The later nine-digit recall error difference from added pupil data had a 95% interval that included zero in 27 ds003838 participants with early recall calibration

We tested whether pupil diameter improved forecasts of later nine-digit recall. We used metadata from OpenNeuro ds003838 version 1.0.6 and preserved raw file versions. The saved main test included 27 people and 279 later trials. Both models used early personal recall and a shared block trend. One model also used the pupil change before spoken recall. The mean squared errors were 3.8587 and 3.8790 digits squared. The error difference was -0.0204, with 95% limits from -0.1900 to 0.0696. A positive difference favors pupil data. The interval included zero. This test did not establish a decrease in forecast error. It does not prove that pupil data have no value. Shared models used early trials from other people. The test used blocks 7 to 9. The package preserves source files, pupil bins, forecasts, bootstrap results, code, source hashes, and raw retrieval instructions. The result applies to this cohort, pupil rule, and model. It is not a clinical or causal result.

pupillometry · working memory · digit span · recall forecasting · out-of-sample prediction · personal calibration · OpenNeuro ds003838