On variability in local field potentials

Abstract

Neuronal variability is a fundamental feature of neuronal coding. In spiking activity, across-trial variance (ATV) normalized by mean spike count (i.e., the Fano factor) shows stimulus-induced quenching. ATV has also been studied in electro/magneto-encephalography (E/MEG) without normalization, revealing effects of stimulation and cognition. Here we show that outside of event-related potentials (ERPs), ATV for both EEG and local field potential (LFP) is nearly identical to intra-trial variance (ITV), which equals signal power. ATV–ITV correlation decreases during ERPs, proportional to how much the ERP explains total power. While EEG ATV shows post-stimulus quenching, LFP ATV does not, particularly in the gamma band, where power increases after stimulus onset. To provide an independent variability measure, we introduce CV(power), the coefficient of variation of signal power, as a mean-normalized metric. CV(power) is the inverse of the signal-to-noise ratio of power and thus related to decodability, offering a variability measure applicable to E/MEG and LFP. Parto-Dezfouli et al. analyze variability in LFP and EEG signals. They find that the previously used across-trial variability is almost identical to within-trial variability, i.e., the usual power metric, while the coefficient of variation of power better captures variability.