Average reference increased mean beta edge concentration at high-power contacts in 16 SEEG models with independent phases

We tested whether a common average reference moves strong graph edges toward high-power contacts when contact phases are independent. We kept beta magnitudes from 16 public SEEG records in OpenNeuro ds004100 version 1.1.3. We generated 199 independent phase draws per record. Each graph kept the top 5% of all contact pairs by squared pooled coherency. The high-power set contained the top 10% of contacts, rounded up. Edges that touched this set accounted for 15.97% of the selected edges before average reference and 28.23% after average reference. These values give equal weight to each record. The paired increase was 12.26 percentage points. A descriptive 95% record bootstrap interval was 5.67 to 20.58 percentage points. The increase was positive in 15 of 16 records. The power-balanced average reference control gave 20.23%. We provide exact cross-spectrum moments for any fixed real montage, code, input magnitudes, graph selections, and source checks. All 16 inputs matched fresh source rebuilds. A complete repeat matched all numerical outputs. The result shows a reference effect under the stated phase model. It does not show neural coupling in the recorded data.

stereoelectroencephalography · common average reference · squared coherency · graph concentration · independent phase model