Temporal dynamics of emotional processing: Parsing trial-wise variance of the late positive potential using Generalizability Theory

The reliability of individual trial event-related potential (ERP) components extracted from electroencephalogram has been consistently questioned since ERP research began. This ambivalence is based on misunderstood assumptions stemming from Cronbach and Classical Test Theory. Contemporary methods allow for the reliability of individual ERP trials to be estimated and for analyses of these trial-level ERP components to be meaningfully parsed. We illustrate the use of Generalizability Theory procedures in estimating the reliability of trial-level ERPs using the late positive potential (LPP), a neural measure of motivated attention toward emotionally evocative stimuli. Individuals (N = 88) completed a passive viewing task while continuous EEG was recorded. Variability in trial-level LPP responses was decomposed into facets corresponding to individual differences, chronological trial within block, stimulus type, their two-way interactions, and specific stimuli. We estimated various reliability coefficients and found that both overall and category-specific person-level LPP estimates have good-to-excellent reliability, while the reliability of within-person differences (i.e., change) between arousal categories was fair for the early LPP. These results were generally consistent across time windows, but were highest early in the LPP time course. We argue that investigating reliability using trial-level data allows researchers to pursue hypotheses focused on neurophysiological dynamics that unfold over the course of an experiment and not risk false inferences (i.e., ecological fallacy) when using person-level aggregates to deduce such processes. Moreover, such analyses provide information that allows researchers to optimize their protocols by potentially reducing the number of individual trials, burden on participants, and cost, while retaining sufficient reliability.