Accounting for Time Dependence in the Analysis of Event-related Brain Potential Data

Ching-Fan Sheu and David Causeur

Abstract

Event-related potentials (ERPs) are recordings of electrical activity along the scalp time-locked to perceptual, motor and cognitive events. In recent years, clinical and behavioral researchers have increasingly used ERPs to study the time courses of mental events.

Because ERP signals are often rare, occurring only in brief moments during trials, and weak, relative to the huge between-subject variability, establishing significant association between ERPs and treatment conditions, as well as behavioral variables of interest, poses major challenges to statistical analysis.

For mass univariate analysis of ERPs, false discovery rate (FDR) procedures, guaranteeing a low erroneous rejection rate of the null under independence, have recently emerged as favorite alternatives to the classical Guthrie-Buchwald procedure (GB), which accounts for time dependence with an autoregressive process. In high throughput settings, however, the detrimental effects of dependence on the accuracy of simultaneous testing has been widely known and a variety of approaches, such as decorrelation, have been developed to counter them.

Noting that ERP time dependence exhibits a block pattern suggesting strong local and long-range autocorrelation components, we propose an adaptive factor modeling of dependence together with a joint modeling of the signal and noise processes in a multivariate linear model framework. We conducted simulation with known signals embedded in real dependence structure extracted from authentic ERP trials of a directed forgetting memory experiment. The proposed procedure outperforms FDR procedures, GB method, surrogate variable analysis, and latent effect adjustment after primary projection by attaining the largest true detection rate, while keeping FDR under the 0.05 level for each comparison.