Hi Orjan, it looks like there are some errors in your design. The "Variables" function is only for continuous variables (covariates) such as Age. Sex and diagnosis are discrete variables and they should be coded as part of the class (eg, if you only wanted to add sex, then you'd have 4 classes Q1-Q4 for males and the same for females). doug
Ørjan Bergmann wrote:
Dear Dr. Greve,
I have been working with FreeSurfer for a while now, amongst others in collaboration with my colleague Dr. Lars-Morten Rimol here in Oslo. We have contected you on issues relating to FreeSurfer before, and hope you have some insights as to how to deal with a particular problem we've come across.
We have a set of subjects divided into four groups, depending on how they score on a particular test; the first group are those scoring in the lower 25% (i.e. Q1), those scoring from 25-50% (Q2), etc. Each of these groups additionally belong to one of 3 diagnostic groups; Scz, BD and NOS.
We have formulated this as a FSGD file: Class Q1 Class Q2 Class Q3 Class Q4 Variables Age Sex Scz BD NOS
We are interested in seeing if, when correcting for the age, sex and diagnosis, there are differences between the groups. E.g. a contrast vector would be [1 -1 0 0 0 0 0 0 0].
The problem is of course that the resulting design matrix resulting from this FSGD file becomes singular since its first 4 colums and its 3 columns each add up to 1. Clearly this design matrix is singular.
In the context of FSGD-files, do you know of any other way to solve this problem?
Mvh Ørjan Bergmann, PhD Post. Doc. TOP-prosjektet Universitetet i Oslo