A novel methodology for data-driven EEG-ERP analyses based on massive univariate statistical methods and Bayesian inference

We present a novel methodology for analyzing electroencephalographic (EEG) event-related potentials (ERP) using massive univariate statistical methods combined with Bayesian inference. This data-driven approach automatically identifies clusters of electrodes and time windows showing potential effects, improving traditional methods that rely on a priori selection of regions of interest and null hypothesis significance testing (NHST). Our methodology addresses key limitations of NHST, including increased risk of type II errors and restrictive experimental design requirements. Through Bayesian inference, we evaluate and quantify the significance of the identified effects, providing a more flexible and interpretable framework for hypothesis testing. We applied this method to EEG data collected from ex-combatants, victims, and civilians involved in the Colombian armed conflict during a modified Implicit Association Test designed to measure implicit bias. Our approach demonstrated increased sensitivity compared to conventional NHST-based ERP analyses, with Bayesian inference offering robust evidence for group differences. This methodology enhances exploratory ERP research by mitigating issues related to multiple comparisons and integrating prior knowledge. It could also be applicable in experimental psychology and neuroscience studies where pre-selecting regions of interest is still challenging.