Positive weights matched equal person and time totals after ART sample deletion in 155 ds000228 movie scans

We tested sample weights for linear group means in OpenNeuro ds000228 version 1.1.0. We used recorded artifact regressors from all 122 children and 33 adults. Positive weights matched equal person totals and equal totals for all 158 movie times. The result held after deletion of recorded artifacts and after deletion within one or two TRs of each artifact. No person or movie time was excluded from the group fits. With the original flags, equal-person averages had a sharp common-signal difference bound of 0.0286133 for signals with absolute value at most one. The fitted bound was below 5e-16. Squared-weight sums increased by 0.144% in children and 0.051% in adults. These sums describe a variance ratio only for independent errors with equal variance. Linear programs and explicit dual certificates established the smallest possible maximum weight multipliers in all six cases. A separate parser and reverse-order fit passed. Fresh retrieval matched all 470 source hashes. A full repeat matched all 13 analysis files. The method cancels a shared time signal in a linear mean. It does not recover arbitrary missing neural values or establish a developmental effect. Iterative proportional fitting is an established algorithm. The package preserves source files, provenance, fitted weights, proofs, code, and reproduction instructions.

iterative proportional fitting · sample weighting · missing data · fMRI · movie viewing · marginal balance · artifact censoring · linear estimands