Hello Miguel,
The core assumption in randomise is of the exchangeability of the input data - there aren't really any other assumptions made about the un-permuted statistic image ( or its distribution ).
Many Regards
Matthew
>
>
> Hi FSLers,
>
> I understand that, in randomise, an unpermuted T map is first computed and then those T values are compared to our 'custom distribution'. I this is true, my question is to what extent the unpermuted T is affected by the assumptions that randomise is trying to avoid. If we get that original T from a multiple regression and our data is not normally distributed, shouldn't that first step have an impact on the following steps (permutations)?
>
> Thanks
> Miguel
