Hi there

the easiest option is to code your design matrix as stim1 attended stim1 non attended stim2 att and stim2 non attended then specify it's a factorial design 2x2

deside this, note that correlation is not an issue in the sense that normal equaltions inversion solved in SPM using pinv ensure beta parameters are mnimal - ie each regressor explains as much as possible it's unique part of variance and common variance (correlation) goes into the error - of course now you see that correlation is a problem but not in terms of maths, more in terms of design since lot of variance is unexplained. 

more on

http://en.wikipedia.org/wiki/Linear_least_squares#Derivation_of_the_ normal_equations

http://mathworld.wolfram .com/NormalEquation.html

http://en.wikipedia.org/wi ki/Orthogonalization

Good luck

Cyril

 
> Hi~
>    I'm doing DCM with SPM. Before DCM I compute the GLM to get VOIs. 
> my experiement is a 2*2 factorial design, with a modulate factor 
> (with or without attention) and a input factor (2 types of 
> stimulus). I thus included 4 regressors: one corresponds to all 
> attended trials (with both types of stimuli), the second corresponds 
> to all inattended trials (with both types of stimuli), the third 
> corrsponds to one type of stimuli (with or without attention) and 
> the last corresponds to the other type of stimuli (with or without 
> attention). However, the regressors so defined highly correlated 
> with each other. Mathametically, this may cause some problems in the 
> model computation. Whether SPM automatically deal with such issue? 
> Or is there any other method to include both the modulate and input 
> factor while at the same time does not invoke the correlatin problem?
>  
> Thank you very much!
>  
> Abraham
>