I wouldn't do that comparison as it won't show you much

what you want to do is to look at the data during the cognitive task

and try to correlate across different subjects what accounts for success at the task

eyes closed data, whether before or after a task, will not give you what you really want to know, which is How does the brain actually work. eyes closed data doesn't tell you that

kirt thornton

I think, both methods will yield valid results in some way. Still, with this normalization procedures important information will get lost and one has to be careful when interpreting the results. It should be checked before normalisation, whether there are significant differences of amplitude between the groups you study. Not only structural variables influence the amplitude. A higher amplitude in one group might be due to an e.g. larger amount of ocillation neurons at the same frequency, thus indicating differences in neurophysiological activity.

@Kirtley Thornton

I think that the understanding of the ongoing activity of the human brain during rest, i.e. the self organizing and self maintaining neural oscillations are essential to get a clue about what is going on during any cognitive task.

Hi Clara,

First, I would agree with Kirtley Thornton. The resting eyes closed state should be independent. It shouldn't depend on the time of the recording - before or after a task. But, I don't know of any study. So, perhaps it's worth looking at it.

Second, the EEG data, if resting or during a task, depend on various parameters: Therefore, you need to regress your values.

EEG values in females are, normally, higher than in male. The EEG values decrease with age and subjects of European descent have higher values than subjects of African descent. So, you certainly need these three parameters but can check with a statistical program if they are significant in your study. If the age range is small, age is perhaps not significant and could be dropped. But you certainly need to regress these factors out.

Then you can z-score the values across both groups to see if there are differences.

Hope this helps.

Thanks for your answers and recommended readings. As per the comments regarding the meaning of looking for traces in the resting state following performance in tasks, it is an interesting and relatively unexplored venue that could potentially shed light on the mechanisms of use-dependent plasticity. I agree that it needs to be coupled (and possibly correlated) with analysis of brain activity during the task, which is in fact what I do. There are certainly challenges but I also think it's worth exploring. Here a couple of recent related papers that might start to persuade you:

Local experience-dependent changes in the wake EEG after prolonged wakefulness.

Hung CS, Sarasso S, Ferrarelli F, Riedner B, Ghilardi MF, Cirelli C, Tononi G.

Sleep. 2013 Jan 1;36(1):59-72. doi: 10.5665/sleep.2302.

PMID: 23288972

Electrophysiological traces of visuomotor learning and their renormalization after sleep.

Landsness EC, Ferrarelli F, Sarasso S, Goldstein MR, Riedner BA, Cirelli C, Perfetti B, Moisello C, Ghilardi MF, Tononi G.

Clin Neurophysiol. 2011 Dec;122(12):2418-25. doi: 10.1016/j.clinph.2011.05.001. Epub 2011 Jun 8.

PMID: 21652261

Repetitive tactile stimulation changes resting-state functional connectivity-implications for treatment of sensorimotor decline.

Freyer F, Reinacher M, Nolte G, Dinse HR, Ritter P.

Front Hum Neurosci. 2012;6:144. doi: 10.3389/fnhum.2012.00144. Epub 2012 May 23.

PMID: 22654748

Translating long-term potentiation from animals to humans: a novel method for noninvasive assessment of cortical plasticity.

Clapp WC, Hamm JP, Kirk IJ, Teyler TJ.

Biol Psychiatry. 2012 Mar 15;71(6):496-502. doi: 10.1016/j.biopsych.2011.08.021. Epub 2011 Oct 5. Review.

PMID: 21974785

@Sebastian: I hear what you say. Indeed that is exactly the core of the issue, disentangling the structural and physiological contributions to differences in power. For example, in the case that I am dealing with, my patient population seem to have on average a much larger power (broadband - up to about 25 Hz) than controls. Before jumping to conclusions about the role of pathology, I wanted to rule out the possibility of other effects related to inter subjects variability, hence my question. Thanks again for your very useful answers.

Cara Clara,

have you ever thought about considering cortical sources intead of channels amplitude? In that case (for example, LORETA) you could use the normalization of a reference-free method with a reduced inter-subject variability.

- Resting state cortical electroencephalographic rhythms in subjects with normal and abnormal body weight.

Babiloni C, Marzano N, Lizio R, Valenzano A, Triggiani AI, Petito A, Bellomo A, Lecce B, Mundi C, Soricelli A, Limatola C, Cibelli G, Del Percio C.

Neuroimage. 2011 Sep 15;58(2):698-707. doi: 10.1016/j.neuroimage.2011.05.080. Epub 2011 Jun 17.

Ciao Antonio,

I had thought of it but not implemented it yet as I am not too familiar with source reconstruction.

I will certainly have a look at the paper and maybe I will contact you with some question!

Grazie !

Hi Clara, just a note to get back to the question that you posed. In my experience I found it it was serviceable to normalize power in each frequency band for each individual at each channel, by dividing those values by the maximum power value observed across frequency bands and channels for each individual. as described in my small paper "Frontal theta cordance predicts ..." and key references therein (work by Cook, Leuchter and colleagues). Thanks

Hi Clara,

Normalization by dividing by total power is tricky indeed, because total power both contains power due to unspecific (and for instance gender specific) 'noise', but also power due to the very brain activity your interested in (for example alpha power with eyes closed has a large contribution). So by dividing by total power you can end up 'mixing' in band specific brain results to all other bands. So what you end up with after normalization depends on the activity (power spectrum) you have, and can be different per experiment (and conditions like eyes open or eyes closed or cognitive task). I would advice you to make topographical and power spectrum plots of the absolute and relative power to see if all looks ok after normalization. You can also try using for instance only the power in higher frequencies (where the power is more due to noise and less to brain) to normalize by to avoid mixing in alpha effects. Hope it helps and good luck. And I'm following this thread with interest since I'm struggling with the same thing at the moment :)

Thank you Ingrid I did not consider that

but my thought was to divide by maximal power (within a particular band), not total. but I wonder what would be some drawbacks to that approach? thanks

In a recent preclinical study in rats, I had significant differences between the vehicle and drug treated groups prior to dosing. In rats, a log transformation of the FFT linearizes the spectral response such that the average power from e.g 0 - 100 Hz (2 Hz bins) is around zero (see e.g. Buzsaki, "Rhythms of the Brain", fig. 5.2). By subtracting this average power value from all pre- and post-dosing power values (e.g. at all frequencies and after log transformation), the pre-dosing curves were not different (i.e. by average power over the entire range, or over delta, theta, gamma frequency bands, however you wish to define them). This a very simple method, which was mentioned above but the initial log-transformation is crucial. Also, of course, the normalization period must be specific for the variable you are investigating - in my case drug. However, as noted above, time of day, eyes open/ closed, cortical position, etc. must also be controlled for by appropriate pre-test recording periods / records specific to each condition. It is not fool-proof, however, as for example some rats in one group may produce huge theta, and then the normalization will still fail. The only way to be sure is to do the normalization on all subjects before you assign them to treatment groups. That should provide the best probability of finding significant treatment effects. If subjects is one of your variables, as apparently in your case, that won't work, but you will be able to better detect changes at particular frequencies.

Thanks for this question, Clara.  I appreciate the responses, as well.  

I note that none of the responses address the Z-score method. I've just seen this method used in recent sleep study paper: Sprecher, Riedner, Smith Tononi, Davidson, Benca 2016.  Anyone else out there have thoughts on using this Z-score method as an alternative? 

Also, it seems there are at least two approaches for dividing by average power. A more traditional method (I think, maybe prior to dense-arrays), was to divide by average power across all frequency bins at each electrode.  An alternative is to divide by average power across all electrodes, but separately for each frequency bin or range that you are examining. I see this method used more frequently with dEEG.  I guess either can be used, depending on your question.  I'm curious if others have thoughts on this.

Thanks!