The Influence of Unequal Numbers of Trials on Comparisons of Average Event-Related Potentials

Abstract

Four studies examined (a) how event-related potentials (ERPs) change as the number of trials averaged increases and (b) the statistical implications of comparing ERPs composed of different numbers of trials. Experiment 1 utilized data from 7-year-old children performing an oddball task. The other three experiments used simulated data with different distributions of P3 peak latency. In all 4 experiments, peak amplitude decreased and the mean amplitude of the 300 to 900 msec interval remained stable as the number of trials averaged increased. The standard deviations of both measures decreased. These data show that the decrease in peak amplitude with increasing numbers of trials that has been found in other studies is not solely due to the elimination of residual noise but is likely to also involve a fundamental aspect of signal averaging and the algorithm used to select peaks. Furthermore, these experiments expose the possibility of statistical errors when investigators compare average ERPs composed of small versus large numbers of trials as is often done when the oddball paradigm is used.