Just to clarify what exactly do you mean by "interaction"?
("interaction" is a somewhat nebulous term, as indicated by Doug's and
mine differing responses).If you want to look for a difference in slopes of two main continuous effects, then a contrast vector can indeed be set up to handle that (as noted by Doug). If by interaction you mean the product of two continuous main effects, then that needs to be coded into the design matrix, and the caveats that I mentioned become applicable. cheers, Mike H. On Wed, 2009-08-05 at 15:10 -0500, Michael Harms wrote: > Hi Tori, > There is no way to investigate the interaction of two continuous effects > with the contrast vector alone, if the design matrix itself only > involves main effect terms. The interaction itself needs to be coded > into the design matrix. > > Note that it can become tricky as to the proper coefficients to use for > the contrast vector when you have both main effect and interaction terms > in your design matrix. (There are in fact different "types" of > statistics that can be estimated, and you want your contrast vector to > be an "estimable" function. Sophisticated statistical packages check > that this estimability constraint is satisfied, but I suspect that > mri_glmfit does not). > > e.g., > http://support.sas.com/onlinedoc/913/docMainpage.jsp > > cheers, > Mike H. > > > On Wed, 2009-08-05 at 15:48 -0400, Victoria Williams wrote: > > Hello, > > > > I apologize if this is a very simple question, but how would you set up a > > contrast matrix to look at the interaction of 2 variables on a measure? > > For example, my fsgd file contains: > > > > class=1 > > variables=2 > > Input Subjid class var1 var2 > > > > When I tried using a contrast of 0_1_1 the significance map looks very > > similar to the glm run only containing var1 (contrast_0_1). I thought > > 0_1_1 was correct, but now I'm not so sure and would like to double check. > > What I am interested in is how the interaction of var1 and var2 is > > correlated with the brain measure per voxel. Is this possible with > > mri_glmfit? > > > > I've checked the tutorials and scoured the internet as not to ask any > > silly questions, but I'm having trouble finding anything helpful. > > > > Thanks in advance! > > Tori > > _______________________________________________ > > Freesurfer mailing list > > Freesurfer@nmr.mgh.harvard.edu > > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
