I am collecting data from an auditory oddball paradigm where, by
definition, the standard tones outnumber the targets and novel sounds. In
order to maintain a constant SNR across all three conditions I do not want
to compare evoked responses created from different numbers of epochs. Is
there a way to control the number of epochs allowed into the average in
mne_process_raw on a condition-wise basis? The only other solution I can
think of is to create a custom events list.
I assume the easiest way is via a custom event list, as you suggest.
However, different numbers of epochs may not always be a problem. When analysing surface MEG data or signed source estimates, different numbers of epochs may result in different variances, but it won't bias the means. If you compute intensities (e.g. unsigned MNEs) or root-mean-squares (e.g. to combined pairs of gradiometers), then different noise levels may result in different means. This can partially be corrected by baseline-correcting unsigned source estimates or RMS curves (this is briefly discussed at the bottom of http://imaging.mrc-cbu.cam.ac.uk/meg/SensorSpm).<http://imaging.mrc-cbu.cam.ac.uk/meg/SensorSpm).> It would be a shame if you had to throw away 80% of your data.