Hi!

I have two between subjects contrasts: Treat & Sham and two conditions: Pre- & Post-.

I  want to use CONN to do a Time*Group ANOVA and a correlation analysis between clinical changes and functional connectivity changes. The question I have is how to add covariates of no interest (i.e:Age; Baseline clinical scores) into these models.

Thanks a lot!

Liu

Hi Liu,

Typically you add covariates of no interest to any given analysis simply by adding the corresponding second-level covariates into the set of selected subject effects (in the 'subject effects' list) and then filling with zero the corresponding entries in the between-subjects contrast. Also typically covariates of no interest are centered (so that the effects of interest are estimated at the average group level) unless the zero-level of the covariate is of particular interest. 

For example, if you want to look at associations between clinical changes and changes in functional connectivity, separately for Treat & Sham groups, and compare the aforementioned associations between these two groups, you would do that by:

a) create two covariates containing the clinical score changes separately for each group, for example 'ClinicalChange_Treat' and 'ClinicalChange_Sham' each containing the clinical score changes for subjects in the corresponding group and 0's for the subjects in the opposite group

b) select 'Treat', 'Sham',' 'ClinicalChange_Treat', and 'ClinicalChange_Sham' in the subject-effects list, and enter a [0 0 1 -1] between-subjects contrast

c) select 'Pre' and 'Post' in the conditions list and enter a [-1 1] between-conditions contrast

Now if, in addition, you want to control these associations by age (e.g. let's say just main associations between age and connectivity-changes, if there are no reasons to expect age-by-group interactions), you may do so simply by adding 'age' to the list of subject-effects and adding a zero to the contrast. For example, by changing step (b) above to:

b) select 'Treat', 'Sham',' 'ClinicalChange_Treat', 'ClinicalChange_Sham', and 'Age' in the subject-effects list, and enter a [0 0 1 -1 0] between-subjects contrast

Note that the within-subject design already controls for main age effects on connectivity (i.e. associations between functional connectivity and age). What you are additional controlling for by adding age to this design are potential associations between age and connectivity-changes (differences between post- and pre- intervention connectivity values). 

All of the above applies similarly to the case where you are mainly interested in Time*Group interactions (e.g. ANCOVA with clinical changes as a covariate), where the corresponding contrasts in (b) would simply look like [1 -1 0 0]. Just as above, this within-subject design is already intrinsically corrected for associations between age and connectivity, but if you want to, in addition, control for potential associations between age and post- vs. pre- connectivity-changes, you could do so simply by adding an age covariate as before and entering a contrast [1 -1 0 0 0]. 

Hope this helps
Alfonso
Originally posted by Yilin Liu:

Hi Alfonso,

Thank you so much for the detailed answer. Some follow-up questions though.

If I define the contrast like 'Treat', 'Sham', 'ClinicalChange_Treat', 'ClinicalChange_Sham', 'age' [0,0,1,-1,0], this allow me to do an one-way ANCOVA to compare pre- vs. post-treatment connectivity associated with clinical changes while use age as covariate.

If the contrast looks like 'Treat', 'Sham', 'ClinicalChange_Treat', 'ClinicalChange_Sham', 'age' [1,-1,0,0,0], this ANCOVA model estimates the between subjects factor of group while clinical changes and age are control variables.

Have I understood correctly?

Another question is that when I did export mask in the results explorer, the dimension of the image changed from 91*109*91 to 74*92*78. Is there any setup that I missed?

Many thanks.
Best regards,
Hi Liu,

Typically you add covariates of no interest to any given analysis simply by adding the corresponding second-level covariates into the set of selected subject effects (in the 'subject effects' list) and then filling with zero the corresponding entries in the between-subjects contrast. Also typically covariates of no interest are centered (so that the effects of interest are estimated at the average group level) unless the zero-level of the covariate is of particular interest. 

For example, if you want to look at associations between clinical changes and changes in functional connectivity, separately for Treat & Sham groups, and compare the aforementioned associations between these two groups, you would do that by:

a) create two covariates containing the clinical score changes separately for each group, for example 'ClinicalChange_Treat' and 'ClinicalChange_Sham' each containing the clinical score changes for subjects in the corresponding group and 0's for the subjects in the opposite group

b) select 'Treat', 'Sham',' 'ClinicalChange_Treat', and 'ClinicalChange_Sham' in the subject-effects list, and enter a [0 0 1 -1] between-subjects contrast

c) select 'Pre' and 'Post' in the conditions list and enter a [-1 1] between-conditions contrast

Now if, in addition, you want to control these associations by age (e.g. let's say just main associations between age and connectivity-changes, if there are no reasons to expect age-by-group interactions), you may do so simply by adding 'age' to the list of subject-effects and adding a zero to the contrast. For example, by changing step (b) above to:

b) select 'Treat', 'Sham',' 'ClinicalChange_Treat', 'ClinicalChange_Sham', and 'Age' in the subject-effects list, and enter a [0 0 1 -1 0] between-subjects contrast

Note that the within-subject design already controls for main age effects on connectivity (i.e. associations between functional connectivity and age). What you are additional controlling for by adding age to this design are potential associations between age and connectivity-changes (differences between post- and pre- intervention connectivity values). 

All of the above applies similarly to the case where you are mainly interested in Time*Group interactions (e.g. ANCOVA with clinical changes as a covariate), where the corresponding contrasts in (b) would simply look like [1 -1 0 0]. Just as above, this within-subject design is already intrinsically corrected for associations between age and connectivity, but if you want to, in addition, control for potential associations between age and post- vs. pre- connectivity-changes, you could do so simply by adding an age covariate as before and entering a contrast [1 -1 0 0 0]. 

Hope this helps
Alfonso
Originally posted by Yilin Liu: [/quote]