No. If -nolegendre is on, the baseline functions for polort=1, at the n-th time point within a run, are
1 and
n, for n=0,1,....
If -legendre is on, the baseline functions will be
1 and
x(n), where x(n)=2*n/(nlast-1)-1 for n=0..nlast-1. The second function is a linear function, but not the same as
n -- it is (nearly) orthogonal to
1, which is the point. In particular, with -legendre, the coefficient of
1 is the correct estimate of the baseline mean, unlike the previous case, where you would have to use the coefficient of
1 plus 0.5*(nlast-1) times the coefficient of
n to get the baseline mean estimate.
Where -legendre is important is for polort > 1. It was a mistake to use power functions for the baseline model. They are still available with -nolegendre, in case people will want those coefficients (for backwards compatibility?), but otherwise -legendre is to be preferred. It will be the default in the next release of 3dDeconvolve (read on).
You must read the -help option very closely and frequently. This is one of a set of changes I am making to 3dDeconvolve at this time, and this one has crept out to the binaries on the Web site. The other changes underway include:
(a) replacing Gaussian elimination on the normal equations to solve the least squares problem with an SVD-based solution [done this morning] (can be disabled with the -nosvd option)
(b) displaying the X matrix condition number to inform the user of potential problems with the numerical solutions [done last week] (can be disabled with the -nocond option)
(c) check stimulus filenames and colums of the X matrix for duplicates [done last week]
(d) add direct input of stimulus timing followed by internal "waver-ing" to allow users to create regressors from timing specs rather than have to use two programs; this option would also allow deconvolution directly from stimulus timing, rather than the clumsy basis function approach which now requires a complex symphone of waver, 3dDeconvolve, glts, and 3dcalc to get any results [this is NOT done or even started]
(e) saving the design matrix in the output header so that post-hoc glts can be done without re-running the whole analysis [NOT done or started]
(f) interpreting a string like "30@0" in the input glt matrix files to mean 30 zeros in a row -- this will make it easier to deal with the large number of elements per row, most of which tend to be zero [working on this RIGHT NOW, if you will all stop pestering me]
Any other suggestions for 3dDeconvolve would be appropriate at this time. My mental energies are focused on this program for the moment, so request now actually have a chance to be carried out. Later may be too late!!!
bob cox