output of timelockstatistics

[Robert Oostenveld]

Hi Marcel,

On 12 Mar 2008, at 14:50, Marcel Bastiaansen wrote:
> I used timelockstatistics (method montecarlo, i.e. cluster
> randomization) to compare the significance between two erp's.
> Here's the output struct.
>
> stats112_113 =
>
>                    prob: [30x600 double]

your input consists of 30 channels and 600 timepoints. For each of
the 30x600 samples a probability is computed. In your case
(montecarlo and correctm=cluster) the probabilities of neighbouring
chan_time points will often be identical, as they belong to the same
cluster.

>             posclusters: [1x5 struct]

There are 5 clusters with a positive summed t-value

>     posclusterslabelmat: [30x600 double]

this contains the cluster number that each sample belongs to, or a 0
if it does not belong to a sampple. Try imagesc
(stat.posclusterslabelmat)

>         posdistribution: [1x1000 double]

This is the randomization distribution for the statistic of interest
for the positive tail, i.e. the sum of the t-values in the largest
cluster in each randomization. Do hist(stat.posdistribution) and look
at the content of stat.poscluster(1) to see where the first cluster
lies in that distribution.

>
>             negclusters: [1x20 struct]
>     negclusterslabelmat: [30x600 double]
>         negdistribution: [1x1000 double]

idem for negative clusters, except that there are 20 in total (note
that not all might be significant).

>                    mask: [30x600 logical]

for each sample contains a 0 if not significant, or a 1 if significant.

>                    stat: [30x600 double]

for each sample contains the t-value (assuming that you used that as
statistic). So you can do imagesc(stat.stat), and then alpha
((stat.mask+1)/2) to add some transparency. The summed cluster
statsitic for the first positive cluster can be retrieved using sum
(stat.stat(find(stat.posclusterslabelmat(:)==1)). This should match
the content of the stat.posclusters(1) structure.

>                  dimord: 'chan_time'
>                   label: {30x1 cell}
>                    time: [1x600 double]

The other fields have the first dimension "channel" and second
dimension "time". For freqstatistics on TFR data you would get
dimord='chan_freq_time'

>                     cfg: [1x1 struct]

this was the configuration used in the computation, including all
defaults that were set.

See also http://www2.ru.nl/fcdonders/fieldtrip/doku.php?
id=fieldtrip:documentation:tutorial:plotting and the clusterplot
function.

best regards,
Robert

----------------------------------
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/fcdonders/fieldtrip.