Dear Kristian
It should work fine on categorical regressors out of the box. If I remember correctly, the probability for each subject being placebo or vaccination should be in the bottom plot of the graphics generated by spm_dcm_loo. If not, you could always put the predicted values through a sigmoid (logistic) function to binarize it.
Let me know if you'd like me to take a closer look.
Best
Peter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Kristian Stefanov
Sent: 13 November 2019 10:41
To: [log in to unmask]
Subject: [SPM] Leave_one_out_cross_validation_categorical
Dear SPM developers,
I would like to know if there is any way of modifying the spm_dcm_loo code so that it estimates and tests the procedure on categorical outcome variables rather than continuous ones. I have a design in which I want to test whether certain connections predict whether the fMRI scan came from a placebo or a vaccination condition (coded for as -1 and 1 in the M.X matrix (second column)). I would really appreciate your help.
Best wishes,
Kristian Stefanov
|