Hi all,

As a brief introduction, in the neuroimaging analysis, all the predictor variables (including grouping variable, independent variables, and covariates) are usually entered into a "design matrix" which is fed with "contrast"(s) to a statistical package/command (e.g. PALM, randomise). This package fits the model into each metric of image (e.g. voxel, vertex) with permutations and reports the results after family-wise error correction.

I have been using this method for a long time but recently encountered an issue analyzing the HCP Wu-Minn dataset with PALM. In this study, I selected a group of subjects based on specific criteria (case group) and matched another group of subjects (control group) based on age, gender, and education and tried to compare the brain activity in a task-fMRI between two groups. The big problem lies here: Do I have to still add these variables as covariates in the design matrix? Even if I have matched the two groups based on these major covariates?

My short answer was if the covariates have the same directionality of linear association with my main outcome variable (here task-fMRI activity), there is no need to add them into the design matrix. However, since the PALM, randomise, or any other statistical package applies the model (i.e. contrast) on each voxel, the linearity or association might be different between two groups in each voxel. For example, I expect "smoking" to be more related to hippocampal volume in schizophrenia subjects compared to unaffected individuals, thus I have to add smoking in the design matrix, even if I have matched my schizophrenia and healthy groups on smoking.

I will be glad to hear your opinions on this issue of adding covariates to the design matrix even if the two groups have been matched on those.

Bests,

Amir