Hi Petter,

surely genetic imager experts have an answer (Tom?) but I'll try to answer

> We have done an fMRI analysis in spm using a multiple regression model.
> We have used genotype (or strictly speaking, presence/absense of one
> allele) as a (binary) regressor and total brain volume (from FreeSurfer)
> as a second regressor. When we include this brain volume regressor and
> set the contrast vector to 1 0 (1 = genotype) we find large activation
> clusters. In contrast to this, we find no activation clusters at all if
> we use a simple regression model with only genotype and no volume
> regressor. We also find no activation if we use a two sample t-test approach with
> the genotypes as two groups.

well coding for the presence/absence I would have used 1 and -1 because the contrast [1 0] look for voxels showing a mean activation for group having the allele which is higher than the overall mean (assuming you modelled it - see bellow) - whereas I guess you want to look at the difference in BOLD between the two groups

beside this, the multiple regression effect can be explained depending on i) if you modelled the overall mean (in the simple reg or paired this is model by default) and ii) the correlation between regressors , though here I can't see how (maybe because of the 10 coding ?)

try the new coding + enter normalized volume values (x-mean/std) ..

hope this will solve your problem


The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.