[Eeglablist] Statistics for ERSP and ITC
Delorme, Arnaud
adelorme at ucsd.edu
Sun Jul 19 18:11:05 PDT 2020
Dear Laurent,
In the EEGLAB STUDY interface, you can plot both subject average and single subject. I would look into single subject ERSP/ITC and also not masked for significance to see if there is any trend. For example, you say you were expecting to see a power increase. Can you see that for even a subset of subjects.
Best wishes,
Arno
> On Jul 17, 2020, at 11:59 AM, Makoto Miyakoshi <mmiyakoshi at ucsd.edu> wrote:
>
> Dear Laurent,
>
> The baseline correction in epoching is one thing, and the baseline
> correction for ERSP/ITC is quite another. The former has no effect on the
> latter.
> It makes sense that you changed the former and you did not see any
> difference in the latter.
>
> To manually specify baseline period for ERSP/ITC for STUDY, you have to use
> optional input in STUDY GUI.
> Please try it again while using 'baseline', [-1500 -1000] so that mean
> value (in dB) between -1500 to -1000 will be used for subtraction.
> I predict that this will change the ERSP results.
> Meanwhile, ITC result will be the same. There is nothing you can do for the
> ITC result.
> I care 'ITC randi' and 'ITC spike' have different background noise
> distributions... at least I can tell that the color scales are different
> between the two plots, if you say the both conditions have the same/similar
> number of trials. Maybe you want to show the color bar for the both plots
> to make sure of it?
>
> Makoto
>
>
> On Fri, Jul 10, 2020 at 12:20 AM Laurent Sheybani <Laurent.Sheybani at unige.ch>
> wrote:
>
>> Thank you Makoto for your help !
>>
>> I actually did not apply a baseline correction (given the high signal to
>> noise ratio, and the fact that I am just checking if I handle correctly
>> Eeglab). For these steps I use EEGLAB's gui. When the window 'Baseline
>> removal' pops up after extracting epochs, I just press cancel (as it is
>> written to press Cancel if one does not want to remove the baseline).
>> However, now that you asked, I tried again, this time extracting epochs
>> from -3000 to +1000 ms around markers, and correcting for baseline, using
>> as baseline the period of -3000 ms to -2000 ms (the statistics are applied
>> after -2000 ms). The results are similar. I also selected an equal number
>> of test and control trials, and again obtained the same results (I can
>> attached them to this email if you need, but there are really just the
>> same...). I feel that I am missing a step in the processing pipeline but I
>> can't find where ?
>>
>> If you have any further idea, I would appreciate again your help !
>> Best
>> Laurent
>>
>>
>>
>> -----Message d'origine-----
>> De : eeglablist <eeglablist-bounces at sccn.ucsd.edu> De la part de Makoto
>> Miyakoshi
>> Envoyé : samedi, 4 juillet 2020 01:30
>> À : eeglablist at sccn.ucsd.edu
>> Objet : Re: [Eeglablist] Statistics for ERSP and ITC
>>
>> Dear Laurent,
>>
>> Wow these are two counterintuitive plots, I agree with you.
>> If I were you, I would check the following two things
>>
>> 1. What's your baseline period in these time-frequency plots? If you use
>> the default, all negative latencies are included, but that would contain
>> half of your spike period.
>> 2. How many trials do you have for each condition? Particularly for ITC,
>> SNR of 'background' is directly affected by the number of trials (since
>> the
>> resultant vector length is normalized to 0-1, data noisiness/quietness
>> in
>> the 'background' is a direct function of number of trials. If two
>> conditions compared have different number of trials, you cannot make
>> exact
>> comparison--sometimes all the 'background' (which I means is
>> time-frequency
>> window of non-interest) will light up simply because of different noise
>> level. I'm not 100% sure though if this is the case in your example).
>>
>> I have never used 'use all single trials...' option, so I can't say for
>> sure about its effect, but assuming that it is irrelevant here.
>>
>> Generally speaking, ITC is a cranky measure compared with ERSP. You should
>> be careful.
>>
>> If it does not make sense, please let me know.
>>
>> Makoto
>>
>> On Mon, Jun 29, 2020 at 10:28 AM Laurent Sheybani <
>> Laurent.Sheybani at unige.ch>
>> wrote:
>>
>>> Dear Eeglabers,
>>>
>>> I am working on an EEG dataset of a cognitive task and I would like to
>>> see whether there is an increase in power and/or ITC around subjects'
>> answers.
>>> Before analysing these data, I am getting used to the EEGLAB gui using
>>> epileptic data that I acquired in the past, just to see whether I
>>> process my data correctly in eeglab. I marked all interictal epileptic
>>> discharges, so that I know when there is an increase in power.
>>>
>>> What is very strange is that I find no significant increase in power
>>> over the whole window (-2000 to 1000 around interictal epileptic
>>> discharges,
>>> 3-50 Hz), while the whole window is significant for ITC. I would have
>>> expected a significant increase in power around interictal epileptic
>>> discharges (as I see it in individual, raw LFP activity) and if there
>>> was an increased ITC, I would have expected it to be more specific (in
>>> terms of time and frequency range).
>>>
>>> The pipeline that I follow is:
>>>
>>> * Load data into EEGLAB from a .mat file (1x28822528): 1 electrode,
>>> 28822528 timeframes at 16000 Hz sampling frequency.
>>> * Load the event file
>>> * Downsampling the data to new sampling frequency = 100 Hz.
>>> * Export epochs of "spike" and "control" period (-2000 to 1000)
>>> * Load these .set file into a study (1 subject, 1 session, 2
>>> conditions)
>>> * Study design: Categorical variable: condition - Values (control -
>>> spike)
>>> * Precompute channel measures, selecting ERPs, power spectrum,
>>> ERP-image, ERSP and ITC (options for ICA are not chosen, as I have
>>> only 1 EEG trace), I keep all default parameters
>>> * Plot channels measures:
>>> * Stats:
>>> * I check "Compute 1st independent variable statistics if any"
>>> and "Use single trials for statistics" (I have only 1 subject)
>>> * Use EEGLAB statistics, permutation statistics, FDR
>> correction,
>>> statistical threshold 0.05
>>> * Params: I do not change the parameters.
>>>
>>> As you can see (power file:
>>> https://urldefense.com/v3/__https://www.dropbox.com/s/rwh3b2u8doc8oab/
>>> power.jpg?dl=0__;!!Mih3wA!R-vXecxt4S9nEJwpZ1Y3TT5lFDzVkvCPQW8ctGvZlHkO
>>> pF7Adw-hqNe9c3QF6ZzjUvrFqg$
>>> ) (ITC file:
>>> https://urldefense.com/v3/__https://www.dropbox.com/s/npa6f7llnv6tbhl/
>>> ITC.jpg?dl=0__;!!Mih3wA!R-vXecxt4S9nEJwpZ1Y3TT5lFDzVkvCPQW8ctGvZlHkOpF
>>> 7Adw-hqNe9c3QF6Zwd2CJ8HA$
>>> ) there is no increase in power, while there is a non-specific change
>>> in ITC.
>>>
>>> Would any one know where in my processing I did a mistake ? It seems
>>> that the ERSP/ITC plots themselves are ok, but the p-values plot are not
>> ?
>>>
>>> Thanks in advance for your help !
>>>
>>> -----------------------------
>>> Dr. Laurent Sheybani, Ph.D
>>> Médecin interne en Neurologie Clinique Clinique de Neurologie Hôpitaux
>>> Universitaires de Genève Rue Gabrielle-Perret-Gentil, 4
>>> 1205 Genève
>>> Switzerland
>>>
>>> _______________________________________________
>>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>>> To unsubscribe, send an empty email to
>>> eeglablist-unsubscribe at sccn.ucsd.edu
>>> For digest mode, send an email with the subject "set digest mime" to
>>> eeglablist-request at sccn.ucsd.edu
>>>
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to
>> eeglablist-request at sccn.ucsd.edu
>>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu
More information about the eeglablist
mailing list