Global/Incremental Update Table

On the Age × Probe × Shifted Along effect, the model computes two targeted custom contrasts across the three shifted_along levels — No Shift, Unchanged dimension, Changed dimension — testing two competing patterns for how activity should look across those three levels:

• Incremental updating — only the changed dimension stands out; the unchanged dimension looks like no-shift baseline. The region is tracking specifically what changed.
• Global updating — both the changed and unchanged dimensions rise together above no-shift baseline. The region responds to any shift event, not the specific dimension that changed.

Each contrast is a single linear combination of the three estimated marginal means (β_Changed, β_Unchanged, β_NoShift), computed per parcel within each age group × probe dimension cell:

Incremental Update = β_Changed − avg(β_Unchanged, β_NoShift)
Global Update      = avg(β_Changed, β_Unchanged) − β_NoShift

Currently shown one-sided (side = ">" in emmeans::contrast()): the null is that the combination is ≤0, and a significant result (the z-statistic and stars shown in each cell) means the combination is reliably greater than zero — a large negative estimate (the pattern running the opposite way) correctly shows p ≈ 1, not a small p-value. This direction is the one predicted a priori for young adults; it is not guaranteed to be the correct direction for older adults, who can show the reverse pattern (see the two-sided option).

Currently shown two-sided: no direction is assumed, so a significant cell can reflect either a positive estimate (the predicted direction, e.g. Changed > baseline) or a negative estimate (the reversed pattern, e.g. baseline > Changed). The z-statistic itself is identical to the one-sided version above — only the p-value (and therefore which cells clear FDR) differs. Watch the sign of the z-statistic in each significant cell to see which direction it went; a negative z on a normally-highlighted cell means the pattern ran opposite to the one-sided prediction, not that it strengthens it.

Raw p-values are Holm-adjusted within each parcel, then FDR-corrected (Benjamini-Hochberg) across all 302 parcels separately within each (contrast × age group × probe dimension) family. Because Incremental and Global share the No-Shift term, a region can look significant on both at once when No Shift is simply much lower than the other two conditions.