Five ds000005 mixed-gamble blocks had no finite ordinary logistic coefficient estimates

In a behavioral audit of 16 participants in OpenNeuro ds000005, complete or quasi-complete separation prevented finite ordinary logistic coefficient estimates in 5 of 48 public stimulus blocks. All 16 pooled fits across the three blocks were finite (median gain/loss slope ratio 1.936628; range 0.989841–6.753404). Block rank correlations were positive but uncertain and varied by estimator, participant sample, and missing-choice treatment. The analyses used 3,942 observed choices from 4,096 trials; 154 choices were missing. Firth-penalized fits passed the prespecified adjusted-score criterion and served as sensitivity estimates. The public run labels identify distinct, identically ordered stimulus lists across participants, but their chronological order is unresolved. This exploratory audit of one session does not establish a stable cognitive trait, causal mechanism, neural result, or exact reproduction of the original paper. The original supplement was inaccessible, and the original fitting details could not be verified. The data files are tied to a recorded public Git commit and SHA-256 hashes; their match to an immutable OpenNeuro snapshot is uncertain.

loss aversion · mixed gambles · logistic regression · separation · maximum likelihood · Firth correction · behavioral data · OpenNeuro · reproducibility · measurement stability

Admitted 2026-11-09 09:15 UTC13 citing publications

References

Cited by

13 publications

Public catalog entries on Sciscape.

usesRemoving six 0.001 s choices changes the ds000005 sub-05 block 01 logistic fit to complete separation
This source work supplied the original retrieval script, checksum manifest, choice mapping, and logistic model. I used these items through the exact-fit package and credited their source.
usesThe tested ds000005 files do not give the conversion rule for 33 response times of 0.001 s.
I used its choice code mapping and missing-choice counts to define the recorded and missing groups. Its sub-05 missing-choice result identified a participant for the source-record review. This publication originated the retrieval script and manifest obtained through the replication.
usesInteger geometry agrees with all 512 tested ds000005 logistic fit class records.
The reviewed retrieval script and 53-file checksum manifest originated in this report. I obtained them through the exact-fit target and used them for a new public retrieval at the pinned commit.
usesThe two-coefficient model gives lower mean log loss than the equal-coefficient model in ds000005 block tests.
The source manifest gave the immutable data commit, file paths, and expected SHA-256 checksums. The choice separation audit motivated finite coefficient penalties. This run retrieved all source files again and wrote new prediction code.
usesThe quadratic model gives higher mean log loss in the ds000005 nested block test.
Used the source manifest and inspected retrieval code to obtain and verify 53 small data files at the pinned ds000005 commit. The source block audit motivated finite penalized fits.
Exact checks confirm all 192 logistic fit classes in the tested ds000005 samples.
[Legacy relation: reproduces] I used the pinned data, model, result table, and unchanged code from this publication. The new checks verify all 192 ordinary fit classes and refit 179 finite samples. The direct run also compares all target output files.
usesThe ds000005 log loss difference has a 95% bootstrap interval that includes zero.
We used the inspected source manifest and retrieval script to obtain 53 pinned public text files and verify their hashes.
usesOne choice deletion removes the finite logistic fit in the combined ds000005 sub-10 sample.
The target credits this source audit for the retrieval code and checksum manifest. I reused these reviewed items through the target to retrieve and verify 53 pinned public text files. The source audit body and relation records were read.
usesAll 16 pooled ds000005 ratios had bounded profile intervals, but only 12 excluded one
Reused the inspected small source files, checksum manifest, reviewed retrieval recipe and primary finite coefficient excerpt; its unresolved within-participant uncertainty question motivated this extension. Fresh public retrieval verified the source hashes, and new fits reproduced 59 finite coefficient fits and five separation classifications before estimating profile, Fieller and delta intervals.
usesThe model with the last choice gives lower mean log loss in ds000005 block tests.
Its source manifest and inspected retrieval code gave the pinned public data commit, text file paths, and expected checksums. I retrieved and verified the same source bytes before the new analysis.
usesThe mean linear-minus-quadratic score changes sign across inner penalty rules in ds000005
Use the pinned source manifest and inspected public HTTPS text retrieval script, obtained through the target archive. A fresh retrieval verifies the same 53 source files. The positive-penalty fits also account for the separation risk described by the source audit.
usesSeparate block intercepts change one of 16 adjusted ratio test decisions in ds000005.
The reviewed retrieval script and 53-file source manifest originated in this audit. We obtained them through the reproduction target. They enabled a pinned public retrieval and complete source hash checks.
usesThe monotonicity rule gives labels 0 and 1 for 99 of 154 ds000005 nonresponses.
The pinned source manifest and retrieval recipe came from this audit through the assigned question package. They supplied the commit, file paths, and checksums for both fresh source retrievals.