[FieldTrip] Inconsistency between stat.mask and p-values in ft_freqstatistics
Guillermo Sanchez-Garrido Campos
gsancam at upo.es
Tue Mar 26 09:21:54 CET 2024
Jan,
First of all, I did check Mikkel suggestions about the tutorial of "the
p-values associated with the cluster"
(https://www.fieldtriptoolbox.org/workshop/oslo2019/statistics/).
However, as I mentioned in my first mail, my data comes from Neuropixels
1.0 recordings, not EEG data from several electrodes around the brain,
but a single tract with 384 channels across 3.8 mm depth (here you have
the link of the first article for further information in case you are
not familiarized with the system:
https://www.nature.com/articles/nature24636). Regarding clusters, after
consulting with a colleague who has more experience with FieldTrip, it
seems that using clusters may not be appropriate for our data analysis.
Our recording system, being based on a single tract (Neuropixels), may
not align well with the typical cluster-based analysis, which is more
commonly used for EEG and recordings from multiple brain regions.
Therefore, we opted for a different approach that better suits our data
structure. According to this, information about clusters was not useful
to this matter, or that is what we thought, maybe we could argue this.
So, as he suggested me, after discussing mask, prob, and cirange
variables after the statistic script, I wrote another issue in the
Github. Just to clarify my little chat with Mikkel, in case I did not
clarify it 😊.
Regarding the use of "fdr" for multiple comparisons, initially, I wanted
to perform a more straightforward analysis of the data without applying
fdr or any other method to have a preliminary view of the results.
However, even after applying fdr, I still encounter the correlation
between mask and cirange, and the inconsistency between mask and prob.
Now, regarding your inquiries:
*
Actually yes, there is new information in the last issue, if you look
carefully. As I checked the "ft_statistics_montecarlo",
"ft_statistics_analytic", and "ft_freqstatistics" in the Fieldtrip
github in order to know how the mask variable is calculated, no clues I
found about this topic. The only part I found about the mask variable
was this: stat.mask = stat.prob<=cfg.alpha;
(https://github.com/fieldtrip/fieldtrip/blob/release/ft_statistics_montecarlo.m#L455
and
https://github.com/fieldtrip/fieldtrip/blob/release/ft_statistics_analytic.m)
when no correction was utilized. Consequently, rechecking the stat
variable that I got from ft_freqstatistics, I realized the cirange and
mask consistency that I have mentioned in the last issue. At this point,
it was my last message with Mikkel, as I did not know if this
correlation is a bug from the script or maybe I did commit some mistake
with my input information.
*
According to the cirange, the same I said before, I do not know if I
have to but the point is that there is a correlation with the mask
variable. When it appears a value of 0 at cirange (let's say at
positions x and y, in the matrix), at those same positions, it appears a
"1" at mask, independently if prob(x,y) < cfg.alpha or not. Repeating
myself in order to clarify, I do not know if this is a random
correlation or if it happens for a bug in the script, but I thought it
was worth asking 😊.
*
I may have missed one line from my code, but yes, you are right, I have
this following line regarding the number of randomizations:
"cfg.numrandomization = 'all';". Another fault from my part and my code,
the "data" you mentioned was referring to "datafftpost" but I may have
copied an old version of the code, so sorry for that.
*
Making statistics with 3 subjects is not definitive, obviously, but you
can have an idea of what differences you have between experimental and
sham group (also if there are any), so it may be helpful although you
cannot believe it until the N is more representative. Also, I was trying
to check if the code was working properly with my data.
Thank you for your support in this matter, and I appreciate any feedback
you can provide. I hope we can discuss any further questions that may
arise from my data calmly and as adults.
Kind regards,
Guille
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20240326/e14b19cd/attachment.htm>
More information about the fieldtrip
mailing list