[Eeglablist] Apparent inconsistencies in study power spectrum computations

[Raza Ward]

Dear Eeglablist,

I'm presently running a study in eeglablist (independent variables: three task/epoch types within each dataset, two conditions per subject with a dataset each), and trying to extract the power spectrum (pre)computations eeglab made for my study design as numerical tables. Opening the .specdat files in matlab appears achieve this; however, I'm running into a worrisome apparent inconsistency: when I precompute the spectral data at default settings ("specmode" "fft" "logtrials" "off")  using the "save single trial measures" function, for every channel/component it faithfully produces a row for every epoch with columns for every 1/3th of a Hz across the frequency spectrum; and when I don't use the "save single trial measures", it outputs a single row with supposedly combined values for every 1/3th of a Hz instead. The problem is this: the combined values are not in fact the average of the individual epoch values, and while the average of all epochs and the outputted combined value are reasonably predictive of each other, their proportionality varies by as much as 0.35. Moreover, when I change the study design to use paired rather than unpaired statistics and tag "save single trial measures", it outputs yet another, different, power value for every epoch, the averages of which are again correlated but not fully consistent with the previously described values.

Where are these differences coming from, and what do I need to do to get my hands on the "true" average power for all epochs of a single type from within a single dataset?

My gratitude and best regards,
~ Raza
 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20150411/fa573816/attachment.html>