Hi Udo,
Thanks for sending. What happens is that, for the PE percentage change, what is used are the EVs themselves as entered in the design, whereas for the COPE percentage change, what is used are the EVs of the model partitioned into EVs of interest and nuisance.
The main difference is that, if the EV of interest isn't perfectly orthogonal with the nuisance, its amplitude (peak to peak height) changes, and this amplitude is used in the computation of the percentage. The partitioning isn't explicitly shown, but it is
necessary to account only for the variability of the contrast itself, thus ignoring sources of variability that might be in the other EVs. If the other EVs were perfectly orthogonal to the EV tested with this contrast, or if there were only one EV, then the
result should be the same (this is what might be going on at the higher levels).
The partitioning is described in the Appendix B of Smith et al (2007), and again in the context of randomise in the Appendix A of Winkler et al (2014):
All the best,
Anderson