Extension of the non-parametric cluster-based time-frequency statistics to small time windows and to single condition tests

Oscillatory processes are central for the understanding of the neural bases of cognition and behaviour. To analyse these processes, time-frequency (TF) decomposition methods are applied and non-parametric cluster-based statistical procedure are used for comparing two or more conditions. While this combination is a powerful method, it has two drawbacks. One the unreliable estimation of signals outside the cone-of-influence and the second relates to the length of the time frequency window used for the analysis. Both impose constrains on the non-parametric statistical procedure for inferring an effect in the TF domain. Here we extend the method to reliably infer oscillatory differences within the full TF map and to test single conditions. We show that it can be applied in small time windows irrespective of the cone-ofinfluence and we further develop its application to single-condition case for testing the hypothesis of the presence or not of time-varying signals. We present tests of this new method on real EEG and behavioural data and show that its sensitivity to single-condition tests is at least as good as classic Fourier analysis. Statistical inference in the full TF map is available and efficient in detecting differences between conditions as well as the presence of timevarying signal in single condition.