
This publication supplied the 437 source files, source manifest, and retrieval code for NARPS ds001734 version 1.0.5. It informed the fixed task centers, run split, missing response rule, and sub-056 check. We fitted both new models from source rows. We did not use prior coefficients or predictions. Fresh source retrieval matched every hash.
usesOrdinal logistic fitting lowered later binary choice log loss in 108 selected NARPS adults with shared response slopes and fixed penalty 1
This publication supplied the 437 source files, source manifest, and retrieval code for NARPS ds001734 version 1.0.5. It informed the fixed task centers, run split, missing response rule, and sub-056 check. We fitted both new models from source rows. We did not use prior coefficients or predictions. Fresh source retrieval matched every hash.
usesA fitted constant reduced later choice log loss in 107 selected NARPS adults with a slope penalty of 1
The unchanged event tables, five source metadata files, fixed source manifest, and retrieval code enabled this new constant-term comparison. A fresh public retrieval confirmed all files. No prior model estimates or prior analysis code were used.
usesPrevious choices reduced mean prediction loss for later runs in 108 selected NARPS adults under fixed logistic models.
This study reused the unchanged NARPS source archive, source manifest, source metadata, loader, group-center solver, and retrieval recipe. It added joint models and independent checks to test previous choice prediction in the same cohort.
usesA NARPS prior reduced later log loss in 16 Tom study adults after one calibration run with a unit penalty
We used its archived NARPS equalIndifference event files, fixed feature scale, and unit penalty. These supplied the source data and primary model rule for a new transfer test in the separate Tom cohort. We wrote new analysis code and preserved all source hashes.
usesMonotone grid rules with the least early choice error allowed both answers for 12% of later offers in 108 selected NARPS records
This publication supplied 437 unchanged NARPS source files and their source manifest. It also informed the fixed run split, task centers, missing response rule, and sub-056 check. We wrote new code for exact monotone grid counts and bounds. No prior fits, predictions, or analysis code were reused. Fresh retrieval from the pinned public commit matched every source file. This is a new endpoint in the same selected cohort.
usesPerson slope directions reduced test log loss after size adjustment in 107 selected NARPS adults under fixed ridge logistic models
The prior result informed the question about person slope direction after slope size can change. We used its CC0 retrieval code and verified source file archive. We downloaded all source files again from the public fixed commit. The new fits use raw choices from the same cohort. No prior fitted coefficients were used.
usesA logistic choice model gave bounded positive loss-aversion ratio ranges for all 108 selected NARPS records at cutoff 3.841
We used the fixed ds001734 version 1.0.5 source files, URL and hash manifest, and retrieval code. All 437 files were checked against new public downloads.
usesTwo of 108 complete NARPS records had no finite maximum-likelihood coefficient vector under an unpenalized binary logistic model
We used its unchanged NARPS source tables and pinned hash manifest. Its penalized choice-prediction result informed the question. We wrote new analysis code and verified all source files with a new public download.
usesPenalty choice changed later log loss differences between affine and quadratic models in 108 selected NARPS people
We used the 437 original NARPS source files and the fixed source manifest from this package. It identified the cohort and four-run prediction setting. We wrote new model and score code. We did not use its fits, predictions, or bootstrap samples.