Hi,
On 26 May 2009, at 14:33, ZhuangJohnny wrote:
> Dear Experts
>
> I am currently learning to use TBSS to compare DTI-derived FA values
> between control and patient groups. I set up my design matrix and
> design contrast as follows.
>
> design matrix:
>
> Gp EV1(Control) EV2(Patient) EV3(demeaned_age)
> EV4(demeaned_education) EV5(sex:female) EV6(sex:male)
> 1 1 0
> 1.7548 0.1133
> 1 0
> 2 0 1
> -1.0252 3.1133
> 1 0
> 1 1 0
> -2.0825 12.1133
> 0 1
> . . . . .
> . . . . .
> 1 1 0
> 13.1848 -5.8867
> 0 1
> 1 1 0
> 2.0948 3.1133
> 1 0
> 2 0 1
> -4.1552 -3.8867
> 1 0
You should not set different (variance) group memberships - randomise
doesn't use those in the same way anyway and you don't need that here.
>
> contrast 1 -1 0 0 (Normal>Patient in FA)
> -1 1 0 0 (Normal<Patient in FA)
>
> Q1: Do i need to split continuous variables, such age, into two age
> EVs for two subgroups, that is, one demeaned age EV for controls,
> and the other one for patients.
> Currently, i didn't split the continuous variables, based on
> examples from previous archives. However, if i didn't do the
> splitting, there will be a error information from FSL: "Problem with
> processing the model: Warning-design matrix uses different groups,
> but these do not contain separable EVs for the different groups." Is
> there anything wrong with my matrix?
Right - but you should not set the group memberships anyway for
randomise (it means something else - see the randomise manual) - then
the error message will go away.
Whether you have separate (e.g.) age EVs for the two groups is up to
you - each approach is valid and does a different thing. If the two
groups have the same mean age then it will make almost no difference
to split it - it they have different mean ages then you need think
about whether you want to be trying to account for this by using a
single EV for this - which may help reduce the bias from the non-
matching - but that can't be guaranteed as the effect of age is
probably not simply additive and separable from the effect of disease.
>
> Q2: If i import more than two categorical variables, in that case,
> EV1, EV2, EV4, and EV5, FSL will indicate that " design matrix is
> rank deficient, one EV is a linear combination of others. How can i
> address this problem by rearranging my categorical EVs.
This is because EVs 4 and 5 should be combined into a single EV -
which also should be demeaned.
Cheers.
>
> Thanks in advance for any help and suggestions
>
>
> Cheers
>
> Johnny Zhuang
>
> 立刻下载 MSN 保护盾,保障Messenger 安全稳定! 现在就下载!
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
|