On 20 September 2010 10:51, Alberto Inuggi
<[log in to unmask]> wrote:
Hi all.
I'm trying to use melodic to denoise an event related task.
I provided melodic the model and contrast file.
melodic reports for each component the F-test value (and the significance of all the contrast created) of the GLM on time series.
can this value be the correct/only/most important way to decide if discard a component or not ????
If I understand you correctly - that you want to use your model, including the components of contrasts of interest, to determine which ICA components not to filter out. This is not valid, as you are selecting your data based on (at least in part) the strength of an effect you want to subsequently test for and report. Even if your data had no signal, your denoised data could likely show a significant effect.
hth
Eugene
