A methodological issue in the study of correlation between psycho physiological variables, The Journal of Parapsychology 1

We used previously accumulated skin conductance (SC) and EEG data to examine the effects of their respective autocorrelations upon hypothesis testing. We found that SC data remain autocorrelated for many seconds, and that EEG data remain autocorrelated for many fractions of a second depending upon filtering parameters. We show that the effect of these non-zero autocorrelations upon the interpretation of correlation coefficients using normal statistics can lead to substantial and artifactually inflated significance levels. With SC, for example, the high autocorrelation can lead to a Pearson’s r correlation of 1.0 even under the null hypothesis. The resulting null-hypothesis z-score distribution range is [-20,20]; whereas, is should be approximately in the range [-3,3]. Alpha EEG, while less autocorrelated than SC, still leads to Pearson’s r correlations in the range [-0.4,0.4] leading to a null-hypothesis z-score distribution in the range [-15,15]. Beta band EEG reduces the null-hypothesis z-score range to [-5,5]. We demonstrate that standard Monte Carlo techniques can provide valid estimates of the significance levels. The underlying assumptions of conventional statistical tests can b