Original scan gaps reduced mean null task rejection rates with grid REML in 431 AOMIC designs under Gaussian AR(1) noise at rho 0.6

We tested nominal 5% task tests in 431 fixed public AOMIC working memory designs. We simulated 5,000 stationary Gaussian AR(1) null responses per design at rho 0.6 per original scan step. Noise was independent of the fixed design and motion mask. Both estimators fitted correlation by REML on a fixed grid and estimated residual scale. Keeping original scan gaps reduced mean rejection rates from 6.5761% to 5.6950% in PIOP1 and from 6.1986% to 5.6367% in PIOP2. The paired reductions were 0.8812 and 0.5619 percentage points. Their 95% Monte Carlo intervals were [0.8547, 0.9076] and [0.5401, 0.5837]. The scan-gap rates still exceeded 5%. Matched known-correlation controls had rates of 4.9864% and 4.9762%. Fixed checks used rho 0 and 0.8 and a finer grid in 24 selected designs. Grid refinement did not remove the excess. All 431 designs passed independent dense covariance checks. A complete offline repeat matched all core result files byte for byte. The package preserves code, public inputs, provenance, seeds, simulations, tables, figures, and repeat instructions. The intervals measure simulation error for these fixed designs. No measured brain image signal was analyzed. The result applies to the stated noise model, removal rule, contrast, fitting grids, and test rule.

task fMRI · AR(1) noise · restricted maximum likelihood · scan gaps · null rejection rate · Monte Carlo simulation · covariance estimation