I have an fmri experiment of 4 runs, in which subject make familiarity judgement about words on a 5-point scale. Each word also has its feature overlap (ranging from 0 to 1) taken from a database. I specified the 1st level design matrix with the 5 possible response options as separate conditions (constant-duration boxcar), and feature-overlap as a parametric modulator.
We would like to test a linear contrast of familiarity ((-2 -1 0 1 2) applied to unmodulated columns), an interaction between familiarity and feature-overlap (which to my understanding is the same linear contrast but applied to modulated columns), and a main effect of feature overlap (which to my understanding would be (0.2 0.2 0.2 0.2 0.2) applied to modulated columns). Since the familiarity condition is based on subjects' judgement, some conditions may not appear in every run (i.e. 0 trial), I also have a script to delete those conditions in the design matrix and adjust the contrast weights for each run to account for difference in the number of conditions. However, for some runs and some conditions, there is only 1 trial, in this case the unmodulated conditions are fine but the modulated ones become all 0s. Based on a previous post on this forum (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=spm;1d7cc8e2.99), that means any contrast weights involving those columns will need to add up to 0, which is not the case for the main effect contrast of feature-overlap.
I can think of two solutions for this: 1)remove those conditions with only 1 trial at the model specification stage. 2)still model them but not include them in the contrast.
I feel like solution 2) is more appropriate. Am I understanding this correctly?
|