Campopiano Allan, van Noordt Stefon J R, Segalowitz Sidney J
Research and Development Services, Halton Catholic District School Board, Burlington, Ontario, Canada; Cognitive and Affective Neuroscience Laboratory, Department of Psychology, Brock University, St. Catharines, Ontario, Canada.
Cognitive and Affective Neuroscience Laboratory, Department of Psychology, Brock University, St. Catharines, Ontario, Canada; Developmental Electrophysiology Laboratory, Yale Child Study Center, Yale University School of Medicine, New Haven, CT, USA.
Behav Brain Res. 2018 Jul 16;347:425-435. doi: 10.1016/j.bbr.2018.03.025. Epub 2018 Mar 21.
Research on robust statistics during the past half century provides concrete evidence that classical hypothesis tests that rely on the sample mean and variance are problematic. Even seemingly minor departures from normality are now known to create major problems in terms of increased error rates and decreased power. Fortunately, numerous robust estimation techniques have been developed that circumvent the need for strict assumptions of normality and equal variances, leading to increased power and accuracy when testing hypotheses. Two robust methods that have been shown to have practical value across a wide range of applied situations are the trimmed mean and percentile bootstrap test. To facilitate the uptake of robust methods into the behavioural sciences, especially when dealing with trial-based data such as EEG, we introduce STATSLAB: An open-source EEG toolbox for computing single-subject effects using robust statistics. With the STATSLAB toolbox users can apply the percentile bootstrap test, with trimmed means, to a variety of neural signals including voltages, global field amplitude, and spectral features for both scalp channels and independent components. The toolbox offers a range of analytical strategies and is packaged with a fully functional graphical user interface that includes documentation.
过去半个世纪对稳健统计的研究提供了确凿证据，表明依赖样本均值和方差的经典假设检验存在问题。现在已知，即使与正态性有看似微小的偏差，在错误率增加和功效降低方面也会产生重大问题。幸运的是，已经开发出了许多稳健估计技术，这些技术无需严格假设正态性和方差齐性，从而在检验假设时提高了功效和准确性。已证明在广泛应用场景中具有实用价值的两种稳健方法是截尾均值和百分位数自助法检验。为了便于在行为科学中采用稳健方法，特别是在处理基于试验的数据（如脑电图）时，我们推出了STATSLAB：一个用于使用稳健统计计算单受试者效应的开源脑电图工具箱。使用STATSLAB工具箱，用户可以将带有截尾均值的百分位数自助法检验应用于各种神经信号，包括头皮通道和独立成分的电压、全局场振幅和频谱特征。该工具箱提供了一系列分析策略，并配有包含文档的全功能图形用户界面。
