How to test for phasic modulation of neural and behavioural responses

Abstract

 The phase of neural oscillations modulates neural responses and behaviour  We test the ability of different methods to detect such a phase effect  We test how this ability changes with nature of effect and experimental parameters  Results represent groundwork for future investigations of oscillatory phase effects AbstractThe question whether perception or other processes depend on the phase of neural oscillations is a research topic that is rapidly gaining popularity. Importantly however, it is unknown which methods are optimally suited to evaluate the hypothesized phase effect. Using a simulation approach, we here test the ability of different methods to detect such an effect and reject an absent one. We reveal optimal methods for different parameters that characterise the phase effect or depend on the experimental design to test for it. We demonstrate that permutation-based methods perform better than parametric ones for all of the simulated parameters. The identity of the optimal parametric method varies with several parameters, in particular with the number of phase bins chosen for the analysis and the width of the phase effect. In contrast, the identity of the best permutation-based method is unaffected by changes in the simulated parameters. In sum, our study lays a foundation for optimized experimental designs and analyses in future studies investigating the role of neural phase for behaviour and neural function. We supply MATLAB code that can be used to identify the ideal method (among all of the methods tested) for a specific scientific hypothesis and set of experimental parameters.

Evidence for the Rhythmic Perceptual Sampling of Auditory Scenes

“…Given that this apparent rhythmic structure may also emerge simply as byproduct of sub-sampling the behavioral sensitivity at a fixed time scale, we repeated the model fitting after shuffling behavioral responses across trials (Zoefel et al, 2019). We computed the probability that the model incorporating the best group-level frequency derived from the original data better explained the data than the trivial model in the shuffled data (based on the cv-AICc): these probabilities were small and revealed the actual effects as (close-to) significant: p = 0.08, 0.076, 0.040, and 0.068 for Experiments 1–4, respectively.…”

“…To determine whether the selection of a specific frequency-dependent model over the trivial model was indeed specific to the behavioral data, and was not induced by any other factor in the analysis such as the temporal binning (see e.g., Vorberg and Schwarz, 1987) for pitfalls involved in testing reaction times for rhythmic patterns), we repeated the model comparison using randomized data (Zoefel et al, 2019). We randomly paired stimuli and responses and computed the probability of selecting the rhythmic model (at the group-level frequency determined using the original data) over the trivial model across 2000 instances of randomized data based on the cv-AICc.…”

Effects of Slow Oscillatory Transcranial Alternating Current Stimulation on Motor Cortical Excitability Assessed by Transcranial Magnetic Stimulation

“…To determine if the effects of SO tACS on motor cortical excitability are specific to the tACS phase a permutation analysis was performed (Zoefel et al 2019). This analysis was performed separately for each of the four time-points (~60 MEPs per time-point per participant) as well as for all online and offline MEPs (~120 MEPs online/offline per participant).…”

Temporal prediction elicits rhythmic pre-activation of relevant sensory cortices

“…The first was a linear model with intercept and slope as free parameters. The second model was a combination of a linear function with a 1 Hz sine and 1 Hz cosine function (which is equivalent to a 1 Hz sinusoid with phase as a free parameter) (Zoefel, Davis, Valente, & Riecke, 2019) Furthermore, the model fits for the rhythmic trials had significantly higher 1 Hz amplitudes than those for non-rhythmic (Wilcoxon signed-rank test of 1Hz amplitudes across participants; Z = 3.86, p = 1.156e-04; Figure 4 C, top right).…”