Hi FSL Experts,
My goal is determine if there is a significant linear relationship between a cognitive variable measured in a disease population and DTI scalar values with randomise. I figured a simple correlation analysis would meet my needs. I'm not using controls in this analysis and my disease population is actually split up into three groups with increasing levels of disease severity. My covariates are age, education years, gender, and site of collection (all are demeaned). I'm also not interested in comparing strength of correlation of one disease group to another disease group.
I'm writing because there seems to be several ways to set up a design matrix for a correlation analysis (or I'm just terribly confused as usual) in randomise and I'm not sure what to use. Mainly, I'm confused on when to use a column of 1s as the first column in the design matrix and if I need to include disease group designation columns.
Here are the options I've come up with to create my design matrix:
1) No intercept or column(s) for group designation because I'm just treating the disease population (that is normally split into three groups) as one big group. Just columns for my covariates and the cognitive variable (column headers shown):
age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
Contrast file: [ 0 0 0 0 0 1] or [ 0 0 0 0 0 -1]
2) Same as option 1 but added an intercept column of 1s, which I guess could represent the disease population, like this link: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Single-Group_Average_with_Additional_Covariate. In all my design matrix options, I'm not showing the Group column that is shown in all your Glm_gui examples.
intercept age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
Contrast file: [ 0 0 0 0 0 0 1] or [ 0 0 0 0 0 0 -1]
3) Same as option 1 but now each disease group has its own column where membership is represented as 1, like this link: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_Adjusted_for_Covariate
group1 group2 group 3 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
0 1 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
0 1 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
0 0 1 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
0 0 1 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
Contrast file: [ 0 0 0 0 0 0 0 0 1] or [ 0 0 0 0 0 0 0 0 -1]
4) Same as option 3 but now has an intercept column.
intercept group1 group2 group 3 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 1 0 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 1 0 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 1 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 1 0 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 0 1 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
1 0 0 1 age_DM educ_DM gender_DM site1_DM site2_DM cog_DM
Contrast file: [0 0 0 0 0 0 0 0 0 1] or [ 0 0 0 0 0 0 0 0 0 -1]
One last question: If I were to run a correlation analysis for one out of my three disease groups, would I have to demean the covariates within that one disease group or across my disease population?
Thank you for reading.
Joy
|