[Eeglablist] whether or not subtracting baseline while calulating coherence

[arno]

Dear Wu Xiang,

our approach is not to subtract baseline coherence. One reason is that
the phase difference between the two processes might change with time,
possibly arising because of different synchronization phenomena. It
would be strange to subtract the coherence amplitude in the baseline (at
one phase) with the coherence after the stimulus presentation (at
potentially another phase).

However, the computation of significance in crossf currently uses the
baseline. Time-frequency regions where the coherence amplitude is larger
than the baseline are shown (and the rest of the time-frequency regions
are masked). Note that the comment above still apply (that is somewhat
strange to compare coherence amplitude of synchronization processes at
potentially different phases) and that this is a not a perfect solution
either.

One possible solution is to compare the distribution of phase difference
values with a random phase distribution. This will help mask regions and
highlight other time-frequency regions (during the baseline or not) that
show a consistent phase difference across trials (more consistent than
random given the p-value of interest). We will implement this solution
in the future in crossf.

Hope this helps,

Arno

>I wonder about that whether or not the baseline coherence should be
>subtracted while calulating coherence.
>
>For baseline may be different in different conditions and subjects, it
>was subtracted during ERP analysis. Also, the spectra-power of baseline
>was subtracted while calculating spectra-power, such as the
>baseline-nomalized ERSP in EEGLAB. However, why the coherence of
>baseline was not subtracted in most papers (I can not find a paper where
>the coherence of baseline was subtracted)?
>
>Would someone recommend some papers where the coherence of baseline was
>subtracted to me for reference? (It's important to me)
>
>Thanks very much
>
>  
>