[FieldTrip] Cluster analysis on correlation statistics

Arjen Stolk a.stolk8 at gmail.com
Tue Dec 9 12:40:58 CET 2014


Hi Chris,

Quick reply:

-          The new ft_statfun_intersubcorr can calculate Spearman
correlation statistics between subjects (but not between trials), and
doesn’t currently have a parametric (e.g., Pearson’s) option, although this
appears to be covered by the ft_statfun_indepsamplesregrT function anyway.

> I implemented a cfg.type, allowing the specification of the correlation
statistics option (e.g. Pearson), but haven't yet tested it.

By the way, in the ft_statfun_intersubcorr description/instructions, some
of the text appears to refer to ft_statfun_depsamplesT instead. Can I
assume that ivar is in fact the behavioural variable of interest rather
than “conditions that must be compared” as currently stated? Also, what is
the reason this function cannot be used across trials within a subject?

> ft_statfun_intersubcorr was implemented last week on request. the
documentation was taken from ft_statfun_depsamplesT and needs to be updated
(thanks for reminding). The function computes a correlation across
observations (subjects/trials) where one input is neural data, and the
other behavior (enforce a match of dimension and size), and then transforms
it to a t-value. Theoretically therefore the function could also be used to
calculate a correlation statistics across trials in a subject.


>> Btw, I didn't see Eric's reply before. But if you get it  to work with
ft_statfun_indepsamplesregr, I'd go for that.

best regards,
arjen



2014-12-09 12:27 GMT+01:00 Dr Christopher Brown <cb802 at cam.ac.uk>:

> Dear Eric, Tom, Arjen
>
>
>
> Thanks for the information and it’s good to know the functionality I need
> is already there. Just to clarify, I understand the situation as thus
> (please correct any errors of understanding):
>
> -          The ft_statfun_indepsamplesregrT function can be used to
> calculate parametric linear regression statistics with ivar as the
> independent variable of interest, e.g. a behavioural variable across trials
> or subjects, and uvar as units (e.g. trials or subjects).
>
> -          The new ft_statfun_intersubcorr can calculate Spearman
> correlation statistics between subjects (but not between trials), and
> doesn’t currently have a parametric (e.g., Pearson’s) option, although this
> appears to be covered by the ft_statfun_indepsamplesregrT function
> anyway. By the way, in the ft_statfun_intersubcorr
> description/instructions, some of the text appears to refer to
> ft_statfun_depsamplesT instead. Can I assume that ivar is in fact the
> behavioural variable of interest rather than “conditions that must be
> compared” as currently stated? Also, what is the reason this function
> cannot be used across trials within a subject?
>
>
>
> Best wishes
>
> Chris
>
>
>
> *Dr Christopher Brown*
>
> *Research Associate | CamPain Research Group | Department of Anaesthesia |
> University of Cambridge*
>
> Box 93
> Addenbrooke's Hospital
> Hills Road
> Cambridge CB2 0QQ
>
> 01223 256 995
>
>
>
> *From:* fieldtrip-bounces at science.ru.nl [mailto:
> fieldtrip-bounces at science.ru.nl] *On Behalf Of *Tom Marshall
> *Sent:* 08 December 2014 15:10
> *To:* FieldTrip discussion list
> *Subject:* Re: [FieldTrip] Cluster analysis on correlation statistics
>
>
>
> Hm... it seems that I may have reinvented the wheel (or perhaps invented a
> cruder wheel before Eric and co provided this more streamlined one).
>
> Eric, can indepsamplesregrT also compute nonparametric correlation
> coefficients (eg Spearman)?
>
> Best,
> Tom
>
> On 12/8/2014 3:56 PM, Maris, E.G.G. (Eric) wrote:
>
> Hi Chris,
>
>
>
> Just to clarify my last email: I'm referring to correlation between either
> single trial data within subject, or averaged ERPs across subjects, with a
> behavioral variable. Not linear regression across conditions as is
> already implemented in the ft_statfun_depsamplesregrT function.
>
>
>
> For correlating/regressing single trial data of a single subject, you use
> the ft_statfun_indepsamplesregrT (with the units-of-observations being the
> trials, which each correspond to one column of your design matrix).
>
>
>
> For correlating/regressing subject-averaged data of a group of subjects,
> you also use the ft_statfun_indepsamplesregrT (with the
> units-of-observations being the subjects, which each correspond to one
> column of your design matrix).  Note, in this case your behavioral data
> (accuracy or RT) or also subject averages.
>
>
>
> Best,
>
>
>
> Eric Maris
>
>
>
>
>
>
> ------------------------------
>
> *Van:* Dr C A Brown [cb802 at cam.ac.uk]
> *Verzonden:* zaterdag 6 december 2014 12:33
> *Aan:* fieldtrip at science.ru.nl
> *Onderwerp:* Re: [FieldTrip] Cluster analysis on correlation statistics
>
> Hello again,
>
>
>
> Just to clarify my last email: I'm referring to correlation between either
> single trial data within subject, or averaged ERPs across subjects, with a
> behavioral variable. Not linear regression across conditions as is already
> implemented in the ft_statfun_depsamplesregrT function.
>
>
>
> Many thanks
>
> Chris
>
>
>
> Dr Christopher Brown
>
> Research Associate
>
>
>
> Sent from a mobile device, please excuse my brevity.
>
>
>
> ----- Reply message -----
> From: "Christopher Brown" <cb802 at cam.ac.uk> <cb802 at cam.ac.uk>
> To: <fieldtrip at science.ru.nl> <fieldtrip at science.ru.nl>
> Subject: Cluster analysis on correlation statistics
> Date: Sat, Dec 6, 2014 07:49
>
>
>
> Dear all
>
> I would like to use the cluster analysis functions in Fieldtrip on
> correlation statistics. I note there has been some discussion about this on
> this list in the past but I haven't come across the necessary
> ft_statfun_corr function to allow this analysis. It would be great if
> somebody has this code and could share it with me; otherwise, I will embark
> on writing it myself and sharing my results with everyone for feedback.
> Being new to Fieldtrip I would greatly appreciate any collaborative help
> with this. May I check my understanding on something before getting
> started: To allow for multiple comparisons correction using monte carlo for
> example, as part of a cluster analysis, I presume the output of the
> function must be t values rather than correlation coefficients? Any other
> hints as to how to write this are most welcome.
>
> Kind regards
>
> Chris
>
> --
>
> *Dr Christopher Brown*
>
> *Research Associate | CamPain Group | Division of Anaesthesia | School of Clinical Medicine | University of Cambridge*
>
>
>
> Box 93
>
> Addenbrooke's Hospital
>
> Hills Road
>
> Cambridge CB2 0QQ
>
>
>
> 01223 256 995
>
>
>
>
> _______________________________________________
>
> fieldtrip mailing list
>
> fieldtrip at donders.ru.nl
>
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20141209/78e6e8c4/attachment.html>


More information about the fieldtrip mailing list