Dear SPM-Experts
 I'm currently struggling with the analysis of an parametric modulated design on the second level 
with SPM2.   

I have a simple ABRABRABRA... design, where A is the rest condition, B is the stimulus condition, 
which is modulated by the perceived intensity of the stimulus, which by itself is recorded in 
condition R.

I modeled the perceived intensity using a 2nd order polynom. In the end I have a design matrix 
with the following columns: B*bf, B^1*bf, B^2*bf, R*bf. The rest condition A is implicitly modeled.

From here on I'm not sure how to proceed on the second level. My guess is that analogously to the 
analysis with several basis functions it should be possible to create t-contrast for each column of 
the design matrix (for each subject) and take them to the second level using an ANOVA (one way). 
There I would create an F-contrast for the effects of interest [1 0 0; 0 1 0; 0 0 1] and possibly t-
contrasts like [1 1 0] to look for voxels that are responding to the the stimulus linearly. Is this 
correct to use an ANOVA in this context and are t-contrasts valid then? 
How can I show the voxels that are responding to the stimulus where this response is in a 
(positive) linear way. Can I create a conjunction contrast of two t-contrasts: [1 0 0] and [0 1 0]?  

Another (possibly stupid)  question I have is, how I can reconstruct the function of voxel response 
vs. rating like I can do for a single-subject analysis using the plot button. Normally I would think 
that I can simply extract the beta for the linear modulation and the beta for the quadratic 
modulation and feed them into the formula b2*x+b3*x, where x is the rating I used when setting 
up the model. But the scaling of the resulting graph is rather odd then and I think this should be 
due to orthgonalization and scaling. Is this valid and if so how can I get back the original scaling?


Thank you in advance!

Rainer