Electrophysiological network (in)stability: The impact of connectivity metric choice and sample size

Poster Session F - Tuesday, March 10, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballrooms

Juee Naik¹, Riley DeHaan¹, Michael Kahana¹; ¹University of Pennsylvania

Numerous studies of electrophysiological functional connectivity (FC) have revealed large-scale, often frequency-specific coordination across brain regions during many cognitive functions, including episodic memory, cognitive control, motor planning, and vision. However, the field uses a large variety of connectivity measures that are rarely compared head-to-head. Given these measures differ substantially in terms of properties such as sensitivity to zero-lag mixing and sample-size bias, the network one observes may depend heavily on the chosen metric, limiting comparability across studies. Here we compare several popular electrophysiological FC measures including coherence, PLV, PPC, ciPLV, PLI, wPLI, dPLI and AEC across the theta band (3-8 Hz) in a large free recall experiment conducted in 383 patients recorded intracranially for epilepsy treatment. All measures indicate stronger overall connectivity across the brain during successful retrieval than during successful encoding. Measure by measure network correlations show high agreement across measures and cluster into phase-synchrony (coherence/PLV/PPC), lagged-phase (PLI/wPLI/ciPLV), amplitude-envelope (AEC), and directed (dPLI) groups, with higher agreement during encoding than retrieval. Reliability analyses demonstrate that population FC matrices depend strongly on sample size: at typical sample sizes for intracranial studies (5-25 subjects), connectivity matrices for population subsamples correlate only weakly with the full-sample matrix for all FC measures (Pearson r = 0.2 to 0.3), with reliability increasing steadily with sample size up through hundreds of subjects. These findings suggest that future studies of electrophysiological FC would benefit from substantially larger samples alongside detailed multiverse comparisons to ensure reliable network inference.

Topic Area: METHODS: Electrophysiology