of a variable band-pass filter which scales with amplitude frequency, can alleviate this issue and improve the sensitivity of detecting PAC (Berman et al., 2012; Voloh et al., 2015). Thirdly, periods which contain non-stationary periods should be avoided. This includes sensory evoked potentials which induce correlations between frequency bands via phase reset (Sauseng et al., 2007), and can be misinterpreted as PAC (Aru et al., 2015). For this reason, we did not analyse the first 300 ms following visual grating presentation, due to the presence of visual evoked potentials (Di Russo et al., 2002). Fourth, given that PAC algorithms produce values ranging from 0 to 1, data are commonly not normally distributed, and therefore the use of non-parametric statistics is paramount. Whilst surrogate data are often employed (Tort et al., 2010b; Aru et al., 2015), this may not be possible where data are organized into short trials and temporal correlations between surrogate and true time-series are high (Dvorak and Fenton, 2014). Therefore, to assess changes in PAC, using a baseline period or contrasting between conditions, combined with non-parametric statistics may prove to be a useful alternative for sensory neurocognitive research.
