Dear Stephen, dear Shashwath,
On Sunday 02 October 2005 19:13, Stephen J. Fromm wrote:
> On Mon, 19 Sep 2005 12:55:25 -0400, Shashwath Ashok Meda
>
> <[log in to unmask]> wrote:
> >Dear SPM group,
> >
> > I would really appreciated if someone could tell me if
> >the 2 sample t-test performed via SPM2 is a one tailed or a
> >two tailed analysis.
>
> 1-tailed:
>
> http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind04&L=SPM&P=188731
>
it might be a little bit confusing, but I think we have to distinguish between
t-contrasts and t-tests (one- and two-sample), the latter being models that
can be accessed by pressing the "Basic Models" button.
t-contrasts in spm are, as Stephen points out, always 1-tailed (independent of
the underlying model).
The models you specify (including the 2 sample t-test) are neither one- nor
two-tailed. What they are (or become) depends on the question you ask in
terms of contrasts you specify. If you specify a t-contrast (e.g. [1 -1 0])
you pose a directional question to the model, i.e. a one-tailed test will be
performed. Unlike t-contrasts, F-contrasts are always non-directional (again
independent of the underlying model). The "effects of interest" in a 2 sample
t-test for example is an F-contrast, which should yield the same
"activation"-blobs at alpha like two converse one-tailed t-contrasts together
([1 -1 0] and [-1 1 0]) thresholded at alpha/2 each.
Thilo
> >Thank You,
> >
> >Thank You,
> >
> >With regards,
> >
> >Shashwath Meda
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >***************************************
> >
> >Graduate Research Assistant
> >WSU MR Research Facility
> >Harper University Hospital
> >3990 John R
> >Detroit, MI - 48201.
> >phone(off): 313-993-8821
> > (home): 313-832-7149
> > (cell): 313-505-6847
> >Email: [log in to unmask]
> >
> >***************************************
> >=========================================================================
--
Thilo Kellermann
RWTH Aachen University
Pauwelstr. 30
52074 Aachen
Tel.: +49 (0)241 / 8089977
Fax.: +49 (0)241 / 8082401
E-Mail: [log in to unmask]
|