Hi Donald and Doug,
Thanks for advice. Can I be a little bit clearer as I suspect I need a little more guidance.
When comparing controls (gp c) and whole patient group (gp p) I see interesting differences in area (DOSS model after checking for no interaction and with age and ICV as correction factors). I also see an interesting interaction between age and group (DODS model with ICV as nuisance factor) such that that thickness correlates stronger with age in one group compared to another group.
The patient group can be split into 4 subgroups (gp p1 to gp p4) depending on level of an endogenous substance. I want to see whether when running gp c against gp p1, if the interaction between group and age (for thickness) is stronger/weaker than the interaction when running gp p4 against gp c. I also want to see whether differences in area between a subgroup and controls are bigger/smaller compared to differences between other subgps and controls.
To do this can I use the original method (granted power will b smaller) or if I follow the method suggested by Donald should I demean individually within subgroups. Also is this kind of model qdec friendly, and if not could you offer any guidance on design matrix and contrasts.
Hope i make some sense !
Thanks.
Mahinda
It sounds like you need a conjunction to show that P4 > P1 AND P1 > C Does that make sense? If so, then run the analysis as Donald suggests, creating contrast matrices for the above contrasts. Then run mri_concat --conjunct p1gtc/sig.mgh p4gtp1/sig.mgh --o conjunction.mgh
doug
On 11/19/2012 01:36 PM, Mahinda Yogarajah wrote:
Hi Donald and Doug,
Thanks for advice. Can I be a little bit clearer as I suspect I need a little more guidance.
When comparing controls (gp c) and whole patient group (gp p) I see interesting differences in area (DOSS model after checking for no interaction and with age and ICV as correction factors). I also see an interesting interaction between age and group (DODS model with ICV as nuisance factor) such that that thickness correlates stronger with age in one group compared to another group.
The patient group can be split into 4 subgroups (gp p1 to gp p4) depending on level of an endogenous substance. I want to see whether when running gp c against gp p1, if the interaction between group and age (for thickness) is stronger/weaker than the interaction when running gp p4 against gp c. I also want to see whether differences in area between a subgroup and controls are bigger/smaller compared to differences between other subgps and controls.
To do this can I use the original method (granted power will b smaller) or if I follow the method suggested by Donald should I demean individually within subgroups. Also is this kind of model qdec friendly, and if not could you offer any guidance on design matrix and contrasts.
Hope i make some sense !
Thanks.
Mahinda
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
On Mon, Nov 19, 2012 at 12:36 PM, Mahinda Yogarajah y.mahinda@gmail.com wrote:
Hi Donald and Doug,
Thanks for advice. Can I be a little bit clearer as I suspect I need a little more guidance.
When comparing controls (gp c) and whole patient group (gp p) I see interesting differences in area (DOSS model after checking for no interaction and with age and ICV as correction factors). I also see an interesting interaction between age and group (DODS model with ICV as nuisance factor) such that that thickness correlates stronger with age in one group compared to another group.
This suggests that you need the DODS model for your analysis.
The patient group can be split into 4 subgroups (gp p1 to gp p4) depending on level of an endogenous substance. I want to see whether when running gp c against gp p1, if the interaction between group and age (for thickness) is stronger/weaker than the interaction when running gp p4 against gp c. I also want to see whether differences in area between a subgroup and controls are bigger/smaller compared to differences between other subgps and controls.
The difference between C and gp1 AND C and gp4 is the same as the difference between pg1 and pg4. If you are only wanting to make comparisons between patients, which it sounds like, then only enter the 4 patient groups. If you want to know where pg1 and pg4 differ from controls, you'll need a conjunction as suggested by Doug.
To do this can I use the original method (granted power will b smaller) or if I follow the method suggested by Donald should I demean individually within subgroups. Also is this kind of model qdec friendly, and if not could you offer any guidance on design matrix and contrasts.
Demeaning does not change the slope estimation. Demeaning only changes the group means. (1) If you do not demean, you get the group intercepts as if your covariate is 0; (2) if you demean across everyone, you get the covariate-adjusted group means; and (3) if you demean by group, then you get the group means. One additional point, if you have an interaction between group and a covariate, then you generally can't interpret the group means.
Hope this helps.
Hope i make some sense !
Thanks.
Mahinda
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Mahinda Yogarajah -
MCLAREN, Donald