After looking carefully at the code I found that there is indeed a
difference between the induced and evoked option which is in the order
of squaring (i.e. computing power) and averaging. In the induced case
each trial is multiplied with Gaussian window then the result is
squared and then accumulated over trials whereas in the evoked case
the average is computed first then multiplied with Gaussian and then
squared. According to Karl with the 'induced' way the mean of the
power is nonlinearly more sensitive to high-amplitude evoked (trial
specific) responses. Another difference can be that if the trial
average does not have consistent sign it will average out in the
'evoked' way but not in the 'induced'. So there is an explanation for
the differences you are seeing. However, I would still recommend you
to use the 'evoked' option since this is something established that
can be justified to the reviewers. The meaning and justification of
'induced' + Gaussian is not completely clear to me and the fact that
you can do it in SPM is just a by-product of having Gaussians and
wavelets projectors integrated in the same function.
On Mon, Aug 15, 2011 at 3:49 PM, Vladimir Litvak
<[log in to unmask]> wrote:
> Dear Urs,
> Maybe you have better eyes but with my limited vision I cannot see
> much difference between evoked and induced options in your results.
> This actually makes a lot of sense because the induced option is only
> different from the evoked if you specify a frequency band. As you can
> see in the plot the projector in both cases is just a gaussian window
> and that can be applied either before or after averaging with the same
> result. Perhaps there is just some kind of minor scaling difference
> that should be removed later anyway when exporting to images.
> Since as implied from your second question, you are interested in ERPs
> and want to relate your results to specific peaks in the ERP, it
> probably makes more sense for you to use the 'evoked' option because
> that corresponds exactly to this way of thinking. The induced option
> is more suitable for questions formulated in terms of time-frequency
> analysis, like 'gamma activity related to face processing'.
> Regarding your second question I don't think there is a way to answer
> it on a principled basis. I think your choice should be based on two
> non-principled criteria (1) you can show what you want to show (2) you
> can justify your choice in your paper so that it doesn't look
> arbitrary. That's what I think most people do. If you want to be
> really rigorous, you can use bootstrap i.e. resample with replacement
> the same number of trials from your data 100 times, compute the peak
> latency every time, get a distribution and take 95% confidence
> interval as your window.
> On 15 Aug 2011, at 13:32, Urs Bachofner <[log in to unmask]> wrote:
>> Dear Vladimir, dear SPM Users,
>> In 3D Source Localisation I'm unsure about the choices to be made in the window section.
>> I'm still working with my ERPs (ca. 180 good trials, -100 to 500 ms) and I'm expecting activation in the frontal regions.
>> 1. Induced vs. Evoked:
>> "The projectors generated at the previous step can either be applied to each trial and the results averaged (induced) or applied to the averaged trials (evoked)."
>> When I'm applying this to the whole time span of the trial (0 - 500 ms), I get much nicer activations in frontal regions when I choose the "induced" option.
>> Now since I'm now interested in P200 and P300 specifically, I choose a timespan +/- 20ms of the peaks of those components.
>> When doing so, all of a sudden I get nicer frontal results (darker and more definite spots) when choosing "evoked".
>> Could someone please explain what the advantages of the two modes are and in what situation which one should be applied?
>> 2. This leads to the question about the time window. I happen to know the exact ms of the Peak of my P200 and P300 when all trials are averaged.
>> What would be the best compromise between a short time window (+/- 4ms, maybe some peaks of some trials are missed) and a long time span (like +/- 50ms, other peaks and eeg components could be included)?
>> Thank you for your inputs if you've got any too share with me.
>> Urs Bachofner
>> NEU: FreePhone - 0ct/min Handyspartarif mit Geld-zurück-Garantie!
>> Jetzt informieren: http://www.gmx.net/de/go/freephone