Dear Freesurfur users,
I have a question about the design for a surface based analysis! I want to compare controls and patients. I have 5 covariates. One is a continous variable. The others are categorial variables (4 with 2 levels and 1 with 5 levels). I want to ajust my result with all of thoses variables. I am interested in the difference in thickness between patients and controls. Is it possible to construct such design? Can I put some of my variables in the "nuisance factors" ? I don't really understand the difference between "nuisance factor" and covariabes. I really hope somebody can help me. Thanks for all! All the best, Charlotte
Charlotte,
There is no difference, from a statistical perspective, of nuisance factors and covariates.
The key questions are: (1) Do any of these covariates have a differential effect on thickness between groups? (2) Do you want to compare the actual group means or the covariate-adjusted group means (e.g. test the means for subjects as if the subjects all had the same covariate values)? (3) Do you have enough subjects to add 5 covariates? (4) Do you have a valid reason for why all the covariates would have an effect on thickness? One can always add more covariates to reduce the error, but its not always a good thing to add covariates just to reduce the error of the model.
On 2/23/12, Charlotte Bernard chakeba@live.fr wrote:
Dear Freesurfur users,
I have a question about the design for a surface based analysis! I want to compare controls and patients. I have 5 covariates. One is a continous variable. The others are categorial variables (4 with 2 levels and 1 with 5 levels). I want to ajust my result with all of thoses variables. I am interested in the difference in thickness between patients and controls. Is it possible to construct such design? Can I put some of my variables in the "nuisance factors" ? I don't really understand the difference between "nuisance factor" and covariabes. I really hope somebody can help me. Thanks for all! All the best, Charlotte
Just to clarify, you can do this analysis in FS, but not in QDEC. You'll have to create an FSGD file. Do a search for FSGD on our wiki to get examples that you can generalize from. doug
MCLAREN, Donald wrote:
Charlotte,
There is no difference, from a statistical perspective, of nuisance factors and covariates.
The key questions are: (1) Do any of these covariates have a differential effect on thickness between groups? (2) Do you want to compare the actual group means or the covariate-adjusted group means (e.g. test the means for subjects as if the subjects all had the same covariate values)? (3) Do you have enough subjects to add 5 covariates? (4) Do you have a valid reason for why all the covariates would have an effect on thickness? One can always add more covariates to reduce the error, but its not always a good thing to add covariates just to reduce the error of the model.
On 2/23/12, Charlotte Bernard chakeba@live.fr wrote:
Dear Freesurfur users,
I have a question about the design for a surface based analysis! I want to compare controls and patients. I have 5 covariates. One is a continous variable. The others are categorial variables (4 with 2 levels and 1 with 5 levels). I want to ajust my result with all of thoses variables. I am interested in the difference in thickness between patients and controls. Is it possible to construct such design? Can I put some of my variables in the "nuisance factors" ? I don't really understand the difference between "nuisance factor" and covariabes. I really hope somebody can help me. Thanks for all! All the best, Charlotte
freesurfer@nmr.mgh.harvard.edu
-
Charlotte Bernard -
Douglas N Greve -
MCLAREN, Donald