Sorry, that should have been interaction among continuous (not necessarily nuisance) variables. You will need to create new variables that are products of your variables of interest in order to test for interactions.
doug
On 3/30/14 9:20 PM, Clint Johns wrote:
I'm not sure I understand (sorry for being obtuse!) We don't want to check for interactions among nuisance variables (inasmuch as we only have the one, the average thickness). Rather, we want to check for interactions among the four factors revealed by our PCA, controlling for the nuisance variable. Without the nuisance variable, we have 5 regressors, and the nuisance variable makes six.
So, if I *do* understand you, we did right by including the mean thickness as a new regressor (the final '1' in the various .fsgd files we created, where the various other combinations of 1s reflect the other potential interactions of our 4 variables). So the nuisance variable effectively doubles the number of contrasts, so we have .fsgd files for possible interactions absent the nuisance variable, and a second set that includes the nuisance variable.
Again, I'm sorry if I'm missing the obvious. I'm really just not sure that we have formatted them.fsgd files correctly!
Thanks for your time and attention.
clint
On Sunday, March 30, 2014, Douglas Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> wrote:
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug On 3/30/14 4:20 PM, Clint Johns wrote:I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6 Null contrast is 1 0 0 0 0 0 Main effect, e.g., of the working memory component (#1) is 0 1 0 0 0 0 Main effect, e.g., of the WM component modulated by the nuisance variable is 0 1 0 0 0 1 So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is 0 1 1 0 0 0 And with the nuisance variable 0 1 1 0 0 1 Do we have this right? THANKS!! Clint -- I remain... Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu <javascript:_e(%7B%7D,'cvml','johns@haskins.yale.edu');> _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <javascript:_e(%7B%7D,'cvml','Freesurfer@nmr.mgh.harvard.edu');> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu mailto:johns@haskins.yale.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer