Dear Donald, Thank you very much for the explanation. I have found much trouble in modeling my data, and you are already helping a lot! I believe the "medication problem" is very hard to solve, but perhaps one or two models can be useful although reliant on a few more premisses. Without them, it would be impossible to attribute group differences to a "diagnosis" if only patients received a certain type of medication. If I got it correctly, using a DOSS model would assume that the slope for medication (covariate) in the controls (unknown, as no control received it) is the same as the patients' slope for medication (known). This is a very strong biological assumption, I know. Another option would be to use a DOSS / DODS (if the patients group is divided, let's say, by gender) model excluding the controls. Am I following your thoughts here? Best, Pedro.
On Mon, May 9, 2016 at 10:15 PM, MCLAREN, Donald mclaren.donald@gmail.com wrote:
Pedro,
Instead of saying that you want to include the covariate in the analysis, it's better to consider why you want to include the covariate and how it changes the interpretation. You don't necessarily want to regress out covariates, especially when the covariate is different between groups.
DO -- Different offsets, no covariate/no slope --> the offsets are the group means not adjusted for covariates. DOSS -- Different offset, same slope model --> If you don't subtract the mean, then the offsets are the group means when the covariate is 0 for all subjects. If you subtract the mean, then the offsets are the group means when the the covariate is the mean covariate for all subjects. The difference in offsets won't change with mean centering the covariate. DODS -- Different offset, different slope model --> Not possible because the covariate for controls is collinear with the group term.
There are other models that you could construct - such as only including a covariate for the drug group, but the interpretation will once again be different.
The bottom line is to decide what you want to test and how you want to interpret the results before adding covariates. With DOSS, you will reduce/increase the group differences (depending on the slope of the covariate) because you are interpreting the results when the covariate is 0 in the drug group.
Best, Donald
Best Regards, Donald McLaren, PhD
On Sat, May 7, 2016 at 3:18 PM, Pedro Rosa pedrogomesrosa@gmail.com wrote:
Dear list, I am running a command line group analysis, and I want to include a covariate that is zero for all subjects in a group, and diverse for all subjects in the second group (medication intake, which is null for all subjects in the control group). This generates a lack of range of that continuous variable within a class (control group), and thus mri_glmfit ends with errors. Is it possible to perform such analysis? Could demeaning procedure work here? If not, how could I "regress out" the effect of such covariate in between-groups thickness/area differences? Many thanks in advance, Pedro.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.