The two calculations agree within 2.3e-9 for all 47 finite ds000005 score statistics.

A separate implementation reconstructs the block design and score statistics from OpenNeuro ds000005 for 16 participants. The dataset contains 4,096 trials, 3,942 recorded choices, and 154 missing choices. All 48 participant/block-pair comparisons have zero shared gain/loss pairs. The 47 finite score statistics agree with the reference within 2.257158016050198e-09; the sub-10 full quadratic fit has quasi separation and no finite score statistic. Only sub-07 and sub-09 have Holm-adjusted p-values below 0.05 in the linear model family. Neither quadratic family has an adjusted p-value below 0.05. A conditional simulation of 649,953 independent choices from fitted null probabilities gives 39,202 undefined draws, including 13 numerical failures. The results reproduce arithmetic and show a model-dependent decision pattern. They do not establish equal slopes or explain the linear decisions. The analysis assumes independent choices and correct fitted null probabilities; it does not compare statistical power, assess brain measurements, or resolve dependence, choice selection, or block chronology.

OpenNeuro · ds000005 · logistic regression · score test · block effects · quasi separation · Monte Carlo simulation · reproducibility