Hi Jeremy,

1. "Does it matter what input you use and is there a benifit to using one in particular (in terms of using the binary 1D file or the preconvolved file) for the glt matrix of 00000000111111-1-1-1-1-1-1?"

Although 3dDeconvolve works for both kinds of input file, it does matter in the sense the user is interested toward two slightly different directions. With a stimulus file indexed with 0's and 1's, the program does deconvolution, and outputs the coefficients of the impulse response functions and their corresponding statistics. If you provide a time series generated from waver, the program basically does linear regression analysis instead of deconvolution.

Both approaches would allow you to do general linear tests among various stimuli, but with the first approach you have the full control of simulating the impulse response functions (i.e. max lag #) while the hemodynamic response functions are specified in waver (i.e. gamma variates) in the second approach.

With stimulus files generated from waver, usually you don't need a max lag number more than 0 because the basis functions you specified in waver intrinsically include the information about the response delay.

2. "What happens when you use just 000000001-1 for your glt, with max lag of 5, and the wavered(preconvolved) input(this is what we were doing)?"

With max lag of 5, 3dDeconvolve would sputter if you use 000000001-1 as your glt vector because there are totally 2 * 4 (baseline coefficients) + 2 * 6 (regressor coefficients) = 20 coefficients, more than you specified in the glt vector, which is 10.

3. "Is there a simple explanation as to how the above glt matrix is comparing the overall magnitude of each wave? Is it that all time points are included and if you don't than you only look at one time point? Or am I looking at it in the wrong way?"

If you are talking about the "area under the curve" approach with stimulus files of 0's and 1's and max lag of 5 , I am not so sure the following explanation of mine would help: Since the pure siginal is modeled as the convolution of the impulse response functions and the stimulus series, those coefficients together estimated in 3dDeconvolve at each voxel measure the "magnitude" of the hemodynamic response corresponding to that stimulus. Therefore the comparision between two stimuli is equivalent to running a general linear test between the two groups of coefficients.

Hope this helps,
Gang