Dear Philippe,
> I ran a PET study where subects had to perform a same task under two
> conditions (A & B), and I want to know regions where the performance
> measure (covariate P) is differentially affected by condition. In
> SPM96, the usual way to test this question was to build a covariate of
> interest which was the interaction between performance (P) and
> conditions (A and B specified as a vector of 1 -1 ). The two members of
> the interaction were also entered as counfound to disclose only those
> regions were the performance was significantly more correlated with
> rCBF in condition A than in condition B.
>
> This strategy will certainly work also in SPM99b, but I was wondering
> if the same result is obtained using factor by covariate interactions
> in the PET statistic "Multi-subj: conditions & covariates". In this
> case, one has first to specify which scans below to each condition A
> and B. Afterwards the option is proposed for the interaction between
> the covariate vector (the performance measure P) and conditions. In the
> resulting design matrix, there ar four columns "of interest" : two for
> conditions A and B, one for the interaction A X P, and one for the
> interaction B X P. Hence I have two related questions :
Nearly. The last two culumns model the simple main effect of P under
levels A and B of the condition factor.
> 1. If I compute a contrast (1 -1) between AXP and BXP, am I right to
> expect to disclose the same regions as in the SPM96 procedure describe
> above ?
This contrast is indeed the interaction (i.e. difference in simple main
effects) and should be identical to the SPM96 analysis.
> 2. Are the coufounds controlled in the same way ? Or in other words,
> did this procedure control for the influence of the two members of the
> interaction, leaving only the product of their interaction ?
Yes it does. All the main effects are modelled and are orthogonal to
the contrast specified with weights -1 1.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|