Working memory accuracy did not reduce Raven prediction error in the AOMIC PIOP1 to PIOP2 linear model test

We tested whether short visual working memory accuracy improved Raven score prediction in a separate public cohort. We used OpenNeuro AOMIC PIOP1 and PIOP2, both version 2.0.0. A fixed linear model trained in 200 PIOP1 participants was tested in 222 PIOP2 participants. The reference terms were age, recorded sex, and passive-trial response rate. Adding active accuracy changed test RMSE from 4.963 to 4.995 Raven points. The gain was -0.031 points, with a 95% interval of -0.469 to 0.241. The interval used 10,000 bootstrap draws with model refitting in the sampled training cohort. It permits both benefit and harm. The reverse test and four sensitivity checks also had gain intervals that included zero. A within-cohort association in PIOP1 did not establish prediction gain in PIOP2. The package preserves all source files, scoring audits, predictions, sample counts, code, and reproduction instructions. Independent scoring and QR checks passed. A second offline run matched all 27 numerical output hashes. The result applies to these cohorts and this fixed linear procedure. It does not show that memory and reasoning are unrelated.

working memory · Raven matrices · prediction · cohort transfer · AOMIC · OpenNeuro · behavioral data