 |
 |
Dear Glen Here are the equation which SPM uses: t=c'*beta/sqrt(var(c'*beta)) ' means transpose and c'*beta is your contrast image. This can be written as: t=c'*beta/sqrt(sigma^2*c'*pinv(X'*V*X)*c) pinv is the pseudoinverse and sigma is sigma^2= e'*e/(J-rank(X), where J is the number of observations V is the covariance matrix which for i.i.d. errors is simply the identity matrix, but if you have chosen to model dependency it is a pooled estimate over all voxels in an effects of interest mask, estimated using spm_reml. Best Torben Torben Ellegaard Lund Associate Professor, PhD Center of Functionally Integrative Neuroscience (CFIN) Aarhus University Aarhus University Hospital Building 10G, 5th floor, room 31 Noerrebrogade 44 8000 Aarhus C Denmark Phone: +4589494380 Fax: +4589494400 http://www.cfin.au.dk [log in to unmask] Den Uge:4 25/01/2012 kl. 15.39 skrev Martin Dietz: > Dear Glen, > > For a single subject, the SPM t or F is simply based on the rank of the first level design matrix and the within-subject variance > > Martin > > Martin Dietz > [log in to unmask] > > > On Jan 25, 2012, at 2:30 PM, Glen Lee wrote: > >> Dear SPMers, >> >> Can anybody explain based on what the SPM t-map was generated for a particular contrast (e.g., condition vs. rest) at a single subject (first) level? >> t-values at the second level RFX seems straight forward to me as we could calculate the variance and df from many subjects, but I'm still not sure how >> t-value was calculated from an individual contrast image. >> >> Thanks in advance, >> Glen >> >>