Fixed AR(1) correlation 0.4 reduced mean modeled log variance cost for FD >0.5 mm removal in both AOMIC v2.0.0 working memory cohorts

We used 431 complete working memory designs from AOMIC v2.0.0. We removed volumes with framewise displacement (FD) above 0.5 mm. We compared independent noise with known stationary first order autoregressive noise, called AR(1), at adjacent volume correlation 0.4. Each noise model used its own full eligible design as the variance reference. The calculation kept the original time gaps after row removal. The mean paired log cost change was -0.012047 in PIOP1 and -0.001994 in PIOP2. Their 95% person bootstrap intervals were [-0.016121, -0.008493] and [-0.003037, -0.001090]. Mean variance increases were 4.449% versus 2.933% in PIOP1 and 0.584% versus 0.372% in PIOP2. The AR(1) model had higher mean absolute variance. All 353,766 models passed the rank rule. Dense calculations agreed within 1.044e-13 relative error. A second analysis gave identical main result files. The package includes source inputs, hashes, code, a reusable tool, results, saved draws, and instructions. These fixed noise results do not measure artifact removal or identify a preferred censor rule.

fMRI · motion censoring · AR(1) · generalized least squares · variance · AOMIC