Print

---

Hi all,
I'm seeking some advice for my first GLM analysis in SPM. I would like to know how to model interaction effects on the second level in SPM. 

For my analyses of my behavioural data, l find that an interaction between “(1) Behaviour (y or n)” and “(2) Memory of that Behaviour (y or n)” explains my dependent variable, i.e. “Behavioural Change”. In other words, there is a significant interaction effect “Behaviour x Memory” explaining “Behavioural Change”.

I now would like to find the neuronal correlates of this “behaviour x memory-dependent behavioural change” effect.

Currently, my first level model in SPM contains the following regressors:
		(1) Stimulus Onset
		(2) Behaviour Onset 
		(3) Behaviour Change (Parametric Regressor for (2), Continuous)
		(4) Memory of Behaviour (Parametric Regressor for (2), Binary Variable)

In my second level analyses I’m now interested in how the “behaviour x memory-dependend behavioural change" effect is associated with changes in BOLD on the group-level.  How do I model this interaction effect?

Do I :
(a) Set the contrasts as [0 0 1 1] ?

Or 

(b) Create a new interaction term a-priori, which is the product of “Behaviour Change” and “Memory of Behaviour” and put this already into my first level model?

or

(c) Do I rather change my first level model such that it closely parallels my behavioural model? Specifically, I could classify the trails into Behaviour (A or B) and Memory (A or B) and put all possible combinations of these trials as onset regressors into the GLM. Then for each trial type I also model the parametric regressor “Behavioural Change” and use that contrast image for the second level analysis.

Is any of the above mentioned analyses a valid approach? 

Thank you very much in advanced for your help! 

Kati