Dear SPM list and Will Penny:
I have a 40 subject mothers listening to infant cries experiment. On each
subject, I have contrast images for fMRI activity during a mother's own
infant's cry minus its just preceeding rest baseline (Own cry contrast) and a
standard (other) infant cry minus its just preceeding rest (Other cry
contrast). I have both breast and bottlefeeders and mulitiparous and
first-time mothers. Bottlefeeders as its turns out are less educated than
breastfeeders. Multiparous mothers are slightly older than first-time
mothers.
I am interested in the effects of parity, breastfeeding status, as well
as their interaction on the fMRI signal and want to adjust out the
age and education effects (effects of no interest).
I have setup a multiple regression analysis for OWN and Standard cries
contrasts separately (I was not sure if this was necessary but it was
easier to think about for an initial analysis)
Here is the one I set up for own cry. I took the scans for own cry minus
rest for each of the 40 moms.
(1) column 1 = covariate has 1's for brestfeeding, first time moms; 0's
elsewhere
(2) column 2 = 1's for brestfeeding, multiparous moms and
0's elsewhere
(3) column 3 = covarite has 1's for nonbrestfeeding, first time moms; 0's
elsewhere
(4) column 4 = 1's for nonbrestfeeding, multiparous moms and
0's elsewhere
(5) column 5= age of mother
(6) education level of mother
My question are:
(1) How is spm handling error terms for a contrast such column 1 minus
column 2 versus a different contrast such as column 3 versus 4--> is it
always the same variance / residual being used or is the adjustment for
individual columns variances hidden in the code.
(2) Is a contrast such as looking at a t-test in this model of column 1
only (1 then all 0 ) valid if I wanted to know brain effects in
breast feeding first time moms only
(3) If I wanted to look at effects of age on brain signal (column 5), how
is spm handling the error term as the beta coefficient would have a much
different value than that of say the ones and zeros in columns 1 to 4.
This is of interest because if the same error term is used for each
column or contrast then I would have to scale this column in order to get
it in range of the others
(4) Do I need to orthogonalize or mean center the covariates?
(5) Do I need extra columns to account for
age and education differing in the 4 groups of the first 4 columns
Thank you for any help.
Sincerely,
Jeffrey Lorberbaum, MD
|