Comparing Methods for Mass Univariate Analyses of Human EEG: Empirical Data and Simulations.

Area of Science:

• Neuroscience
• Cognitive Science
• Biostatistics

Background:

• Electroencephalography (EEG) is crucial for studying human brain dynamics.
• Traditional EEG analysis often requires predefined regions of interest (ROIs), limiting exploratory research.
• Mass univariate analyses offer a complementary approach for high-dimensional EEG data.

Purpose of the Study:

• To review and evaluate statistical methods for delineating spatial and temporal effect boundaries in EEG data.
• To compare permutation-based and Bayesian approaches for within-subjects comparisons across different EEG data types.
• To assess the sensitivity and conservativeness of various statistical inference methods in high-dimensional EEG analysis.

Main Methods:

• Review of permutation-based approaches (e.g., cluster-based permutation tests, tmax procedure) and Bayesian alternatives.
• Evaluation across three distinct EEG datasets: steady-state evoked responses, event-related potentials, and time-frequency data.
• Focus on within-subjects comparisons to identify condition differences.

Main Results:

• Cluster-based permutation tests demonstrated a liberal approach with high sensitivity for detecting large effects.
• The permutation-based tmax procedure proved to be the most conservative method across all evaluated datasets.
• Bayesian approaches require careful threshold selection for meaningful hypothesis support due to their continuous nature.

Conclusions:

• Mass univariate analyses, particularly permutation-based methods, are valuable for statistical inference in high-dimensional EEG data.
• The choice of method (cluster-based permutation vs. tmax vs. Bayesian) impacts sensitivity and conservativeness in effect detection.
• These findings aid researchers in selecting appropriate statistical tools for exploring spatiotemporal effects in EEG studies.