Two 95% intervals for prediction gain from target angle terms included zero in 61 ds008037 v1.0.1 adults under circular mixtures

We tested whether target angle terms improve orientation report prediction in separate people. The source was OpenNeuro ds008037 version 1.0.1. Training used 60 adults. Testing used 61 other adults. Each person supplied 60 reports at set sizes 2 and 4. The reference used two von Mises densities plus uniform error. The added model changed the mean and concentrations with target angle. The mean test log loss gain was 0.00460 nats per report. The conditional 95% interval was -0.00340 to 0.01252. The 95% interval with fitting effects was -0.00903 to 0.01317. Both included zero. The result did not meet the saved rule for positive evidence. It does not establish no angle effect. All five planned secondary overall intervals also included zero. All 250 source files matched a fresh download. A separate calculation checked all trial scores and all bootstrap draws. A full repeat gave 32 analysis files with identical bytes. The package preserves data, code, a local plan, source hashes, fits, scores, draw indices, figures, and repeat instructions. The result applies to the stated models and fixed split in this cohort. It gives no evidence for a brain mechanism.

visual working memory · orientation report · target orientation · circular mixture model · prediction gain · von Mises distribution · bootstrap interval · OpenNeuro ds008037