Dear Ivan,

Yes, exactly. Within-subject contrasts (i.e. between-conditions and/or between-sources) are very easy to understand. Whatever linear combination you specify in the corresponding contrast vector is going to be applied to the condition- and source-specific connectivity values (typically fisher-transformd correlation coefficients) and the resulting numbers are the ones being entered into your second-level model. So in your example a [.5 .5 -1] contrast means that for each subject the (r_task_level1+r_task_level2)/2-r_baseline effect is going to be computed and the results are going to be entered into your second-level general linear model for between-subject analyses.

If you are interested, the rational for this is that a likelihood ratio test for the hypothesis:
  C*B*D' = 0  (a test including a between-subjects contrast C and a within- subjects contrast D)

in a linear model of the form:
  Y = X*B + error  (GLM using your original data matrix Y)

is exactly equivalent to a test for the alternative hypothesis:
  C*B2 = 0 (a test including only between-subjects contrast C)

in an alternative linear model of the form:
  Y*D' = X*B2 + error  (GLM using a linear combination of your original data matrix Y)

(where Y is your subjects by within-subject effects functional connectivity estimates, X is your subjects by between-subject effects design matrix, C is your between-subjects contrast vector/matrix, D is your within-subjects contrast vector/matrix, and B or B2 are the vectors/matrices of regression coefficients estimated by the corresponding general linear models)

Hope this helps
Alfonso

Originally posted by Yang Yang: