[Eeglablist] Double Dipping with ICA

[Dr. Michael Villanueva]

Nick
Your use of the word "distorted" in reference to ICA caught my eye.  Independent Component Analysis is a linear transformation.  Thus, ICA no more "distorts" data than subtracting the number 3 from 12 distorts the number 12.

Michael

-----Original Message-----
From: eeglablist <eeglablist-bounces at sccn.ucsd.edu> On Behalf Of Scott Makeig
Sent: Sunday, November 8, 2020 10:51 AM
Cc: EEGLAB List <eeglablist at sccn.ucsd.edu>
Subject: Re: [Eeglablist] Double Dipping with ICA

Re the double-dipping example Makoto provides: Don't neglect to check, however, whether and to what extent IC ERPs do contribute to the ICs signal variance - this may be less than 1% (i.e., removing the IC ERP from every trial reduces the overall IC signal variance by ~1%). If you wish, you can remove the ERPs from every trial after clustering, and then test differences in remaining cluster variance-not-explained-by-the-ERPs. Or, you can normalize the ERPs *before* clustering based on equivalent dipole location and normed ERPs, then test differences in cluster variance/power/etc.

Scott

On Sun, Nov 8, 2020 at 1:39 PM Makoto Miyakoshi <mmiyakoshi at ucsd.edu> wrote:

> Dear Nick,
>
> Two things.
>
> 1. The second ICA you mentioned does not change the result. Just try 
> it and you see what happens.
>
> 2. Double dipping refers to a procedure that is known to inflate Type 
> I error. ICA uses mutual information, and you test amplitude/phase 
> metric--these two do not have apparent correlation. If you say, 
> cluster ICs by ERP waveforms, then perform ERP amplitude difference, 
> this is double dipping (why? because IC cluster's ERP variance is 
> minimized by clustering, but this variance information is being used 
> in the subsequent statistical
> test)
>
> Given that,
>
> > Are there reasons not to do this or to do it?
>
> You may do so but it is meaningless because the second ICA in that 
> case does nothing. And this has nothing to do with the double dipping problem.
>
> Makoto
>
> On Sun, Nov 8, 2020 at 8:22 AM Nicholas Dogris < 
> drdogris at neurofieldneurotherapy.com> wrote:
>
> > Will my data be distorted if I run ICA, remove IC's such as eye 
> > blinks,
> and
> > then run ICA again on the same dataset?  Are there reasons not to do 
> > this or to do it?
> >
> > Cheers,
> >
> > Nick
> > _______________________________________________
> > 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 92093-0559, http://sccn.ucsd.edu/~scott _______________________________________________
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