Dear nilearn devs/users,

I am very much interested in using the `non_parametric_inference`

function in a second-level analysis for a paired t-test.

In that case, care must be taken when the permutations are randomly drawn as they must respect some block structure of my data/design (i.e. it should respect the paired structure to avoid mixing data of different subjects with each other).

The type of analysis I have in mind is presented e.g. in the Example 4 in the Winkler et al paper. In this case the blocks as specified explicitly with the “EB” column.

As far as I understand, there is no way of telling `nilearn.glm.second_level.non_parametric_inference`

how the blocks are organized. Does it assume that all rows pertain to the same block? Or is there something I am missing here? I would have naively expected a new keyword argument, like `exchangeability_blocks=EB`

where `EB`

is an array of length = number of rows of the design matrix.

Thanks a lot for developing and maintaining nilearn, it is really awesome!

Regards

AR