On Tue, 8 Mar 2005 18:06:54 EST, John D. West <[log in to unmask]>
wrote:
>Dear SPM group:
>
>We are currently trying to setup an ANCOVA with multiple covariates for
VBM. Unfortunately, the basic ANCOVA in SPM2 appears to only permit one
covariate (as a nuisance variable).
Not sure this is what you're looking for but you can modify the SPM code
to allow multiple covariates.
Here's an example: in spm_spm_ui.m, replace
D = [D, struct(...
'DesName','AnCova',...
'n', [Inf Inf 1 1], 'sF',{{'repl','group','',''}},...
'Hform', 'I(:,2),''-'',''group''',...
'Bform', 'I(:,3),''-'',''\mu''',...
'nC',[0,1],'iCC',{{8,1}},'iCFI',{{1,1}},...
'iGXcalc',[-1,2,3],'iGMsca',[1,-9],'GM',[],...
'iGloNorm',9,'iGC',12,...
'M_',struct('T',[-Inf,0,0.8*sqrt(-1)],'I',Inf,'X',Inf),...
'b',struct('aTime',0))];
with
D = [D, struct(...
'DesName','AnCova',...
'n', [Inf Inf 1 1], 'sF',{{'repl','group','',''}},...
'Hform', 'I(:,2),''-'',''group''',...
'Bform', 'I(:,3),''-'',''\mu''',...
'nC',[0,Inf],'iCC',{{8,1}},'iCFI',{{1,1}},...
'iGXcalc',[-1,2,3],'iGMsca',[1,-9],'GM',[],...
'iGloNorm',9,'iGC',12,...
'M_',struct('T',[-Inf,0,0.8*sqrt(-1)],'I',Inf,'X',Inf),...
'b',struct('aTime',0))];
(Of course, don't simply cut and paste this as it's not certain the
version to modify is the same. But the idea behind the modification
should be relevant.)
>
>We could use any suggestions you might have to setup this analysis.
>
>The basics of the VBM analysis we wish to perform are as follows:
>There are two between subjects factors that we wish to investigate in a 2
X 3 design yielding 6 independent cells (Factor A with level 1 and 2 and
Factor B with level 1, 2, and 3).
>We think that the best design to use would be to combine these factors so
that we have 6 groups. (11 12 13 21 22 23). We also want to covary for 3
other potential confounding variables (age, sex, ICV). It seems that we
could do this by using a multiregression analysis and a combination of
dummy variables to code for the six groups. However, it seems to us that
we would lose the ability to look at paired contrasts betwen subgroups
(e.g. 11 vs 12) and to extract the adjusted data for profiling the
results. Similar issues have come up with complex fMRI models.
>
>Any suggestions on how to compute a complex factorial design with
multiple covariates would be most appreciated.
>
>Thank you in advance.
>John West
>Brain Imaging Lab, DHMC
|