Hi Peter,
That sounds right. The baseline number is only really meaningful at the
first level and as you say, it should be around 10000 when using Feat. At
higher levels, the "time series" is really just a concatenation of PE
scores from individual runs (or subjects) and thus there is no meaningful
baseline. What I do is to take the % signal change from the 1st level
analyses and manually combine (in excel or SPSS) as in the higher level
analyses. So for instance, if I have 3 runs per subject and 10 subjects, I
extract 30 mean % signal changes from the 1st level and average the 3 per
subject into one mean % signal change per subject which I then use in
subsequent analyses or bar plots. This way the degrees of freedom are
correct as are the variance measures.
Hope that is some help.
Joe
|