There are several questions here. First, the "optimal" design is to have an equal number in both groups. But this assumes that you have a fixed total number that you need to divide between the two groups. It is always better to have more subjects, even if the groups are not balanced in numbers. Second, it is not "either-or" in terms of matching your groups and using the GLM. Matching is generally good experimental design because you don't know that the GLM will always account for differences. That said, many, many studies use unmatched designs. Third, even if you have matched subjects, it can be a good idea to include nuisance variables if they can explain extra variance.

On 7/1/18 11:16 AM, John Absher wrote:

External Email - Use Caution

In group analysis, is it better to have matched controls (1:1) or to use age, education and other characteristics as confounds in the glm? In the latter situation, we would control for these demographic variables in the glm and the control group n could be smaller, the same or larger than the test group. We are looking at TBI+ and TBI- subjects across a broad age range to look at patterns of cortical thinning. Thanks.

John R. Absher, MD, FAAN

GHS Neurosciences

jabsher@ghs.org

864-350-6655 (mobile)
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer