Hello,
I understand that there are two types of models that can be set up for an unpaired two-group design with additional covariates (i.e. age or gender). The first is a design in which BOLD is compared between groups while correcting for group differences in the covariate. The second is an interaction design in which the within-group effect of the covariate on BOLD is determined for each group and then compared between groups. I am confused about how to demean (mean center) both of these types of designs.
The first type (between-group BOLD comparison while correcting for group differences in covariate) is of greater interest because we have a significant difference in age between groups. Different sources tell me different things regarding demeaning for this setup.
This source tells me to demean using the overall mean: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_Adjusted_for_Covariate
And this source tells me not to demean at all: http://mumford.fmripower.org/2010_summercourse/setting_up_designs_2010.pdf#page=26
Which is correct for this model?
The two sources are also contradictory when describing the interaction setup.
This source says to demean using the overall mean, and set the group column to all ones: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two_Groups_with_continuous_covariate_interaction
And this source says to demean using separate group means, and set the group column to 1's and 2's depending on group: http://mumford.fmripower.org/2010_summercourse/setting_up_designs_2010.pdf#page=15
Which is correct for the interaction model?
Thank you in advance for your help.
Musen Lab
|