Hi,

The same applies: at first-level you only model signals over and  
above baseline, i.e. do not model rest as a separate EV.  'Rest' in  
the FEAT sense is everything that has not been modelled out  
otherwise, i.e. given implicitly once we know the entire design.  
Given that there is no parameter estimate which models 'rest' it  
cannot appear in any of the contrasts. The "Win" question then is a 1  
0 0 0 contrast in the case where you have 4 original EVS to model the  
4 outcome types. This means the 1st PE indicates where there is any  
response under the win condition (over and above baseline, i.e.  
rest). Note that in the case of this design you might want to model  
the Stimulus and maybe even 'Anticipation' as separate EVs -  
otherwise these types of events might introduce additional non- 
modelled variance. You might find it useful to go through the 'model- 
based analysis' sections of the FSL course (see http:// 
www.fmrib.ox.ac.uk/fslcourse/ ).

Wrt the second question: The important thing to remember is that  
contrasts define linear combinations of fixed parameter estimates,  
i.e. are applied _after_ model fitting and as such are not part of  
the GLM fitting itself. In your case this means that if your design  
models the two conditions adequately then the PE for 'context' is the  
amplitude necessary to scale the EV1 ('context') in order to best  
fits into the date while simultaneously(!) fitting the parameter  
estimate for EV2. That is, even if you decide to use a contrast which  
only involves one of the two EVs it still uses the correct parameter  
estimate found when fitting the entire model simultaneously.  
Therefore, your "context > rest" question is answered by asking  
"where in the brain is PE1 > 0 (when fitting EV1 as part of the  
larger design which models all other signals)", i.e. you get a 1 0  
contrast provided that the entire design matrix models all(!) signals  
adequately. Note also that in your design the 'context' and 'no  
context' conditions give orthogonal EVs, so the PE for either one  
necessarily is unaffected by the other and PEs are not biased.
Hope this helps
cheers
Christian



On 3 Mar 2007, at 03:24, Ananth Narayanan wrote:

> Hi, thanks for the explanation. That helped me out a lot.
>
> What I described previously was essentially a block design task. We  
> have
> another paradigm which is jittered event related. Each event is  
> described
> below.
> Rest: 3,6,9 seconds (average of 6)
> Stimulus: 3 seconds
> Anticipation: 3 seconds
> Fedback: 3 seconds
>
> Feedback contains several outcomes, events that are modeled as Win,  
> NoWin,
> Loss and NoLoss. In addition, the paradigm we are trying to replicate
> compares each one of the four conditions to the "Rest" period. From  
> what
> you've described, their contrasts would be as follows
> 				Win	NoWin	Loss	NoLoss	Rest
> "Win" = Win-Rest		1				-1
> "NoWin" = NoWin-Rest			1			-1
> "Loss" = Loss-Rest				1		-1
> "NoLoss" = NoLoss-Rest					1	-1
>
>
> On top of that, there is an additional contrast, which is defined as
> ("Win" - "NoWin") - ("NoLoss" - "Loss"). For this contrast, I would  
> think
> the contrast would be defined as
> 			Win	NoWin	Loss	NoLoss	Rest
> 			1	-1	1	-1	0
> This is a much more complicated design. So will the Rest period  
> completely
> cancel out? Or is there a different model to it?
>
> *************************************
> Going back to the original contrasts, I am still confused on the  
> actual
> meaning of what you said. I took a look at the contrasts you  
> suggested.
> Suppose I consider the "Context vs Rest" and "NoContext vs. Rest"  
> contrasts,
> which would be
> 1 0  (context > 0)
> -1 0 (context < 0)
>
> My understanding is that it is comparing times when the "Context"  
> stimulus
> is on with times when the "Context" stimulus is not on. The latter  
> includes
> true rest period and the "NoContext" stimulus period. If there is
> significant activation in the "NoContext" stimulus period compared  
> to the
> true rest period, won't the signals get averaged out during the  
> subtraction
> causing a bias?

--
  Christian F. Beckmann
  Oxford University Centre for Functional
  Magnetic Resonance Imaging of the Brain,
  John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
  Email: [log in to unmask] - http://www.fmrib.ox.ac.uk/~beckmann/
  Phone: +44(0)1865 222551 Fax: +44(0)1865 222717