Bayesian Parallel Factor Analysis for Studies of Event-Related Potentials

The aim of the present work was to develop a Bayesian probabilistic model for parallel factor analysis of event-related potentials (ERP) in the human brain. Twelve statistical models considering the specific features of signals from ERP sources are proposed. Procedures for constructing sets of random parameter values based on Markov chain Monte Carlo methods were developed for these models. The effectiveness of these procedures was evaluated using both synthetic data with different signal:noise ratios and a set of ERP recordings obtained from 351 people in a Go/NoGo test. The procedure yielding the most accurate parameter assessments for models was selected. Analysis of the relationship between signals in the model and the type of activity performed by human subjects showed that Bayesian parallel factor analysis identifies functional differences between ERP components.