[Eeglablist] ERP analyses and average referencing

Sat Oct 25 09:02:40 PDT 2008

Carlos - Filters learned by ICA are both reference free and data adaptive.
This is because the information content and spatial sources are not altered
by multiplication by a fixed (well-conditioned, dimension-preserving)
matrix, which is what re-referencing involves. Both the IC time courses and
equivalent dipole source locations should remain the same under
re-referencing -- within numerical and statistical error.

I'm sure a laplacian filter plug-in for EEGLAB does exist. To my mind, the
shortcoming of Laplacian filters, vis a vis ICA, is that they are not data
adaptive. In particular, they cannot well model or represent a source whose
equivalent dipole is not radially oriented (e.g., tangential sources located
in cortical sulci).

Scott Makeig

On Thu, Oct 23, 2008 at 3:16 PM, carlos hamame <comediante.x at gmail.com>wrote:

> I'm surprised not to see a free-reference option like the laplassian
> transform here.
>
> On Thu, Oct 23, 2008 at 7:29 PM, Steve Luck <sjluck at ucdavis.edu> wrote:
>
>> I'd like to make one last set of comments about the choice of the
>> reference electrode:
>>
>> 1) Alex is right that there is no truly correct reference.  It is always a
>> matter of balancing the advantages and disadvantages of different reference
>> montages.
>>
>> 2) He is also correct that there may be statistical power advantages to
>> using the average reference.
>>
>> 3) The major advantage of mastoids (and earlobes, etc.) is that they are
>> widely used in ERP research (although not in all subareas).
>>
>> 4) The most important thing is to recognize that your data are always
>> influenced by your choice of reference.  As long as you don't forget this
>> (and make sure that others don't forget it as well), you will be fine.
>>
>> 5) For experiments with broad, dense, and uniform distributions of
>> electrodes, the wisest choice is usually to look at the data _both_ with an
>> average reference and with a mastoid (or earlobe, etc.) reference.  That
>> gives you the best of both worlds.
>>
>> 6) For experiments with a limited or nonuniform distribution of
>> electrodes, avoid using the average reference (unless you are being
>> extremely careful).
>>
>> Steve Luck
>>
>> *From: *"Alexander J. Shackman" <shackman at wisc.edu>
>> *Date: *October 21, 2008 12:22:47 PM PDT
>> *To: *eeglablist at sccn.ucsd.edu
>> *Subject: **Re: [Eeglablist] ERP analyses and average referencing*
>> *Reply-To: *ajshackman at gmail.com
>>
>>
>> Arno prefaced his comments by noting that "average referencing is always
>> incorrect." But as Joe Dien notes in his excellent '98 paper, it would be
>> equally appropriate to say that "mastoids montages are always incorrect" or
>> "ALL referencing schemes are always incorrect."
>> Steve and Arno are correct in noting, as Dien did, that the topography and
>> waveforms will differ across montages, making it difficult to compare
>> average to the more commonly used (for ERPs, at least) mastoids montage.
>>
>> But there are at least two other reasons, at least with high-density
>> recordings, to consider using the average reference. 1) Dien suggests that
>> the average reference, which is employed by both dipolar and distributed
>> source modeling algorithms, potentially provides more insight into the
>> underlying cerebral generators. 2) The average reference is likely to more
>> psychometrically reliable (cf. S. Gudmundsson et al., Clinical Neurophys,
>> 2007).
>>
>> Alex Shackman
>>
>>
>>
>>
>> On Sun, Oct 19, 2008 at 2:29 PM, Steve Luck <sjluck at ucdavis.edu> wrote:
>>
>>> I would like to echo and expand upon Arno's comments about average
>>> referencing.  Under the most optimal conditions this can be perhaps a decent
>>> approximation (see Dien, 1998).  However, under most conditions it is a poor
>>> and misleading approximation (and, as Arno pointed out, is is never
>>> completely correct).  Your waveforms will look completely different
>>> depending on what electrodes you happen to be using (see Figure 2 and the
>>> related text in chapter 3 of An Introduction the Event-Related Potential
>>> Technique).  As a result, your data may look quite different from the data
>>> of other researchers, even if they are also using the average of all sites
>>> as the reference (because they probably don't have exactly the same set of
>>> sites that you have).
>>>
>>> So, what to do?  Lately, my lab has been seeing the same sort of problem,
>>> with lots of muscle activity being picked up by mastoid reference
>>> electrodes.  The best thing to do is to try to get subjects to sit in a more
>>> neutral position so that they do not need to contract the neck muscles to
>>> keep the head upright. However, if you already have this noise in your
>>> mastoid data, you can try referencing to scalp sites that are close to the
>>> mastoids (e.g., P9 and P10), which may have less muscle noise.  Or, if you
>>> have a sufficiently dense array of electrodes, you could use the average of
>>> a small cluster around the mastoids on each side as the reference.
>>>
>>> The most important thing is to realize that you are _always_ looking at
>>> the potential between two electrode sites (or groups of sites).  There is no
>>> such thing as potential at a single site.
>>>
>>> Steve Luck
>>>
>>> *From: *arno delorme <arno at ucsd.edu>
>>> *Date: *October 18, 2008 4:15:53 AM PDT
>>> *To: *Yvonne Tran <Yvonne.Tran at uts.edu.au>
>>> *Cc: *eeglablist at sccn.ucsd.edu
>>> *Subject: **Re: [Eeglablist] ERP analyses and average referencing*
>>>
>>>
>>> Dear Yvonne,
>>>
>>> average referencing is always incorrect. The amount of current going in
>>> and out of the head is assumed to be 0. Using that properties, average
>>> referencing means that the average potential across all electrode is 0 at
>>> all times. However, you cannot expect that the electrode spatial
>>> distribution will be homogenous over the head (because first you cannot put
>>> any within the neck, and there is usually no electrode on the face etc...).
>>> It is generally assumed that the current flowing within the neck is
>>> negligible (because of high conductances).
>>>
>>> As an answer to your question, if your electrode repartition is
>>> relatively homogenous on the scalp, then you may use average reference.
>>> Nevertheless, average reference will not make it easy to compare between
>>> montages.
>>>
>>> Best regards,
>>>
>>> Arno
>>>
>>> On 17 sept. 08, at 04:20, Yvonne Tran wrote:
>>>
>>> Dear All
>>>
>>>
>>> We are currently working with spinal cord injured participants and have
>>> recorded some oddball data. We have been using A1 and A2 mastoid for
>>> reference channels, however, with this particular group we are experiencing
>>> increased muscle tension in this region (which cannot be prevented, as some
>>> participants are unaware that they are tensing up), and therefore when the
>>> data are re-referenced the other EEG channels become flooded with muscle
>>> tension noise. This can be overcome when we re-reference using average
>>> referencing. My question is how many electrodes (evenly distributed around
>>> the scalp) will be ok for average referencing for ERP analyses? We have 26
>>> EEG channels.
>>>
>>>
>>> Any suggestions/opinions appreciated!
>>>
>>>
>>> Thank you
>>>
>>> regards
>>>
>>> Yvonne
>>>
>>>
>>>   --------------------------------------------------------------------
>> Steven J. Luck, Ph.D.
>> Professor
>> Center for Mind & Brain and Department of Psychology
>> University of California, Davis
>> 267 Cousteau Place
>> Davis, CA 95618
>> (530) 297-4424
>> sjluck at ucdavis.edu
>> --------------------------------------------------------------------
>>
>>
>>
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to
>> eeglablist-request at sccn.ucsd.edu
>>
>
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>

-- 
Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation, University of
California San Diego, La Jolla CA 92093-0961, http://sccn.ucsd.edu/~scott
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20081025/20ed305d/attachment.html>