@article {637970, title = {Does Data Cleaning Improve Brain State Classification?}, journal = {Journal of Neuroscience Methods}, volume = {328}, year = {2019}, abstract = {\ BackgroundNeuroscientists routinely seek to identify and remove noisy or artifactual observations from their data. They do so with the belief that removing such data improves power to detect relations between neural activity and behavior, which are often subtle and can be overwhelmed by noise. Whereas standard methods can exclude certain well-defined noise sources (e.g., 50/60 Hz electrical noise), in many situations there is not a clear difference between noise and signals so it is not obvious how to separate the two. Here we ask whether methods routinely used to {\textquotedblleft}clean{\textquotedblright} human electrophysiological recordings lead to greater power to detect brain{\textendash}behavior relations.New methodThis, to the authors{\textquoteright} knowledge, is the first large-scale simultaneous evaluation of multiple commonly used methods for removing noise from intracranial EEG recordings.ResultsWe find that several commonly used data cleaning methods (automated methods based on statistical signal properties and manual methods based on expert review) do not increase the power to detect univariate and multivariate electrophysiological biomarkers of successful\ episodic memory\ encoding, a well-characterized broadband pattern of neural activity observed across the brain.Comparison with existing methodsResearchers may be more likely to increase statistical power to detect physiological phenomena of interest by allocating resources away from cleaning noisy data and toward collecting more within-patient observations.ConclusionsThese findings highlight the challenge of partitioning signal and noise in the analysis of brain-behavior relations, and suggest increasing sample size and numbers of observations, rather than data cleaning, as the best approach to improving statistical power.\ }, url = {https://www.sciencedirect.com/science/article/pii/S016502701930278X}, author = {Steven L. Meisler and Michael J. Kahana and Youssef Ezzyat} }