All 16 pooled ds000005 ratios had bounded profile intervals, but only 12 excluded one
In 16 participants from one session of OpenNeuro ds000005, free-intercept logistic fits to 3,942 observed mixed-gamble choices produced finite pooled loss/gain slope ratios for all participants. All 16 pointwise 95% asymptotic profile-likelihood intervals were bounded; 15 point estimates exceeded one, 12 intervals lay wholly above one, and 11 two-sided likelihood-ratio tests remained significant after Holm correction. In exploratory stimulus-block fits, five blocks had complete or quasi-separation; of the remaining 43 finite fits, 42 profile sets were bounded. One highly incomplete block had an unbounded profile set, and two blocks had bounded profile intervals but all-real covariance-based Fieller sets. These are model-conditional results from one cohort and session. The analysis assumes independent Bernoulli choices under a constant logistic model; missingness, serial dependence, small blocks, and likelihood shape may limit interpretation. No neural or causal claim, coverage calibration, or independent cohort verification was made.
loss aversion · mixed gambles · logistic regression · profile likelihood · ratio estimation · Fieller interval · individual differences · OpenNeuro ds000005
References
4- usesFive ds000005 mixed-gamble blocks had no finite ordinary logistic coefficient estimates sci:sha256:91c6fdc6f992b1e2c6a66611dbde4774f4dcfb141e2049c2f8a3e754dc9eb16e
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.
- doi:10.1126/science.1134239
- doi:10.1093/scan%2Fnsp011
- doi:10.1037/dec0000141
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesThe simulation method measures 92.41% to 94.90% coverage for 16 ds000005 ratio models.
Its profile ratio constraint and chi-square-1 cutoff define the confidence-set procedure tested in the fixed-design simulations.
- usesThe score test rejects equal block slopes for two of 16 ds000005 participants under independent choices.
I used its source manifest, retrieval code, and pooled estimates. They fixed the data version and enabled comparison with a test of common block slopes.
- A second calculation gives the reported confidence sets for 59 ds000005 samples.
[Legacy relation: reproduces] The new numerical code repeats all 59 finite fits and confidence sets, 64 sample counts and separation labels, and 16 Holm-adjusted tests. It uses trust-exact fits, a different separation LP, and a convex likelihood bracket instead of the target optimizer and endpoint grid. The target provides the model, reported comparison values, source manifest, text files, and reviewed retrieval script.
- One choice deletion removes the finite logistic fit in the combined ds000005 sub-10 sample.
[Legacy relation: reproduces] The new code repeats all 64 sample counts and separation labels, all 59 finite coefficient fits, and 236 reported likelihood statistics and finite profile boundaries. The target supplies the model, comparison table, and source retrieval recipe. The new analysis then tests every single recorded-choice deletion.
- Separate block intercepts change one of 16 adjusted ratio test decisions in ds000005.
[Legacy relation: reproduces] We repeat 64 sample classifications, 59 finite coefficient fits, 236 likelihood-ratio checks, and 16 combined-sample profile intervals and adjusted p values. We then test separate block intercepts on the same choices. We do not repeat the Fieller intervals.