Pearson tests with exact variance matching gave a mean of 0.786% false calls at a 1% target under independent phases in 2-second SEEG beta windows from 21 people

We tested whether an exact variance match gives correct Pearson correlation tails. We used 700 channel pairs from 35 public SEEG recordings in 21 people. Each pair used 2 seconds before stimulation and the 13 to 30 Hz band. We fixed the measured Fourier magnitudes and drew independent uniform phases. At a 1% target, the corrected Pearson rule gave a mean of 0.786% false calls. Its 95% interval from resampling people was 0.735% to 0.824%. All 700 pair rates were below target. At a 0.1% target, the mean was 0.0538%. A full-spectrum rule gave 1.0014% in 35 million independent phase draws at the 1% target. A proof and two valid spectra show that equal variance does not give equal tails. A fresh raw-source extraction and a full repeat analysis matched the saved inputs and results. The package contains three figures, compact data, code, source hashes, checks, and reproduction instructions. This result can help check correlation rules used for BCI channel selection. It does not establish neural causality or improved decoder or stimulation performance.

Pearson correlation · fixed-spectrum phase null · tail calibration · effective sample size · SEEG · beta band · Fourier analysis · brain-computer interfaces