You would typically include the variable as a confounding covariate
within your GLM. This would show where there appear to be differences
that can not be explained (linearly) by the TIV. Alternatively, you
may want to re-phrase your question in terms of analysing the
proportion of the TIV that each region accounts for (ie use a
proportional scaling correction).
When you write up the work, it is probably worth stating that the TIV
is significantly correlated with your effect of interest.
Best regards,
-John
On 24 April 2012 16:39, CARLES M. FALCON FALCON <[log in to unmask]> wrote:
> Dear spmers,
> in a VBM analysis I am performing, the total intracranial volume (TIV) has notable collinearity with group defining variable (different mean value of the TIV in each group). Is it valid to orthogonalize both regressors (group and TIV) to remove this collinearity?
> Many thanks
>
> Carles Falcon
> Medical Image Lab. IDIBAPS
> GIB-UB. CIBER-BBN. Barcelona
|