non-parametric cluster analysis

Eric Maris e.maris at DONDERS.RU.NL
Wed Sep 22 22:25:00 CEST 2010


Dear Emanuel,


> I would like to ask a question regarding your article: Nonparametric
> statistical
> testing of EEG- and MEG-data in Journal of Neuroscience Methods 2007. I
> hope you will answer it.
> My question is about the following. I want to use the non-parametric
> cluster
> analysis, the one you described in your article, for my ERP analysis.
> In this article you quantified the difference between the two
> conditions
> (congruent and incongruent) by calculating sample specific t-values,
> based on
> the t-distribution. These sample specific t-values were calculated
> based on a
> two sided test statistic and an alpha of .05. Subsequently clusters
> were
> chosen based on the uncorrected critical t-value. This uncorrected
> critical t-
> value is also based on a two sided test statistic.
> Now my question. I understand that this uncorrected critical t-value
> threshold
> affects the sensitivity of the statistical test rather than the
> validity. So is it
> also possible to calculate the sample specific t-values belonging to
> the one-
> sided test statistic, instead of two-sided?

I guess you mean here whether one may also select samples for subsequent
clustering based on whether their t-statistics exceed the critical value of
a one-sided statistical test. Yes, this is allowed. However, when you use
the cluster-based permutation tests in Fieldtrip, you can do this with the
same function that (per default) does the selection on the basis of the
critical values for a two-sided statistical test. You only have to double
the value of the alphathresh-parameter (e.g., putting it at 0.10 such that
it uses 0.05 as the critical p-value for the selection of the samples), and
you can evaluate the permutation p-values of the clusters by comparing them
with your nominal alpha-level (typically, 0.05). Researchers that, unlike
you, are interested in both positive and negative clusters, must compare the
permutation p-values with half their nominal alpha-level (typically, 0.025).

Good luck,

Eric



The same question accounts
> for the
> uncorrected critical t-value used as threshold.







> Thanks in advance,
> Best,
> Emanuel
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users of
> the FieldTrip  toolbox, to share experiences and to discuss new ideas
> for MEG and EEG analysis. See also
> http://listserv.surfnet.nl/archives/fieldtrip.html and
> http://www.ru.nl/neuroimaging/fieldtrip.

----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.



More information about the fieldtrip mailing list