[FieldTrip] clusterstatistics

[Gio Piantoni]

Hi Mark,

It seems to me that clusterstatistics is giving the appropriate
answer. Indeed it takes advantage of the information which can be
sparse over multiple channels/frequency/time bins. The only criterion
is that this activity is contiguous. I don't follow your example about
significant activity at X and X+100 on two different channels. There
should be some continuity between the two. As you mentioned, the
cluster can span one channels between 0 and 100 and in another
neighboring channel between 50 and 150. The cluster will correctly
include both channels and the time window between 0 and 150. This is
because there is some overlap in time between the two channels.

If you think in terms of activity, you can report that activity is
significantly different from baseline between 258 and 370 ms, over
your channels. The p-value only belongs to the cluster, not to the
single time point or channel. If you use bonferroni correction, you
can attach single p-values to each timepoint/channel.

If you have a priori hypothesis that you expect activity in Fz, you
can specify a single channel in timelockstatistics. Depending your
data, you might see more power (because you're testing fewer channels
at the same time).

Hope this helps,

Gio

-- 
Giovanni Piantoni, Ph.D. student
Dept. Sleep & Cognition
Netherlands Institute for Neuroscience
Meibergdreef 47
1105 BA Amsterdam (NL)

+31 (0)20 5665492
g.piantoni at nin.knaw.nl
www.giovannipiantoni.com


On Mon, Feb 20, 2012 at 12:25, Mark Noordenbos
<mark.noordenbos at gmail.com> wrote:
> Dear Fieldtrippers,
>
>
> I have a question about how to report the findings using clusterstatistics.
> I used the cluster method to determine at which channels activity is
> significant different from zero.
>
> The cluster method revealed a significant cluster ranging from about 258-370
> ms (using ft_clusterplot). If I only look at a certain channel (e.g., Fz)
> the activity is only significant during 272-322 ms.
>
>
> The time interval of the cluster is determined by the first and the last
> sample where the activity is significant, regardless if a certain channel
> shows significant activity during the complete time interval. For instance,
> if a certain channel shows significant activity only at sample X and another
> channel only at sample X+100, the cluster ranges from sample X to X+100.
>
>
> My problem is how to report the finding of the significant cluster. If I
> report that a significant cluster was found from 258-370 ms and show a plot
> of the activity at Fz (272-322 ms). Then the time intervals and p-values do
> not correspond. The stat structure contains only the p-value for the time
> interval of the cluster and not of the individual channels.
>
>
> Any suggestions or examples of how to report clusterstatistics.
>
>
> Regards,
> Mark
>
> --
> Mark Noordenbos, MSc
>
> Radboud University Nijmegen
> Behavioural Science Institute
>
> P.O. Box 9104, Room A05.36
> 6500 HE Nijmegen
> The Netherlands
>
> Email: m.noordenbos at bsi.ru.nl
> Telephone: +31 24 3612070
> Fax:          +31 24 3616211
>
> http://www.ru.nl
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip