[Mne_analysis] Unbalanced number of trials between individuals: problem in sources

[Eric Larson]

If you plan to take the absolute value of the data (or the norm of the
three components) it's advisable to equalize the event counts. This
nonlinear operation takes what is typically thought of as being zero-mean,
normally distributed noise, and turns it into a folded normal distribution.
A folded normal distribution has a mean that is dependent on the standard
deviation of the un-folded version of the distribution. Different trial
counts yield different effective noise standard deviations, which then
become different noise *means* once the distribution gets folded by the
abs()-like operation.

In other words, if you don't equalize the trial counts before a nonlinear
norm operation, you can end up with potentially biased estimates. At one
point I tried correcting for this bias, but was unsuccessful -- the noise
distribution, even if turned zero-mean, was skewed. Perhaps there is some
good way to deal with it, but subsampling the data (equalizing trial
counts) has seemed to be the safest thus far. It's probably not a big deal
if it's 35 vs 40 trials, but if you have an unbalanced design of e.g. 30
trials in one condition and 60 trials in another, it's more likely to
matter.

Eric


On Thu, Mar 3, 2016 at 11:02 AM, Laetitia Grabot <laetitia.grabot at gmail.com>
wrote:

> Ok, thanks, I was still applying the old advice of equalizing epochs
> between conditions! But so, why in certain statistical examples of the
> website, you're still equalizing conditions (like in the Reapeated-measures
> ANOVA in sources)?
>
> 2016-03-03 16:43 GMT+01:00 Denis-Alexander Engemann <
> denis.engemann at gmail.com>:
>
>> Yes that's the idea. Amplitudes of the inverse solution are scaled by the
>> number of trials (.nave). This should work as an heuristic unless you
>> compare rare to frequent events (e.g. oddball).
>> Hope this helps.
>>
>> On Thu, Mar 3, 2016 at 4:38 PM, Laetitia Grabot <
>> laetitia.grabot at gmail.com> wrote:
>>
>>> So does that mean that we need not to equalize epochs between conditions
>>> (within one subject)?
>>>
>>> 2016-03-03 16:23 GMT+01:00 Denis-Alexander Engemann <
>>> denis.engemann at gmail.com>:
>>>
>>>> Hi Laetitia,
>>>>
>>>> I I'm understanding you correctly this should not be an issue for
>>>> evoked data as long as you are using the .average method that will tell the
>>>> inverse routines how to scale the data via its .nave attribute.
>>>> Did you have any particular problems?
>>>>
>>>> Cheers,
>>>> Denis
>>>>
>>>> On Thu, Mar 3, 2016 at 3:09 PM, Laetitia Grabot <
>>>> laetitia.grabot at gmail.com> wrote:
>>>>
>>>>> Dear MNE-users,
>>>>>
>>>>> I am aware of the fact that when contrasting two conditions, the
>>>>> epochs within each condition should be equalized before reconstructing in
>>>>> source space, otherwise the SNR would be different between the two
>>>>> conditions. That also means that if two subjects have a different number of
>>>>> epochs, the SNR of each subject will be different. It is often the case
>>>>> that the number of epochs doubled between subjects (50 vs. 100) so I guess
>>>>> that it is an issue when looking at group averaged stc for instance.
>>>>> A possible solution would be to normalize the stc to correct for the
>>>>> number of trials. What do you think of that? What would be the proper
>>>>> normalization?
>>>>>
>>>>> Thanks,
>>>>> Best,
>>>>> Laetitia G.
>>>>>
>>>>> _______________________________________________
>>>>> Mne_analysis mailing list
>>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>>>
>>>>>
>>>>> The information in this e-mail is intended only for the person to whom
>>>>> it is
>>>>> addressed. If you believe this e-mail was sent to you in error and the
>>>>> e-mail
>>>>> contains patient information, please contact the Partners Compliance
>>>>> HelpLine at
>>>>> http://www.partners.org/complianceline . If the e-mail was sent to
>>>>> you in error
>>>>> but does not contain patient information, please contact the sender
>>>>> and properly
>>>>> dispose of the e-mail.
>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> Mne_analysis mailing list
>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>>
>>>>
>>>> The information in this e-mail is intended only for the person to whom
>>>> it is
>>>> addressed. If you believe this e-mail was sent to you in error and the
>>>> e-mail
>>>> contains patient information, please contact the Partners Compliance
>>>> HelpLine at
>>>> http://www.partners.org/complianceline . If the e-mail was sent to you
>>>> in error
>>>> but does not contain patient information, please contact the sender and
>>>> properly
>>>> dispose of the e-mail.
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Mne_analysis mailing list
>>> Mne_analysis at nmr.mgh.harvard.edu
>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>
>>>
>>> The information in this e-mail is intended only for the person to whom
>>> it is
>>> addressed. If you believe this e-mail was sent to you in error and the
>>> e-mail
>>> contains patient information, please contact the Partners Compliance
>>> HelpLine at
>>> http://www.partners.org/complianceline . If the e-mail was sent to you
>>> in error
>>> but does not contain patient information, please contact the sender and
>>> properly
>>> dispose of the e-mail.
>>>
>>>
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>
>>
>> The information in this e-mail is intended only for the person to whom it
>> is
>> addressed. If you believe this e-mail was sent to you in error and the
>> e-mail
>> contains patient information, please contact the Partners Compliance
>> HelpLine at
>> http://www.partners.org/complianceline . If the e-mail was sent to you
>> in error
>> but does not contain patient information, please contact the sender and
>> properly
>> dispose of the e-mail.
>>
>>
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>
>
> The information in this e-mail is intended only for the person to whom it
> is
> addressed. If you believe this e-mail was sent to you in error and the
> e-mail
> contains patient information, please contact the Partners Compliance
> HelpLine at
> http://www.partners.org/complianceline . If the e-mail was sent to you in
> error
> but does not contain patient information, please contact the sender and
> properly
> dispose of the e-mail.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20160303/ae677825/attachment.html