Estimating statistical power for event-related potential studies using the late positive potential.

Abstract

The late positive potential (LPP) is a common measurement used to study emotional processes of subjects in ERP paradigms. Despite its extensive use in affective neuroscience, there is presently no gold standard for how to appropriately power ERP studies using the LPP. The present study investigates how the number of trials, number of subjects, and magnitude of the effect size affect statistical power in analyses of the LPP. Using Monte Carlo simulations of ERP experiments with varying numbers of trials, subjects, and synthetic effects of known magnitude, we measured the probability of obtaining a statistically significant effect in 1,489 experiments repeated 1,000 times each. Predictably, our results showed that statistical power increases with increasing numbers of trials and subjects and at larger effect sizes. We also found that higher levels of statistical power can be achieved with lower numbers of subjects and trials and at lower effect sizes in within-subject than in between-subjects designs. Furthermore, we found that, as subjects are added to an experiment, the slope of the relationship between effect size and statistical power increased and shifted to the left until the power asymptoted to nearly 100% at higher effect sizes. This suggests that adding more subjects greatly increases statistical power at lower effect sizes (<1µV) compared with more robust (>1.5µV) effect sizes. We confirmed the results from the simulations based on the synthetic effects by running a new series of simulated experiments based on real data collected while participants looked at emotional images.