Hello Yufeng:

A1: The numerator df for the GLT F-stat is equal to the number of linear
constraints; i.e., the number of rows in the GLT matrix. In this case,
since there is one row, the numerator df = 1. The denominator df is equal
to the (error) degrees of freedom from fitting the Full Model. This can
be calculated as follows (for Example 1.4.4.5):

den df =
20 (number of data points)
- 2 (max. of the maxlags)
- 2 (constant and slope baseline parameters)
- 3*3 (3 stimuli, 3 time lags each)
= 7

The df for the t-stat is the same as the denominator df of the F-stat (i.e.,7).
Note that, in the special case where the numerator df of the F-stat is 1,
the F-stat is equal to the square of the t-stat (1.5420^2 = 2.37776).


A2: As illustrated above, the denominator df is essentially equal to the
number of data points minus the number of estimated parameters. Neither the
numerator nor denominator df depends on the number of trials of each stimulus.
This does NOT mean that the statistical result itself is independent of the
number of trials of each stimulus.


For more details, please see Sections 1.2.11 and 1.4.4 of the documentation
in file 3dDeconvolve.ps.

Doug Ward