In my analysis on a MEG data, after computing the source reconstruction from the sensors using LCMV beamformer (using individual anatomy), I smoothed the sources and projected them on default anatomy before computing the time-frequency plots using Morlet wavelengths (and perform further analysis such as averaging...).
However, I saw in a paper a similar analysis to mine where after computing the source reconstruction from the sensors, they computed the time-frequency plots, performed averaging across trials, then they smoothed the TF plots and projected them on default anatomy.
I was wondering if the two methods should give similar results or if one is better than the other. I am worrying that in my method, the raw data became fairly distorted by the time I compute the TF plots, which may affect the accuracy of these plots, but I don't know how significant the differences would be.
I smoothed the sources and projected them on default anatomy before computing the time-frequency plots using Morlet wavelengths (and perform further analysis such as averaging...).
However, I saw in a paper a similar analysis to mine where after computing the source reconstruction from the sensors, they computed the time-frequency plots, performed averaging across trials, then they smoothed the TF plots and projected them on default anatomy.
Exchanging the order of the smoothing and projection on the template should lead to similar results.
Exchanging the computation of the TF power and the smoothing/projection would lead to different results, but I'm not sure one is better than the other.
For one subject, try both and see how much they differ.
If you process all the subjects in the same way, it should not introduce much bias in the results.
Note that the statistical differences visible at the source level are also visible at the sensor level, with a much lighter processing pipeline. Make sure all your source-level findings match the sensor-level power results.