Five SEEG spectral features yield about four effective tests, but Holm remains safer than an M_eff Sidak correction

In one 60 s memory-task segment from 229 patients and 27,607 bipolar pairs in the RAM FR1 and catFR1 cohorts, five standard SEEG spectral features had an effective test count of 4.0 by Li–Ji and 4.39 by Nyholt. Within patients, relative beta power and the 2–40 Hz aperiodic exponent were strongly correlated (Spearman rho −0.84, 95% CI −0.85 to −0.83). The feature correlation pattern replicated across six sites (r 0.93–1.00), although four of ten within- versus between-patient correlations differed after Holm correction. In a 200,000-draw null simulation that assumed test-statistic correlations matched feature correlations, Holm over five features had family-wise error 0.043; a Sidak correction using Li–Ji M_eff = 4 had error 0.054. These results support retaining Holm for these five features and caution against treating beta and exponent effects in the same pairs as independent findings. Two preregistered predictions were not supported: Li–Ji M_eff was not below 4, and the lower 95% CI for the first two components’ variance share was 0.6998, just below 0.70. The analysis uses RAM cohorts, one segment per patient, and a simulation assumption that may not hold for every test design.

SEEG · spectral features · multiple testing · Holm correction · effective number of tests · family-wise error rate · intracranial EEG · free recall · preregistered analysis