[FieldTrip] Cluster-level permutation test statistics

[Eelke Spaak]

Hi Valerie,

The crucial thing to note here is that it does not make sense to pair
any individual trial in condition A with another individual trial in
condition B. There are no pairs, the design is
between-units-of-observation, thus a two-sample (= independent) t-test
is used. Note furthermore that when doing this test you are testing a
hypothesis about a population of trials, and not about a population of
people. If you want to test a hypothesis about a population of people,
you would typically compute some average value for condition A and
condition B for each subject, and then do a test over subjects. Note
that here you would do a paired-sample (= dependent) t-test, since now
it does make sense to pair each value in condition A with another
value in condition B (namely, those belonging to the same subject).
Now you have one value per condition per subject, and each subject was
tested in two conditions.

Does this help?

Best,
Eelke

On 3 October 2013 16:22, Valerie Nunez <vnune at hunter.cuny.edu> wrote:
> Hello,
>
> I'm a little confused about the notion of unit of observation (UO) as
> subject or trial in the cluster_permutation_timelock tutorial:
> http://fieldtrip.fcdonders.nl/tutorial/cluster_permutation_timelock
>
> Specifically, I don't quite understand how the example of congruent and
> incongruent trials for a single subject is considered a between-trials
> design and therefore uses an independent t-test. If two trials have taken
> place for one subject, why isn't this considered a within-subject design,
> thereby requiring a dependent-samples t-test?
>
> What am I missing?
>
> Thanks,
>
> Valerie
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip