Brain responses vary in duration-modeling strategies and challenges

Typically, event-related brain responses are calculated invariant to the underlying event duration, even in cases where event durations observably vary: with reaction times, fixation durations, word lengths, or varying stimulus durations. Additionally, an often co-occurring consequence of differing event durations is a variable overlap of the responses to subsequent events. While the problem of overlap, for example, in fMRI and EEG is successfully addressed using linear overlap correction, it is unclear whether overlap correction and duration covariate modeling can be jointly used, as both are dependent on the same inter-event distance variability. Here, we first show that failing to explicitly account for event durations can lead to spurious results and thus are important to consider. Next, we propose and compare several methods based on multiple regression to explicitly account for stimulus durations. Using simulations, we find that non-linear spline regression of the duration effect outperforms other candidate approaches. Finally, we show that non-linear event duration modeling is compatible with linear overlap correction in time, making mass-univariate models combining non-linear spline regression of duration and linear overlap correction a flexible and appropriate tool to model overlapping brain signals. This allows us to reconcile the analysis of brain responses to stimuli in situations where durations differ between conditions, for example, different reaction times, different stimulus durations, or fixation-related potentials with different fixation durations. While in this paper we focus on EEG analyses, we additionally show that our findings generalize to fMRI BOLD responses, and argue that they should generalize to other overlapping signals such as LFPs, or pupil dilation responses.