[FieldTrip] Running sensor space stats then source?
jan.schoffelen at donders.ru.nl
Fri Jan 18 16:55:33 CET 2013
I don't have a definite answer to your question, but note that analysing the data at the level of the sources is something a bit more involved than just taking the log-transform...
Here are some pseudo-random thoughts:
Essentially, sensor level signals present themselves as a linear mixture of the underlying sources of interest, and the source modelling attempts to unmix the sensor signals using biophysical (and additional) constraints.
In general, source level analysis will allow for a 'cleaner' interpretation of the possible location of the sources, where one should always account for the fact that the spatial resolution of EEG/MEG source reconstruction is not typically very high. In addition, the more relevant reason to try to unmix the sensor-level data, is to get a cleaner account of the temporal structure in and between the underlying neural generators, allowing for less problematic interpretation of univariate and bi/multivariate (connectivity) quantities estimated from the data. As a result of the source reconstruction, it could be that results are uncovered which are not easily visible from the sensor data alone.
It is perfectly valid to constrain yourself to sensor-level analysis, if this allows you to make your scientific point. Also, once you manage to reject your null-hypothesis (allowing you to speculate about the alternative hypothesis in your discussion section of your paper), there is no need to go to the source level.
I guess that to some extent it is also a matter of taste, familiarity with the methods, and opportunity which drives researchers to choose for one or the other approach.
These are just my thoughts and ideas, and it could be that other people on this forum have more sensible things to say about it.
On Jan 15, 2013, at 6:00 AM, Peter Goodin wrote:
> Hi fieldtrip list,
> I'm about to start running some stats on my data but have run into a bit of a problem when it comes to the correct method.
> I'm interested in looking at ER "p3" activity and so will first be using the cluster based tests in fieldtrip to examine for significant differences between my two groups, limiting the TOI to around 300 - 500ms.
> The problem comes when I want to to do source localisation, as I'll be using the same time window, just in a different (more assumption filled) space. In my head it's like running a t-test on "raw" data then doing it again transformed (eg, log transform) numbers, removing any assumption of independence and by conventional wisdom shouldn't be done.
> Most of the articles I've read seem to go for one or the other but those that use both don't make any discussion regarding it. Can anyone point me in the right methodology direction?
> Peter Goodin,
> BSc (Hons), Ph.D Candidate.
> Brain and Psychological Sciences Research Centre (BPsych)
> Swinburne University,
> Hawthorn, Vic, 3122
> Monash Alfred Psychiatry Research Centre (MAPrc)
> Level 1, Old Baker Building
> Commercial Road
> Melbourne, Vic, 3004
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
Jan-Mathijs Schoffelen, MD PhD
Donders Institute for Brain, Cognition and Behaviour,
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands
Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands
J.Schoffelen at donders.ru.nl
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip