Hi Francois,
I wanted to ask something about z -scores, for baseline normalization. I understood from this forum that you mostly calcualte z-scores on averaged data. I am not sure if I am correct in this. I was trying to do baseline normalization for each trial, but i think in the end I could not see it in final averages. That is, i did not see any difference between averages with and without z scores calculation. What do you think about this, is it the best to apply baseline normalization on averages?
Also, when we apply z-score, what should be written on the scale? I always see “No units”, so I am afraid that I am maybe doing something totally wrong.
Thank you again.
Svjetlana
Hi Svjetlana,
When importing MEG recordings, you have an option “remove DC offset”, sometimes referred to as “baseline removal” or “baseline correction”. This calculates the average of the signal for each sensor over a baseline period and and removes it from all the imported epoch. This is to get rid of the arbitrary DC level that is recorded by the MEG sensors.
After calculating the source maps with a minimum norm model, we recommend to run a Z-score normalization to increase the visibility of some deeper sources who might have low values just because of the way the minimum norm method works (more superficial sources always get higher MNE values).
This process calculates the average m and the standard deviation s over a baseline period (usually a pre-stim time window in the case of evoked related study). Then it centers the values (subtracts the average) and normalizes (divides by the standard deviation). If a source has high values during the baseline, its standardized value will be low.
The Z-score values are ratios (Value - Avg)/Std. In terms of units, this ratio is unit-less: (pAm - pAm)/pAm = no units.
They have to be interpreted in “number of times the standard deviation”.
Hope this answers all your questions.
Francois
Hi Francois,
I just would like to clarify something so that I am sure that I am doing it correctly.
When I calculate time frequency (morlet wavelet, on sources), and then do “Save average time-frequency maps (across trials)” - my question is: does it calculate time frequency for each trials and then save averages, or first average and then calculate TF. After I see results, it seems it it is first averaged then calculated TF power. If I want to have time frequency for each trials, do I need to put as Output: Save individual time-frequency maps (for each trials), and then average across trials. I went through tutorials, I hope i did not miss something, but just wanted to be sure.
Thanks.
Svjetlana
Hi Svjetlana,
The option “Save average time-frequency maps” does the following:
- calculate the time-frequency decomposition for each individual trial (but does not save the results)
- average the power of the TF maps
It is equivalent to run the process “time-frequency” with the option “Save individual time-frequency maps” immediately followed by the process “Average files” (except that this option keeps all the TF maps of all the trials in the database)
Francois
Thank you very much Francois!