There are several different approaches to analyze event-related potentials (ERPs) at single-subject level, and the aim of the current study is to provide information for choosing a method based on its ability to detect ERP effects and factors influencing the results. We used data from 79 healthy participants with EEG referenced to mastoid average and investigated the detection rate of auditory N400 effect in single-subject analysis using five methods: visual inspection of participant-wise averaged ERPs, analysis of variance (ANOVA) for amplitude averages in a time window, cluster-based non-parametric testing, a novel Bayesian approach and Studentized continuous wavelet transform (t-CWT). Visual inspection by three independent raters yielded N400 effect detection in 85% of the participants in at least one paradigm (active responding or passive listening), whereas ANOVA identified the effect in 68%, the cluster-method in 59%, the Bayesian method in 89%, and different versions of t-CWT in 22-59% of the participants. Thus, the Bayesian method was the most liberal and also showed the greatest concordance between the experimental paradigms (active/passive). ANOVA detected significant effect only in cases with converging evidence from other methods. The t-CWT and cluster-based method were the most conservative methods. As we show in the current study, different analysis methods provide results that do not completely overlap. The method of choice for determining the presence of an ERP component at single-subject level thus remains unresolved. Relying on a single statistical method may not be sufficient for drawing conclusions on single-subject ERPs.
Keywords: Data analysis; Electroencephalography; Event-related potentials; N400 evoked potential; Statistical data interpretation.
Copyright © 2019 The Authors. Published by Elsevier B.V. All rights reserved.