[Eeglablist] How to read "pvaf" in envtopo()

[Scott Makeig]

Since independent components are not spatially orthogonal, their
projected signals can (naturally) partially cancel each other at the
scalp electrodes when their signs disagree. Thus, the projected
variance from all the independent component processes is typically
larger than the scalp-recorded data (!).

Pecent variance accounted for (or its siblings) tells by how many
percent the scalp signal is reduced when the component(s) in question
are removed. Thus it has an upper bound of 100%. But if subtracting
the component(s) actually increases signal variance (e.g. if the
component contributions are of opposite sign to other components),
then a 'negative decrease' in the form of negative pvaf is returned.

Scott Makeig

On Nov 30, 2007 8:23 AM,  <louis.renoult at douglas.mcgill.ca> wrote:
>
> I just found on eeglab website that there was a bug in the envtopo()
> function
> https://sccn.ucsd.edu/eeglab/bugzilla/show_bug.cgi?id=398
>
> So I took the new envtopo file. However, I still do not understand how to
> interprete negative pvaf (nor sortvar values). For some selected epochs,
> even though I have components clearly pointing at the main ERP deflection
> (e.g., P1), they are sometimes associated with negative values (e.g., -68%)
> and not suming to 1.
>
> Thank you very much,
> Louis
>
>
> -----Louis Renoult/Hopital Douglas/Reg06/SSSS a écrit : -----
>
> Pour : eeglab at sccn.ucsd.edu
> De : Louis Renoult/Hopital Douglas/Reg06/SSSS
> Date : 28/11/2007 04:08PM
> Objet : How to read "ppaf" in pop_envtopo()
>
>
> Hello,
>
> While testing ICA on some of our data, we have been faced with the following
> question:
>
> We are studying visual ERPs effects (N100 and P100) and have applied ICA to
> a set of 28 channels data.  We are using the envtopo function to study which
> components make the largest contribution to the ERP in selected time
> windows. However, we are not sure which is the percentage of variance
> accounted for by each component. Here is a copy of our command window (also
> included is the resulting plot):
>
> pop_envtopo(EEG)
> Subtracting requested components from plotting data: 2 3 4 5 7 8 9 10 11 12
> 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
> Data epoch is from 0 ms to 136 ms.
> Plotting data from 0 ms to 136 ms.
> Comparing maximum projections for components:
> IC1 maximum mean power of back-projection: 0.610197
> IC2 maximum mean power of back-projection: 0.280823
>   in the interval  70 ms to  85 ms.
> Plotting envelopes of 2 component projections.
> Topo maps will show components:    1     6
>     with max var at times (ms):   76    79
>                   epoch frames:  147   150
>     Component sortvar in interval:  0.61 0.28
>     Summed component 'ppaf' in interval [  70   85] ms: -108.78%
>     Plot limits (sec, sec, uV, uV) [0,0.136,-2.10574,5.20694]
>
> ans =
> figure; pop_envtopo(EEG, [0  136] ,'limcontrib',[70 85],'compnums',[1
> 6],'title', 'Largest ERP components of Sub_-200_to_330','electrodes','off');
>
> We understand that the max of variance explained by component 1 occurs at
> time 76ms and 79ms for component 6, but do not know how to interprete
> sortvar or ppaf (and why ppaf is negative).
> Also, is there a way to relate a component ppaf and residual variance (RV,
> as mentionned in dipfit2) ?
>
>
>
> Thank you very much (and sorry to ask many questions),
> Louis
>
>
>
> Louis Renoult
> Ph.D. candidate
> Lab of Human Neurocognitive Sciences FBC-1
> Douglas Hospital Research Centre
> 6875 Boul LaSalle,
> Verdun,
> Montreal, Qc H4H 1R3
> Canada
> Tel (514) 761 6131 #3423
> Fax (514) 888 4099
>
>



-- 
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