[FieldTrip] assessing significance in using ft_timelockanalysis results

[Kanal Eliezer]

Maybe I had been using it incorrectly, then. In my study, I have numerous datafiles for each subject (each block gets split into it's own .fif file, due to file size), and I use those functions to combine the preprocessed and timelocked datafiles. Should I be doing this differently?

Elli


On Aug 30, 2011, at 3:13 PM, Eric Maris wrote:

> Hi Elli,
> 
> 
> 
>> Thanks for the response. It looks like the fixed vs random analysis is
> exactly
>> what I'm referring to. From what I understood, it looks like the
> difference
>> really only shows up in the variance of the resultant distribution; with a
>> random, the variance also takes into account the betwee-subjects variance.
>> Is there a way to specify whether I want to do a fixed or random effects
>> analysis in FieldTrip when I'm running ft_timelockgrandaverage or
>> ft_freqgrandaverage? Thanks!
> 
> 
> I don't get this. Ft_timelockgrandaverage and ft_freqgrandaverage will only
> be called when a random effects analysis is performed.
> 
> 
> Best,
> 
> Eric
> 
> 
>> 
>> Elli
>> 
>> p.s. - In case anyone else is trying to figure this out, chapter 12 of
> Friston's
>> book "Statistical Parametric Mapping" does an excellent job explaining the
>> difference between fixed and random analyses, as well as how to implement
>> it algorithmically.
>> 
>> 
>> 
>> On Aug 26, 2011, at 3:00 PM, Eric Maris wrote:
>> 
>>> Hi Kanal,
>>> 
>>>> The event related statistics tutorial
>>>> (http://fieldtrip.fcdonders.nl/tutorial/eventrelatedstatistics) talks
>>> about
>>>> assessing significance parametrically by running t-tests on pooled
>>>> timelockanalysis data. My question is, does the fact that the averages
>>> were
>>>> created from N trials make a difference? If I'm condition A has twelve
>>>> averages and condition B has another twelve, and each average contains
>> 70
>>>> trials, is there a way to "inform" the statistical test that the power
> in
>>> this
>>>> dataset is greater than 24? Is this only possible if I run the t-test
>>> comparing
>>>> each set of 840 (70*12) trials?
>>>> 
>>>> I'm also curious whether this is possible with non-parametric analyses,
> as
>>>> well. Thanks -
>>> 
>>> In an analysis over subjects (called random-effects analysis in the fMRI
>>> literature), "informing" the statistical test about the number of trials
> per
>>> condition only makes sense if this number is different for the two
>>> conditions. I propose that you have a look the fMRI papers that deal
> with
>>> the issue of fixed-versus-random effect analyses. The conceptual issues
>>> involved are the same in fMRI and electrophysiology.
>>> 
>>> 
>>> Best,
>>> 
>>> Eric Maris
>>> 
>>> 
>>> 
>>> 
>>>> 
>>>> Elli Kanal
>>>> 
>>>> 
>>>> --------------------
>>>> Eliezer Kanal, Ph.D.
>>>> Postdoctoral Fellow
>>>> Center for the Neural Basis of Cognition
>>>> Carnegie Mellon University
>>>> 4400 Fifth Ave, Suite 110A
>>>> Pittsburgh PA 15213
>>>> P: 412-268-4115
>>>> F: 412-268-5060
>>>> 
>>>> 
>>>> _______________________________________________
>>>> fieldtrip mailing list
>>>> fieldtrip at donders.ru.nl
>>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>> 
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>> 
>> 
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> 
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip