 |
 |
Hi Michael, Randomise needs fewer assumptions and can be used in higher level designs with complete data as these without difficulties. The most important bit, however, refers to violation of compound symmetry for between-subjects factors, which can lead to incorrect (possibly inflated) error rates. This has been discussed in the list a few times and is mentioned briefly in the FSL-GLM manual. It's safe to use randomise, but one would have to run it twice: one for the contrasts that are within-subject, another for those between-subject. For mixed factors (within and between-subjects) the current version cannot yet deal with it, but a solution will become available soon I hope. All the best, Anderson On 30 October 2014 20:10, Michael Spilka <[log in to unmask]> wrote: > Thank you Veronica for asking this useful question, and to Anderson for > providing an helpful reply. If I could ask a follow-up question: Anderson, > is there something specific about this type of design that leads you to > recommend using Randomise instead of running the mixed-effects analysis in > FEAT? > > Thank you, > > Michael >