 |
 |
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;3b1281b8.05 https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;ca40e28e.1206 I am looking at GLM at the first level, and encountering an error “Warning: at least one EV is (close to) a linear combination of the others.” I found two posts that talk about this issue in my case (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;3b1281b8.05 and https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;ca40e28e.1206), and both indicated that the contrasts will still be fine. However, this does not seem to be the case because as the error states, “Contrasts involving these combinations will be set to zero.” Rather than explain my whole model let me offer a simplified model where we show a subject pictures of people. We have males and females and we have old and young people. We believe that both age and sex are important determinates of where activation occurs. We can set up a model that has repressors male, female, old, and young. This is great, except that male + female = old + young, so we cannot invert the covariance matrix. FSL is sitting my contrasts to zero, they are flat lines in the time series plots, even thought they indicate max z-stats in the 70’s which is insanely high—like we have probably not observed a random occurrence like that since the beginning of the universe. IF I were just setting up a GLM for this with paper and pencil, I would set up: Y = b0 + b1*Male + b2*Old + b3*Male*Old However, I know FSL demeans everything so that there is no b0 term. I thought about just using Y = b1*Male + b2*Old + b3*Male*Old, and letting the demeaning take care of making b0 = 0. However, as Jesper points out, “…but you canNOT look for A (over and above some unobserved baseline) through [1 0]”, which in this case would be [1 0 0]. The thing is Jesper put the “NOT” in all caps so this is BAD. So, the question is, how does one set up the first level model where the important contrasts are linear combinations. Specifically, they are of the form A and NotA and B and NotB? I realize that these are valid contrasts, but FSL will not actually run them. It tells me it is going to set them to zero and it does so.