Thanks again for the clarification, Steve.
I am just a little bit confused by you last point:
> Case 1 would be fine. This is a good idea, as you can use the output
> from the group-mean non-differential testing as a pre-threshold mask
> for the differential testing, which will reduce the multiple-
> comparison-correction.
What exactly do you call "differential testing"? Here I just want to have the mean affect for each condition at the group level, not compare them with one another...
Thanks, your help is really appreciated!
Best,
Stephane
> > On Wed, 29 Nov 2006 11:08:33 +0000, Steve Smith
> > <[log in to unmask]> wrote:
> >> Hi - this is pretty much right, though it does include cross-session
> >> and cross-subject variance within one analysis, which means that if
> >> you intend this to be a cross-subject mixed-effects analysis (which
> >> you probably do) then you are overestimating the degrees-of-freedom
> >> in this analysis and your results will be over-liberal. Ideall, you
> >> should move the cross-session analyses into a second-level analysis,
> >> and bring single image summaries for each condition of each subject
> >> up into the quadrupled-t-test-across-subjects for a third-level
> >> analysis (ie a combination of what you have here and the three-level
> >> example in the manual).
> >>
> >> Cheers, Steve.
> >>
> >>
> >> On 28 Nov 2006, at 19:05, Stephane Jacobs wrote:
> >>
> >>> Hello everyone,
> >>>
> >>>
> >>> Sorry to bug you with the details of my analysis, but I would need
> >>> to check
> >>> that what I'm doing is correct... I have of course looked up the
> >>> archives,
> >>> but did not find the answers to ALL my questions.
> >>>
> >>> I have 16 subjects, each scanned under 4 conditions, 2 or 3 runs per
> >>> condition (depending on the subject). I would like to contrast these
> >>> conditions with each other. I therefore used an extended version of
> >>> the
> >>> Tripled T-test described in the FEAT manual, with a design as
> >>> follows (with
> >>> only 2 subjects, 3 runs per condition, for the sake of clarity):
> >>>
> >>> EV1 EV2 EV3 EV4 EV5
> >>> S1 – Ar1 1 1 1 1 0
> >>> S1 – Ar2 1 1 1 1 0
> >>> S1 – Ar3 1 1 1 1 0
> >>> S1 – Br1 -1 0 0 1 0
> >>> S1 – Br2 -1 0 0 1 0
> >>> S1 – Br3 -1 0 0 1 0
> >>> S1 – Cr1 0 -1 0 1 0
> >>> S1 – Cr2 0 -1 0 1 0
> >>> S1 – Cr3 0 -1 0 1 0
> >>> S1 – Dr1 0 0 -1 1 0
> >>> S1 – Dr2 0 0 -1 1 0
> >>> S1 – Dr3 0 0 -1 1 0
> >>> S2 – Ar1 1 1 1 0 1
> >>> S2 – Ar2 1 1 1 0 1
> >>> S2 – Ar3 1 1 1 0 1
> >>> S2 – Br1 -1 0 0 0 1
> >>> S2 – Br2 -1 0 0 0 1
> >>> S2 – Br3 -1 0 0 0 1
> >>> S2 – Cr1 0 -1 0 0 1
> >>> S2 – Cr2 0 -1 0 0 1
> >>> S2 – Cr3 0 -1 0 0 1
> >>> S2 – Dr1 0 0 -1 0 1
> >>> S2 – Dr2 0 0 -1 0 1
> >>> S2 – DR3 0 0 -1 0 1
> >>>
> >>>
> >>> And I have set the contrasts as following, assuming that the weight
> >>> for each
> >>> contrasts would be A = a + b + c; B = -a; C = -b; D = -c (a for
> >>> Ev1, b for
> >>> EV2 and c for EV3).
> >>>
> >>>
> >>> EV1 EV2 EV3 EV4 EV5
> >>> A – D 1 1 2 0 0
> >>> D – A -1 -1 -2 0 0
> >>> B – C -1 1 0 0 0
> >>> C – B 1 -1 0 0 0
> >>> A – B 2 1 1 0 0
> >>> B – A -2 -1 -1 0 0
> >>> D – C 0 1 -1 0 0
> >>> C – D 0 -1 1 0 0
> >>> AB – CD 0 2 2 0 0
> >>> CD - AB 0 -2 -2 0 0
> >>>
> >>>
> >>> So, basically my question is simple: is this correct? :-)
> >>>
> >>> I think it is, if I understood the Tripled T-test example, but it's
> >>> counter-intuitive enough for me to still doubt... Especially since
> >>> I average
> >>> across sessions within each subject.
> >>>
> >>>
> >>> Also, I will run a separate second level analysis to get the mean
> >>> group
> >>> effect for each condition, with an EV per condition, and an input
> >>> per run
> >>> within each subject. Still correct?
> >>>
> >>>
> >>> Thanks a lot for all your help!
> >>>
> >>> Best regards,
> >>>
> >>> Stephane
> >>
> >>
> >> ---------------------------------------------------------------------
> >> ---
> >> ---
> >> Stephen M. Smith, Professor of Biomedical Engineering
> >> Associate Director, Oxford University FMRIB Centre
> >>
> >> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> >> +44 (0) 1865 222726 (fax 222717)
> >> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> >> ---------------------------------------------------------------------
> >> ---
> >> ---
> >>
>
>
> ------------------------------------------------------------------------
> ---
> Stephen M. Smith, Professor of Biomedical Engineering
> Associate Director, Oxford University FMRIB Centre
>
> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> ------------------------------------------------------------------------
> ---
>
|