[pymvpa] Group level RDMs

Nick Oosterhof n.n.oosterhof at googlemail.com
Wed Feb 24 12:08:05 UTC 2016

> On 24 Feb 2016, at 12:49, Martin Sjøgård <martinsjogard at gmail.com> wrote:
> When doing RSA, is there a preferred way to construct representational distance matrices at the group level (RDMs averaged across subjects) in pyMVPA? Reading the example data, it seems that it might skip directly to the group RDM?

Indeed taking the average is quite common and would seem a reasonable thing to do. 

If you do ROI analysis and if there is sufficient agreement across participants, you may consider using STATIS/DISTATIS [1]. There was a PR long time ago [2] but it seems it  never got in a ready state to be merged in the official master branch. If you're comfortable using Matlab / GNU Octave, DISTATIS functionality is present [3] in CoSMoMVPA (disclaimer: I'm its main developer).

[1] Abdi, H. & Valentin, D. in Encyclopedia of Measurement and Statistics (ed. Salkind, N.) 42–42 (SAGE Publications, 2007).
[2] https://github.com/PyMVPA/PyMVPA/pull/121
[3] http://cosmomvpa.org/matlab/demo_fmri_distatis.html

More information about the Pkg-ExpPsy-PyMVPA mailing list