Hi - yes, the first and second should indeed give the same results,
although see the FEAT manual which recommends the second as you can then
_also_ specify different variance estimation groups for the two A levels
(which _may_ help improve things).
The reason why the final example is different is in the residuals. Because
the model now has more flexibility (ie fitting the different C levels
independently) it can fit the data better and so the residuals go down. On
the other hand, the degrees of freedom also go down, so the final t/f
stats could go either way. But this model does probably make sense to use.
Cheers.
On Thu, 21 Apr 2005, Marcus Lauer wrote:
> I am trying to figure out the implications of different notations for
> describing a contrast.
>
> I am working on a study in which we manipulate three factors
> simultaneously, leading to a total of eight conditions (2x2x2). I want to
> compare one pair of conditions (A) summing across a second condition (B) at
> one level of the third (C). It seems to me that I could code this in several
> ways:
>
> EV1 EV2
> A1B1C1 1 1
> A1B2C1 1 1
> A2B1C1 -1 1
> A2B2C1 -1 1
> (repeat for each subject)
>
> Contr: 1 0
>
> or...
>
> EV1 EV2
> A1B1C1 1 0
> A1B2C1 1 0
> A2B1C1 0 1
> A2B2C1 0 1
> (repeat for each subject)
>
> Contr: 1 -1
>
> or...
>
> EV1 EV2 EV3 EV4
> A1B1C1 1 0 0 0
> A1B2C1 0 1 0 0
> A2B1C1 0 0 1 0
> A2B2C1 0 0 0 1
> (repeat for each subject)
>
> Contr: 1 1 -1 -1
>
> or...
>
> EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8
> A1B1C1 1 0 0 0 0 0 0 0
> A1B2C1 0 1 0 0 0 0 0 0
> A2B1C1 0 0 1 0 0 0 0 0
> A2B2C1 0 0 0 1 0 0 0 0
> A1B1C2 0 0 0 0 1 0 0 0
> A1B2C2 0 0 0 0 0 1 0 0
> A2B1C2 0 0 0 0 0 0 1 0
> A2B2C2 0 0 0 0 0 0 0 1
> (repeat for each subject)
>
> Contr: 1 1 -1 -1 0 0 0 0
>
> The advantage of the last coding scheme is that several comparisons
> can be made using the same .mat file. We might have multiple contrasts in
> the .con file comparing several sets of conditions. However, there is a
> problem with this coding scheme. I don't get exactly the same results (t and
> F maps) as I do with the first and second coding schemes (haven't tried the
> third yet). The results are similar, but not identical.
>
> Can anyone tell me why this is the case? I'm pretty certain that
> these differences are not caused by a typo or a misordering of files in the
> cope or varcope images. Whatever the reason, I need to understand what's
> going on so that I can figure out what notation is appropriate for the
> analyses I'm working on.
>
> --
> Marcus Lauer
> Lab Manager for the Curtis Lab, NYU Psychology Department
> 6 Washington Place, Rooms 875-876
> Phone: (212)998-8347
> http://psych.nyu.edu/curtislab/
>
--
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|