Posted by B. Douglas Ward on March 15, 2001 at 10:07:08:
In Reply to: 3dDeconvolve for t-test posted by Phoebe on March 14, 2001 at 18:02:48:
Phoebe:
If your intention is to do a "pure" t-test, i.e., without considering any
temporal dependence in the data, then you must set maxlag to 0 (or simply
delete the -stim_maxlag option).
I used both 3dfim+ and 3dDeconvolve to analyze a sample time series.
I created file bold.1D to represent the "measurement" time series for a single
voxel. Try this using your own data:
The script file for running 3dDeconvolve is as follows:
3dDeconvolve \
-input1D bold.1D -nfirst 20 -nlast 139 -polort 1 \
-num_stimts 1 -stim_file 1 para.1D -stim_label 1 "para" \
-fout -rout -tout -bucket bold.buck
The screen output is:
-------------------------------------------------------------------------------
Program: 3dDeconvolve
Author: B. Douglas Ward
Initial Release: 02 Sept 1998
Latest Revision: 11 Jan 2001
Results for Voxel #0:
Baseline:
t^0 coef = 16186.6399 t^0 t-st = 156.2191
t^1 coef = 1.8223 t^1 t-st = 1.5848
Stimulus: para
h[ 0] coef = -9.9112 h[ 0] t-st = -1.1730
R^2 = 0.0116 F[ 1,117] = 1.3760 p-value = 2.4317e-01
Full Model:
MSE = 190377.2188
R^2 = 0.0116 F[ 1,117] = 1.3760 p-value = 2.4317e-01
-------------------------------------------------------------------------------
The script file for executing 3dfim+ is:
3dfim+ \
-input1D bold.1D -nfirst 20 -nlast 139 -polort 1 \
-ideal_file para.1D -out All
And the corresponding screen output is:
-------------------------------------------------------------------------------
Program: 3dfim+
Author: B. Douglas Ward
Date: 18 May 2000
Results for Voxel #0:
Fit Coef = -9.9113
% Change = -0.6069
% From Ave = -0.6081
Baseline = 16331.5029
Average = 16298.4658
Correlation = -0.1078
% From Top = -0.6106
Topline = 16232.3906
Sigma Resid = 436.3224
-------------------------------------------------------------------------------
Now, I claim that the above results are equivalent. From the screen output,
we see that 3dDeconvolve calculated t-stat = -1.1730, and program 3dfim+
calculated CC = -0.1078.
To find the corresponding p-values, use program cdf (taking care to enter the
correct parameters):
cdf -t2p fitt 1.1730 117 <-- t-stat with 117 dof
p = 0.243178
cdf -t2p fico 0.1078 120 1 2 <-- correlation coefficient
p = 0.24323
As you see, the p-values are (approx.) equal.
Doug Ward