Stats gurus within the SPM community:
I have perused the listserv regarding the appropriate way to set up a
contrast matrix to test a group X continuous predictor interactions at the
second level. As an example, let's assume we have a study with two groups
(GP1 and GP2) and some continuous predictor variable (PRED). Let's say we
have 5 subjects per group (to make it easier for me to create the contrast
matrix). It seems that in order to test the group X continuous predictor
interaction, the consensus view is that we should set up the contrast matrix as:
GP1 GP2 GP1_PRED GP2_PRED
1 0 24 0
1 0 21 0
1 0 26 0
1 0 19 0
1 0 17 0
0 1 0 32
0 1 0 23
0 1 0 28
0 1 0 23
0 1 0 26
And the contrast would be: '0 0 1 -1' to test GP1_PRED > GP2_PRED AND '0 0
-1 1' to test GP2_PRED > GP1_PRED (to test the group X PRED interaction)
Why is another column for PRED AND CONSTANT not included in the contrast
matrix? Isn't the standard way to test an interaction to enter both
variables into the model with a constant, i.e.,
Y = a + b1group + b2PRED + b3groupXPRED
|