The ds000005 models for 3,700 trials include no latent criterion state.
In 16 participants from OpenNeuro ds000005, four logistic models were compared on 3,700 trials with recorded responses at the center and both adjacent trials. The models included current and adjacent gamble values, block position, and different combinations of the center participant's past and future responses. Mean log loss for the past-response model was 0.2559322983523329 nats per trial; for the future-response model it was 0.26329420112272073. The future-minus-past difference was 0.00736190277038785 nats per trial (pointwise 95% participant-bootstrap interval 0.0028785078162980615 to 0.012177521417114288); 12 of 16 participants had a positive difference. The future response is unavailable at choice time, so this is a retrospective score comparison. These models do not include a latent criterion state. The result does not identify a direct choice-update cause, and the run did not test model recovery or fit drift-only, update-only, or combined mechanism models. The analyzed event files are pinned by source URL, Git commit, and SHA-256 checksums; the exact correspondence to a version-specific dataset DOI is not established.
choice history · criterion drift · risky choice · logistic regression · log loss · model comparison · model recovery · OpenNeuro · ds000005
References
3- The last response decreases mean log loss with response lags 2, 4, and 8 in ds000005. sci:sha256:431956868d514aa58dd34786660ece4a7f1394b14d20bdc907588b88d19fc9dd
[Legacy relation: reproduces] The new code calculates the primary older-versus-all comparison on the same 3,319 eligible trials. All 32 reference participant scores agree within 4.661826480401032e-12 nats. The reference manifest and model specification enabled this check.
- supportsdoi:10.1126/science.1134239
The cited paper and its supporting methods describe the original mixed-gamble task used to interpret the recorded event tables.
- doi:10.1038/s41598-020-66502-y
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesIn this test, the criterion drift method gives two score differences more than their recorded ds000005 values.
Used the four-model specification, inspected code, source manifest, and reference scores to compare simulated and recorded score differences.
- usesThe log loss difference changes sign with the penalty scale in the ds000005 two-state model tests.
We used its question about the missing latent criterion and short blocks to choose the state model and artificial-data checks.