Dear Angela,
The mapping between the T-value and p-value is nonlinear, monotonic
and analytic.
However this mapping changes with the search volume. The only useful
p-value is the
p-value that has been adjusted to control for false positives
(referred to colloquially as
the corrected p-value). [The uncorrected p-values (for height and
extent) can only be
applied to single voxels or clusters that have been identified a
priori. In the absence
of this a priori specification are not true p-values].
The p-values are a function of the T-value, and the search
volume. This function is
encoded in spm_P and its sub-functions (the help is below). You will
see that there are
other arguments such as the degrees of freedom, but these will be
fixed for any SPM{T}.
The search volume R can change depending on the degree of anatomical
constraint used
in inference. If you look at the p-values before and after a small
volume correction (SVC)
you will see that they change.
I hope this helps - Karl
>I have a question about SPM. Is there a one-to-one linear mapping between
>the t-statistics stored in spmT.img files and the p-values?
Returns the [un]corrected P value using unifed EC theory
FORMAT [P p Em En EN] = spm_P(c,k,Z,df,STAT,R,n,S)
c - cluster number
k - extent {RESELS}
Z - height {minimum over n values}
df - [df{interest} df{error}]
STAT - Statisical feild
'Z' - Gaussian feild
'T' - T - feild
'X' - Chi squared feild
'F' - F - feild
'P' - Posterior probability
R - RESEL Count {defining search volume}
n - number of component SPMs in conjunction
S - Voxel count
P - corrected P value - P(n > kmax}
p - uncorrected P value - P(n > k}
Em - expected total number of maxima {m}
En - expected total number of resels per cluster {n}
EN - expected total number of voxels {N}
___________________________________________________________________________
