clusterrandomization

Olga Sysoeva olga at GRAPHICMIND.INFO
Thu May 11 18:29:44 CEST 2006


Thank you very much for the help.

I continue my succes with Clasterrandomization analysis on within-subject
disign.
I applied clasterrandomization to compare two condition. I mostly use the
settings
from the tutorial (both cfg.alphathres =0.05 and cfg.alpha=0.05) and got
positive and negative clusters
 (both clusrand.posclusters and clusrand.negclusters not empty). But when I
look for the pval for neg cluster
(which is only one, size 8) it appeared to be 0.51. It is bigger than our
threshold (0.05). Why it is mark as
significant cluster than (clusrand.negclusterlabelmat =1 for 8 elements)?
Could somebody comment on it?

 Best Regards,
Olga.

----- Original Message -----
From: "Robert Oostenveld" <r.oostenveld at FCDONDERS.RU.NL>
To: <FIELDTRIP at NIC.SURFNET.NL>
Sent: Monday, May 08, 2006 11:16 AM
Subject: Re: [FIELDTRIP] first success!!!


>> I want to share my first success in using Filedtrip. Although I  could
>> still not properly load my current 64 electrodes data, I  succeeded in
>> importing my old 19 channel data. Moreover, I succeded  in conducting the
>> between-trial clasterrandomization analysis!
>
> It might be that the automatic projection of the 3D electrodes to the  2D
> surface for plotting is not optimal for your 64 channel set,  perhaps due
> to a few very low electrodes. I suggest that you  construct a layout file,
> see my previous mail. You can also look into  the code in
> private/createlayout, where at the bottom you will find  some code that
> helps you to  visualise the layout.
>
>> But I still have problems with within-subject cluster- randomization. The
>> error "did not find the gradiometer or electrode  information" appears,
>> line 461. What could it be?
>
> Cluster randomization requires that the neighbourhood geomerty in the
> data is known, i.e. the neighbouring electrodes. You can specify them
> with cfg.neighbours. It is documented in "help clusterrandanalysis".  One
> method to automatically specify the neighbourhood is by having  electrode
> information in the dataset, i.e. an elec structure and to  specify the
> cfg.neighbourdist option (default is 4 cm, which is  suitable for the CTF
> gradiometers). I do not know the units of your  electrodes (could be m,
> cm, mm, or unit-scaled dimensionless), but  the cfg.neighbourdist should
> be specified in the same units.
>
> best regards,
> Robert



More information about the fieldtrip mailing list