 |
 |
Dear Torben, I am afraid my pen got a little ahead of my brain when I wrote yesterdays mail. My sincere appologies for that. Below is what I intended to write. > > Thank you very much for your reply, you seemed to have got my point. So now for > the explanation of why would I like non-integer df's. I second level analyses > one uses contrast images from different persons whom are likely to have > different degrees of paradigm related motion. If one of the motion parameters > have correlation with the paradigm of more than some threshold (say 0.4 ), I > consider the contrast image belonging to the paradigm to uncertain, and exclude > it from the second level analysis. If some correlation between one of the motion > parameters are close to this threshold, I would like to let this information go > into the second level analysis, as a covariate. Clearly the six motion > parameters are not orthogonal, but in the current setup of df calculation they > will take 6 degrees of freedom. If I were to include a single correlation > coefficient in the second-level analysis I would need some apriori knowledge of > how to weight the different parameters, and here my common sense is not good > enough. > > Torben If I understand you correctly you would like to extract some (single) parameter from each subject that somehow indicates to what degree that data set has been corrupted by movement. It doesn't sound easy. Spontaneously I would guess that a pretty kosher way of doing that would be a canonical correlation analysis between the (first few) eigenvariates (following an SVD) of your data, and your estimated motion parameters. That would give you the linear combination of your (6) motion parameters that best explains observed variance in your data. CHANGES FROM HERE The correlation coefficient between that linear combination and the condition regressor could then be used at the second level. For studies involving more than one trial type (condition) I guess the correlation coefficient between the design matrix multiplied with the first level contrast weight vector and the linear combination described above should be used. END CHANGES On a pragmatical note I guess you could pick any of the z-translation or the pitch (x-rotation) since those are almost always the largest, and strongly correlated with each other. Finally, my choice would be to handle the motion induced task correlated variance at the first level to as large an extent as possible. Including the motion parameters in the first level model will ensure that you err on the conservative side. Granted, there is an extra little complication in that a large degree of task related motion -> a lot of "true" activation variance will be removed at the first level -> variance of activations across subjects may increase AND in a task_by_group interaction one might conclude that one group activates less, while in reality they just moved more with the task. Good luck Jesper