Abstract
When recording eye movement and EEG data from observers free-viewing visual stimuli, a researcher faces numerous possible confounds. Two very pervasive confounds are temporally overlapping neural responses and variance in EEG signals that is related to eye movement parameters (such as fixation position or saccade amplitude). Typically, researchers have avoided these confounds by constraining eye movements, e.g. by instruction or stimulus design, or by limiting analysis to a-typically long fixations and to "similar" datasets (often relying on an incoherent application of null hypothesis significance testing to argue for similarity). These common approaches inevitably lead to the inclusion of only a constrained subset of eye movements, possibly not representative of general gaze behavior. Similarly, to truly capture the relationship between eye movements (EMs) and neural activity, it is suboptimal to influence or diminish eye movement effects between conditions before evaluating EEG data. Moreover, NHST should not be used to argue the absence of meaningful differences (Sassenhagen & Alday, 2016). Here, we present a way to address confounds by applying continuous-time regression with numerical covariates (Smith & Kutas, 2015), which involves explicitly modelling overlap and eye movement parameters. We show that this method accurately estimates the modelled confounds in real Eye Tracking-EEG data, thereby controlling for them. Additionally, we discuss EM-EEG relationships that require additional investigation, as well as practical considerations for the application of the method. We conclude that continuous-time regression opens up new venues for investigating neural correlates of visual processing in more natural contexts, such as during free-viewing.
Meeting abstract presented at VSS 2018