[Eeglablist] How ERSP, ITC comparisons are computed?
Arnaud Delorme
arno at ucsd.edu
Wed Jan 30 15:36:41 PST 2008
The comparison for ERSP is a simple difference of log-power values after
common baseline subtraction (common subtraction if 2 conditions).
To compute statistical difference, each time-frequency point is
considered independently. Say A(f,t) is the ensemble (across trials) of
time-frequency decomposition for condition A at frequency f and time t
(noted simply A below) and B(f,t) is the ensemble of time-frequency
decomposition for condition B (f and t are fixed in the whole
procedure). C is the union of time-frequency decomposition trials A and
B. We take a randomly elements in C (each element is only taken once) to
construct two permutation ensemble A' and B' (A' being the same size as
A and B' the same size as B but both A' and B' usually contains elements
both from A and B). We then recompute the difference between the average
power of trials in A' and the average power of trials in B'.
We repeat the procedure in between 200 and 5000 times to build a
distribution of difference power values under the hypothesis that A and
B are drawn from the same population of trials (thus the original power
difference should lie within the permutation distribution). If it does
not lie within this distribution (for instance it is the tail of the
distribution - lower 2.5% or the upper 2.5% - then we can say that we
reject the Null hypothesis at 5%). It is easier to explain with graphics.
The same is performed for ITC.
Hope this helps,
Arno
Soyoung Kwon wrote:
> Dear all,
>
> I wonder how ERSP and ITC comparisons are computed while using newtimef.
>
> Is ERSP comparison merely a mathematical difference between two
different conditions?
> Or is it statistically calculated?
> If so, please let me know the simple protocol of statistical method.
> Equation is also welcome.
>
> Simliar question is arising with ITC comparisons.
>
> Please give me any advice regarding ERSP, ITC and differences of these.
>
> Thanks in advance.
>
>
> Soyoung Kwon
More information about the eeglablist
mailing list