Towards improved methods for detecting behavioral rhythms: Addressing the shortcomings of previous methods

Theories suggest that cognitive processes like attention have a rhythmic nature, however there is rising controversy surrounding the empirical findings supporting this theory. The controversy centers around the statistical methods employed by studies to test for rhythms in accuracy time courses (ATCs) around a behavioral event. To detect rhythms, previous studies surrogates via shuffling-in-time, and then evaluated if significance occurs in a narrow frequency band (i.e. the presence of a spectral peak). In a recent critique, which challenged the methodological foundations of an entire research field, Brookshire (2022) argued that the shuffling-in-time procedure does not effectively distinguish between periodic and aperiodic processes. Brookshire therefore proposed a new method to test for rhythmic modulations, by creating surrogates via a 1st-order autoregressive process (AR(1) method). Strikingly, Brookshire reports that his AR(1) method does not replicate previously reported evidence for rhythmic modulations in all of 23 analyzed datasets, challenging a significant body of empirical work and a major theory of behavior and cognition. Here, we show that the evidence and arguments provided by Brookshire (2022) against the shuffling-in-time procedure are flawed. Moreover, Brookshire's proposed AR(1) method has a low sensitivity making it irrelevant for the use cases it is designed for. Hence, the reported null-findings by Brookshire (2022) need to be reassessed. We briefly outline a new method (APECS - APEriodiC Surrogate testing) that addresses the shortcomings of previous methods.