[FieldTrip] second-level statistical inference

[Sara Bögels]

Hi Eric,

I understand that is another option. However, I do not see how to look 
at an interaction between two variables in this way. Do you have any 
advice if I want to do that?

Thanks,
Sara

On 30/11/2011 19:21, Eric Maris wrote:
> Hi Sara,
>
>
> I'm replying to your initial question (somewhere below in this email).
>
> Why don't you calculate per-subject averages in the two within-subject
> experimental conditions (using timelockgrandaverage with
> keepindividual='yes'), and then do your second-level inference on these
> averages? See also the Fieldtrip statistics tutorials.
>
> Best,
>
> Eric Maris
>
>
>> -----Original Message-----
>> From: Sara Bögels [mailto:s.bogels at psy.gla.ac.uk]
>> Sent: vrijdag 25 november 2011 12:52
>> To: fieldtrip at donders.ru.nl
>> Subject: Re: [FieldTrip] second-level statistical inference
>>
>> Hi Arjen,
>>
>> Thank you very much for your answer. That sounds good, but step 2 does
>> not work straightforwardly, since matlab gives the error message that it
>> cannot find an avg field (which would not be in the structure created by
>> ft_timelockstatistics). Just saying cfg.parameter = 'stat' does not
>> work. I tried to get around that by inserting an avg field which is the
>> same as the stat field for each participant. Matlab also asked for an
>> fsample field, which I inserted from an earlier datafile. Then it
>> worked. Is it ok to do this?
>>
>> I did step 3 as well, using the field individual (which you get by
>> keepindividuals = 'yes'). In step 4, I should just use cfg.statistic =
>> 'depsamplesT', right (because the variables are within subject)?
>>
>> Thank you!
>> Sara
>>
>> On 25/11/2011 09:04, Stolk, A. wrote:
>>> Hi Sara,
>>>
>>> If I understand correctly, you want to test intra-subject differences
>> (between conditions) at the second level? This would require the following
>> steps:
>>> 1) subject-level statistics, which you have done already
>>>
>>> 2) grandaverage all these, with keepindividuals=yes.
>>>
>>> 3) copy the output of the grandaverage (into a dummy variable), and
>> replace the fields containing the subject T-values with zeros (for
>> timelock
>> data this may be the trial fields?)
>>> 4) again timelockstatistics, as in step 1, now with the variables
>>> following
>> step 3. this should give you the resulting statistics of contrasting
>> intra-
>> subject differences vs. null at the group level.
>>> Hope this helps,
>>>
>>> Arjen
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> ----- "Sara Bögels"<s.bogels at psy.gla.ac.uk>   schreef:
>>>
>>>> Van: "Sara Bögels"<s.bogels at psy.gla.ac.uk>
>>>> Aan: fieldtrip at donders.ru.nl
>>>> Verzonden: Donderdag 24 november 2011 15:49:17
>>>> Onderwerp: [FieldTrip] second-level statistical inference
>>>>
>>>> Hi all,
>>>>
>>>> I have been trying to do second-level statistical inference (as
>>>> described in one of the FAQs) on ERFs, but I am not sure whether I am
>>>>
>>>> doing everything correctly.
>>>>
>>>> In the first step I calculate the T-values for the difference between
>>>>
>>>> two conditions (twice), which are between items, with
>>>> ft_timelockstatistics. I put the output of all participants in a cell
>>>>
>>>> (called 'stat1a' and 'stat1b'). (I tried to use
>>>> ft_timelockgrandaverage
>>>> to combine the subjects together but it needs a field avg).
>>>>
>>>> Then I use ft_timelockstatistics again but  subject level. I first
>>>> want
>>>> to look at the difference between the two conditions. This difference
>>>> is
>>>> reflected in the T-values of the first step so I create a dummy which
>>>> is
>>>> the same as 'stat1' but I replace all the values in the field 'stat'
>>>> per
>>>> participant with zeros. Then I call (with appropriate cfg
>>>> parameters):
>>>>
>>>> stat2a = ft_timelockstatistics(cfg,stat1a{:},dummy{:});
>>>> stat2b = ft_timelockstatistics(cfg,stat1b{:},dummy{:});
>>>>
>>>> To compare the two differences (stat1a and stat1b) and thereby look at
>>>>
>>>> an interaction, I call:
>>>>
>>>> stat2a-b = ft_timelockstatistics(cfg,stat1a{:},stat1b{:});
>>>>
>>>> I am uncertain whether the dummy works (or is there a way to compare
>>>> the
>>>> t-values to zero directly?) and whether the stat1a{:} trick works with
>>>>
>>>> ft_timelockstatistics.
>>>>
>>>> Thanks in advance for your answer.
>>>>
>>>> Sara
>>>>
>>>>
>>>> _______________________________________________
>>>> 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