Dear David, dear Helmut,
In principle one could of course compute a phantasy "accurate" Z score.
I believe the actual threshold is (or at least used to be) 8 rather than
7, as explained by Karl in 2000 and dug out in 2002:
" When the p value is too small to compute. Z value equivalents are
computed from the p values associated with the T statistic. If this is
value is less then 'eps' the Z score is set to 'Inf'. Simply report the
Z
score as >8.
I hope this helps - Karl"
I, too, hope this helps :-)
BW, A
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Helmut Laufs
Sent: 27 October 2007 20:08
To: [log in to unmask]
Subject: Re: [SPM] inf values for z-scores at the voxel level
Dear David,
As the Z-values correspond to the p-values, it's actually the p-value
that's
[near] zero, in turn depending on the T statistic. Hence, for the Z, an
Inf
results. For reporting in a manuscript, you could replace the Inf by a
">
[max on your colourbar|some other Z-score >7 e.g.]".
Hope this helps!
Helmut
----- Original Message -----
From: "David Carmichael" <[log in to unmask]>
Sent: Friday, October 26, 2007 4:55 PM
Subject: inf values for z-scores at the voxel level
> Dear All,
>
> When reporting the voxel level Z statistic values of 'Inf' are
reported
> in the graphical display for some positions.
>
> While comfortable with reasoning that this is due to a very
significant
> value (so X-mean(X) is a significant value and sd is small) I don't
> understand why you get an Inf rather than simply a large Z or what
(if
> any) is the cut-off for reporting Inf?
>
> Thanks
> David
>
|