A control file is the "top level" that defines the mixed-effects or random-effects model to be used for analysis.
The first several lines describe options to used in the analysis. After the key-word delimited top section of the file. Subsequently, the keyword “event” denotes that subsequent lines in the file should be interpreted as 1st-level analyses containing maps of signal changes and errors. Generally, one would use square (constant across groups) analysis regressors. However, if a weight is applied to each subject following the file specification, then the 2nd-level regressor will assume the shape specified by the list of weights, which presumably would come from some independently measured variable.
An example GLM control file follows:
time-model MEM # or “REM” , “rem”, “men"
smoothing 3 6 # [1st level resolution] [smoothing to control sliding scale between fixed and random effects analyses]
conditions A B A-B # list of conditions to test
event A # this indicates that following lines will list 1st-level analyses
../subject1/S-a-b.nii # a 1st-level differential test of 2 conditions
../subject2/S-a-b.nii # optionally, a weight could be applied to each subject
../subject3/S-a-b.nii
…
event B # “event” or “subjects” or “sessions"
../subject21/S-a-b.nii # a 1st-level differential test of 2 conditions
../subject22/S-a-b.nii
../subject23/S-a-b.nii
…
The following keywords are defined for use in the control file:
Keyword argument default value purpose
time-model rem/mem/REM/MEM
multiple-comparisons [integer # of resels] corrected p-values on output
conditions a list of conditions GLM determines these values & errors
event/subjects/sessions none separates the file into options (above) and files (below)
output-format [type] NIFTI use “jip” or “nifti” output files
smooth/smoothing [resolution] [smoothing factor] The first number is the effective 1st-level resolution, including smoothing
The 2nd number controls the sliding scale between fixed and random effects