Hi - this would depend on the exact details of what you've done, but 0.1% doesn't sound very high
to me. However most likely the simple explanation is that you have used the default group-stats
option of FLAME, which is more accurate than OLS (ordinary least squares) for the higher-level
analysis, but the averaging that you have done yourself is equivalent to an OLS analysis, hence the
difference?
Cheers, Steve.
On Thu, 18 Jan 2007 11:18:02 +0000, Lori Minini <[log in to unmask]> wrote:
>Hello
>
>I have run Featquery (with analyze mask images) on a series of FEAT
>higher-level folders generated in FSL 3.3, but that used first-level
>analyses from FSL 3.2. I found differences (as big as 0.1%) between
>the COPE % values of the higher-level analyses and the average of the
>first-level results.
>
>I checked whether using first-level analyses from FSL 3.2 could have
>generated this difference by comparing the Featquery results for
>first-level analyses carried out in both FSL 3.2 and FSL 3.3 for the
>same data. This resulted in some minor differences, however it seems
>that they are unlikely to account for the discrepancy reported above
>as this was also present for a subset of the data for which both
>first- and higher-level analyses were run in FSL 3.3.
>
>Has anyone encountered a similar difference and do you know where it
>stems from? In general, should we expect to find differences in the
>Featquery results for higher-level analyses depending on the FSL
>version used for the first-level?
>
>Many thanks
>
>Lori
>
>
>_______________________________________________________
>
>Dr Lori Minini, PhD
>
>Department of Physiology, Anatomy and Genetics & FMRIB Centre
>University of Oxford
>Sherrington Building
>Parks Road
>Oxford OX1 3PT
>
>tel +44-1865-272555
>fax +44-1865-272543
>email [log in to unmask]
>
>_______________________________________________________
>===========================================================
==============
|