Hi,

both methods (orders of processing) might be of interest. The first - computing the ERP and then the spectral power per subject - is producing the evoked oscillation, in which you only see the frequency response that is locked to the stimulus onset (mostly early components of the ERP or big ones like the P300 if it is not 'jittering' too much). The second - computing the spectral power of your trial and then averaging over all these spectograms - produces the induced oscillations. They also reflect oscillatory activity that is not stimulus locked - often later responses that are not immediately caused by the stimulus but some higher processes/feedback from regions further down the processing stream. Have a look at Tallon-Baudry & Bertrand 1999 for a nice introduction to this topic: Oscillatory gamma activity in humans and its role in object representation.

C Tallon-Baudry, O Bertrand - Trends in cognitive sciences, 1999 - Elsevier

In the end your choice depends probably on the processes you are most interested in :)

cheers,

C.

It might be useful to display both, but yes it is a reasonable method to do a grand average topo plot of the frequency band that you are interested in. You might want to consider breaking this down into smaller chunks of time so that you can see the progression. So if your experiment looks at a 1000ms epoch then you would probably want to consider plotting the distribution of the beta band power during the course of this (in which case you would definitely want a grand average). In this case you might look to the literature to see what the typical time course for movement related beta power is like and how this is distributed. Or you could draw up some arbitrary time points that cover the duration of the epoch - the length of these time windows will impact on the accuracy of any conclusions that you draw on the nature of the topography (50 ms windows might be a nice starting point).

Also, the beta band is a very large frequency band, and is commonly considered in terms of low and high beta. You will want to know which is the peak frequency within this band before plotting (you could do an average of the band as a whole but again, consult the literature), accomplish this using a power spectrum density analysis and going with the peak beta component.

I hope that this provides somewhat of an answer that can be of use to you. Good luck with your research.

Thank you for you kind help. It helps me a lot. Thank you for reminding the low and high beta part as normally I deal the beta band as a whole part.

I still have a question about the process part. For the total 60 trials, is it right to first average across trials along time and then calculate the band power. Or is it to first calculate band power from each trial and then average all band powers. I think the second one is right to generate topo plot.

You want to compute the power spectrum density using the continuous data for each subject, then compute the average psd for the group. You can then plot this.

Hi,

both methods (orders of processing) might be of interest. The first - computing the ERP and then the spectral power per subject - is producing the evoked oscillation, in which you only see the frequency response that is locked to the stimulus onset (mostly early components of the ERP or big ones like the P300 if it is not 'jittering' too much). The second - computing the spectral power of your trial and then averaging over all these spectograms - produces the induced oscillations. They also reflect oscillatory activity that is not stimulus locked - often later responses that are not immediately caused by the stimulus but some higher processes/feedback from regions further down the processing stream. Have a look at Tallon-Baudry & Bertrand 1999 for a nice introduction to this topic: Oscillatory gamma activity in humans and its role in object representation.

C Tallon-Baudry, O Bertrand - Trends in cognitive sciences, 1999 - Elsevier

In the end your choice depends probably on the processes you are most interested in :)

cheers,

C.

Alternatively - and possibly a somewhat controversial suggestion, you could compute both of the above methods but also compare these results to an event related synchronisation /desynchronisation. Sure you will choose a method that suits your needs.

You should definitely look for peak power frequencies inside your wide beta band. Grand averaging across subjects may wash out your effect, especially with the full band you collected. For presentation in a grand average topography you will want to determine the strongest frequency source and limit your grand average to it. As mentioned, you may also find substantial individual differences in beta power, which could justify normalizing your beta result(s) to it's/their peak level(s).

Thank you all of you. Now I better understand the two orders of processing.

Mouraux A. and Iannetti GD Magnetic Resonance Imaging 26 (2008) 1041 - 1054

Has a nice, down to earth explanation of the difference between the two ways of analysis, one revealing only time- and phase-locked events and the other all time-locked events whether phase or non-phase locked.

When I plot a spectral topographic map from a group of subjects, I usually follow these steps, for each subject:

- spectrum of each single epoch (i.e. with FFT methods)

- average of all the spectra

- absolute power extraction for the desired bands (i.e. beta)

- normalization of this power to the absolute total power, to obtain a relative power

then average the relative powers of all the subjects.

Normalization is crucial when comparing topographies. Absolute spectral values can differ too much among subjects.

I agree with Marco Cusi but here I would like to ask  whether EEGLAB allows one to plot average power topographic maps in  frequency bands  like delta, theta, alpha, beta, gamma. From the  'Plot channel spectra and maps' command one can only plot spectral plots  for individual frequencies. Is there a way of to plot average power and represent it in topographic maps?

If you do use the international 10-20-system for EEG electrode placement the individual skull shape and sizes will be normalized inherently. Averaging each corresponding electrode position across all subjects should be fine after having averaged the single trials for each subject.