Hi - you're using randomise, which (unlike the stats in FEAT) doesn't use the variance groups information - so don't bother with the "grouping" setup.

However - yes, there is no problem with different group sizes in randomise - this will be fine and valid. You should use all the subjects you have.

Cheers.


On 23 Mar 2012, at 01:10, E. Coffey wrote:

Dear FSL experts,

Does unbalanced group size affect the variance calculation and therefore whether the groups should be entered separately in the design matrix?

I am running a simple "Unpaired Two-Group Difference (Two-Sample Unpaired T-Test)" using "randomize" on DTI-FA data, in which I am not expecting the brains of each group to be wildly different as you might for certain patient groups. I have therefore elected to model them in a single group for the purposes of variance. However, I have unequal group sizes: N= 16 and 23. My questions are:

1) Should this imbalance be taken into consideration when deciding how to model the groups?

2) Would be better to leave some subjects out to have N=16 in both groups?

3) Is there some way of calculating or quantitatively supporting my equal variances assumption?

Best regards,

Emily



---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------