AW: [Eeglablist] statistics on coherence values

[Christopher Summerfield]

hi Gabor:

In addition to Johannes' excellent paper you might also want to have a
look at a recent approach we have
developed, to be published shortly in Neuroimage.  Summerfield,
C., & Mangels, JM: Coherent theta-band activity predicts item-context
binding during encoding.

Detailed information and relevant code are available at the following site:
www.columbia.edu/~cs2028/beast/beast.htm

briefly, nuisance variables such as volume conduction, low trial counts
(<30) and even just generalised phase-locking can contribute to your
coherence estimates.  one way to reduce the influence of these factors is
to shuffle and permute your electrode-trial structure in the following
way:

pointe estimate
        elec1 elec2
trial   1       1
        2       2
        3       3
        n       n

shuffled
	elec1 elec2
trial	1	3
	2	7
	3	2
	4	n, etc

to gain a distribution of overall coherence which is (reasonably)
independent of these nuisance factors.  The approach we adopt is to
convert both our point estimate coherence estimates and the mean of these
'shuffled' coherence estimates to z-scores, to put them on the same scale,
and subtract.  In simulations we have found that this method is highly
reliable and a good way to eliminate potential confounds.

With regard to direct significance testing on these measures, we adopt an
ICA-base approach to reduce electrode pairings (lots
of them) to spatially correlated components. these components reflect
clusters of electrodes which exhibit similar time-frequency structure to
the coherence across conditions.  You can then bootstrap your estimates
across time-frequency coherence space for each component, correcting only
for the number of components rather than the number of pairings  - a
considerable gain in statistical power.

good luck!
chris

Christopher Summerfield
summerfd at psych.columbia.edu

On Thu, 16 Dec 2004, Sarnthein Johannes wrote:

> Hi Gabor,
> you can have a look at a possible approach to this problem in our paper
> J. Sarnthein, H. Petsche, P. Rappelsberger, G.L. Shaw and A. von Stein "Synchronization between prefrontal and posterior association cortex during working memory tasks in humans", Proceedings of the National Academy of Sciences 95 (1998) 7092-7096.
> Best,
> Johannes
>
>
> -----Ursprüngliche Nachricht-----
> Von: eeglablist-bounces at sccn.ucsd.edu [mailto:eeglablist-bounces at sccn.ucsd.edu]Im Auftrag von Stefanics Gabor
> Gesendet: Mittwoch, 15. Dezember 2004 16:13
> An: eeglablist at sccn.ucsd.edu
> Betreff: [Eeglablist] statistics on coherence values
>
>
> Dear All,
>
> I am analyzing channel cross coherence for 32 channel eeg data, I have the
> resulting matrices of the crossf function for all the possible channel
> combinations with a 0.05 bootsrap significance mask applied on the prestimulus
> baseline. My question is what is the correct statistical procedure to compare
> different experimental conditions? In many papers I have read that coherence
> values were Fisher-z-transformed before paired Wilcoxon tests on individual
> channel-pairs or before ANOVA for a number of channel pairs. So my above
> question is manifold:
> - Is it necessary to apply Fisher-z-transformation on my data?
> - Which statistical test is the optimal? Because if I do the Wilcoxon test,
> the significance levels are not corrected for multiple comparisons, but if I
> make ANOVA I can test only a subset of the channel pairs to avoid loss of
> statistical power of the ANOVA. Is there any way to avoid this trap?
>
> Thanks for any help and advice in advance!
>
>
> Gabor Stefanics
>
> _______________________________________________
> eeglablist mailing list eeglablist at sccn.ucsd.edu
> http://sccn.ucsd.edu/mailman/listinfo/eeglablist
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
>
> _______________________________________________
> eeglablist mailing list eeglablist at sccn.ucsd.edu
> http://sccn.ucsd.edu/mailman/listinfo/eeglablist
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
>