On variability in local field potentials

Neuronal coding and decoding would be compromised if neuronal responses were highly variable. Intriguingly, neuronal spike counts (SCs) show a reduction in across-trial variance ( $$ ) in response to sensory stimulation, when SC variance is normalized by SC mean, that is, when using the Fano factor ¹. Inspired by this seminal finding, $$ has also been studied in electroencephalography (EEG) signals, revealing effects of various stimulus and cognitive factors as well as disease states. Here, we empirically show that outside of evoked potentials, the $$ of the EEG or local field potential (LFP) is very highly correlated to the intra-trial variance ( $$ ), which corresponds to the well-known power metric. We propose that the LFP power, rather than the raw LFP signal, should be considered with regard to putative changes of its variability. We quantify LFP power variability as the standard deviation of the logarithm of the power ratio between an active and a baseline condition, normalized by the mean of that log(power ratio), that is the coefficient of variation (CV) of the log(power ratio). This CV(log(power ratio)) is reduced for gamma and alpha power when they are enhanced by stimulation, and it is enhanced for alpha power when it is reduced by stimulation. This suggests a potential inverse relation between changes in band-limited power and the corresponding CV. We propose that the CV(log(power ratio)) is a useful metric that can be computed for numerous existing and future LFP, EEG or MEG datasets, which will provide insights into those signals' frequency-specific variability and how they might be used for neuronal coding and decoding.