Posted by B. Douglas Ward on April 02, 2001 at 10:47:31:
In Reply to: compare the area under the curve posted by Jinghui Zhao on April 02, 2001 at 02:01:37:
Jing-hui Zhao:
Let me restate the problem, to see if I understand it correctly. You have
used 3dDeconvolve to compare the activations produced by two different tasks.
The -glt option was used to compare the "area under the curve" of the
corresponding impulse response functions. The results, for individual subjects,
show statistically different responses for the activated areas. However, when
you attempt to combine the results from different subjects, using 3dttest,
the results are not significant.
This is not too surprising. The claim that "Point A shows a significant
difference in response for the different tasks across the general population"
is much stronger, and hence more difficult to prove, than the claim that
"Point A shows a significant difference in response for the different tasks
for Subject #123".
A single time series can provide many measurements (many degrees of freedom)
for a single subject, and thus the statistical result can be highly
significant. However, this does not carry over to across-subjects analysis.
For example, if you have 100 data points for Subject A, and 100 data points
for Subject B, this does NOT mean that you have a sample of size 200 when
you combine the results from Subjects A and B. Unfortunately, the sample size
is only 2.
So, part of the problem may simply be due to the size of the sample (i.e., the
number of subjects) that you are using.
Also, it is VERY important that the Orig to Tlrc conversion be done correctly.
Due to anatomical differences between subjects, comparisons across subjects
are inherently difficult. Therefore, the Orig to Tlrc transformation must be
as precise as possible. This is often followed by Gaussian blurring, to make
allowance for differences in anatomical structures.
Doug Ward