[FieldTrip] A question about the (sensor, freq, time)-sample permutation test in your paper

Robert Oostenveld r.oostenveld at donders.ru.nl
Thu Jan 30 09:21:31 CET 2020

Dear Yijun, also CC to the FieldTrip mailing list

The FieldTrip plotting functions - but also the way we prefer to plot the results - have changed over the years. The ft_clusterplot function is limited in how it can plot the data: the sequence of topographies have to be averaged either over time, or over frequency. And if you have channel*time*frequency data, averaging over time or frequency collapses the cluster in that dimension. In that case the details of the cluster in that dimension get lost, e.g. when averaging over frequencies you would loose information that for low frequencies the cluster might span different channels than for high frequencies. 

Rather than ft_clusterplot, I recommend that you write your own for-loop, and use ft_selectdata with avgovertime or avgoverfreq (and keeptimedim=no or keepfreqdim=no, please see the code for that) to average your selections and plot those using ft_topoplotER.  After averaging over a dimension, you will see that stat.mask is not 0 or 1 any more, but also can be a number in between if it is not part of the cluster all the time (or for all ferquencies). 

But also please note this FAQ http://www.fieldtriptoolbox.org/faq/how_not_to_interpret_results_from_a_cluster-based_permutation_test/ <http://www.fieldtriptoolbox.org/faq/how_not_to_interpret_results_from_a_cluster-based_permutation_test/> and don’t overinterpret the clusters.

best regards,

PS please address folow-up questions regarding the use of FieldTrip on its http://www.fieldtriptoolbox.org/discussion_list/ <http://www.fieldtriptoolbox.org/discussion_list/>. There are a lot of knowledgeable people on it that can help.

> On 30 Jan 2020, at 00:31, Yijun Ge <xxx> wrote:
> Hi Eric and Robert,
>    I hope this email finds you well! 
>    My name is Yijun, a PhD student in the University of Minnesota. I have recently read your 2007 Journal of Neuroscience paper "Nonparametric statistical testing of EEG- and MEG- data". This paper is wonderful and helps me a lot!  I am kind of new to this field and recently working on my own MEG project and try to do the permutation test using Fieldtrip. 
>     There is one question about the permutation test results on (sensor, freq, time)-samples. In the paper, you did the permutation test across different frequencies and found the clusters were in the 15-30Hz. And Fig4 showed the topography of significant cluster averaged over 15-30Hz. 
>      But when I tried similar procedure and got stats results, functions like ft_clusterplot only allows singleton dimension of frequency. Should I average the stats result across frequency band? (The significant cluster were shown as 0/1 in a 3D matrix).
>     I would appreciate it a lot if you could give me some guidance! Many thanks.
> Best,
> Yijun Ge
> Department of Psychology
> University of Minnesota
> 6514408588

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20200130/bfd17b22/attachment.htm>

More information about the fieldtrip mailing list