Allowing different independent scan error variances changed the sign bounds for right hippocampal total change covariance in 176 selected DLBS adults, with empirical moments fixed

We tested two rules for independent scan error in 176 selected adults from OpenNeuro ds004856 v1.3.0. The outcome was covariance between successive total log volume changes. We held the empirical three-visit matrix fixed. For the right hippocampus, equal error variance at all visits permitted only negative latent covariance. Allowing a different variance at each visit permitted both signs. The upper endpoints were -2.248 and 50.920 squared log percentage points. For the sum of both sides, both rules permitted both signs. We proved sharp bounds under independent additive log scale errors and a positive semidefinite latent matrix. The method extends a related equal variance study of yearly rates. The package includes source files, provenance, code, a proof, all results, and run instructions. Independent checks passed. A full replay from fresh source files matched all 11 analysis files byte for byte. No scan error variance was estimated. The bounds apply to the fixed empirical matrix under stated assumptions. They do not identify the actual biological covariance sign.

MRI · hippocampus · measurement error · covariance bounds · longitudinal study · sensitivity analysis