OK- so the problem is that the Anticipation and Outcome conditions are
correlated. Definitely need some expert advice here. If you have access to
Statistical Parametric Mapping (2007),  Friston et al. (Ed.), check out
Chapter 15: Efficient Experimental Design for fMRI by R. Henson.

It is true that a "rest" condition, which is not present in your design
anyway, is uninterpretable, making a contrast such as Outcome minus Rest (0
1 -1) uninformative. However, the contrast defining Outcome Only (0 1 0)
would be valid, as the model takes into account all of the conditions of the
experiment while the contrast does not include the activation during the
Rest condition. 

Regards,

Julie

Julie E. McEntee, M.A., C.C.R.P.
Senior Research Program Coordinator
Department of Psychiatry and Behavioral Sciences
Division of Psychiatric Neuroimaging
Johns Hopkins University School of Medicine
600 N. Wolfe St./ Phipps 300 (office: room 317)
Baltimore, MD 21287
Phone: 410-502-0468
Fax: 410-614-3676


-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of Andreas Pedroni
Sent: Tuesday, July 01, 2008 5:41 PM
To: [log in to unmask]
Subject: Re: [SPM] influence of non-jittered block-event

>
> Hi Julie,
> thanks a lot for your reply!
> I will give some further insights:
>
> Am 01.07.2008 um 22:22 schrieb Julie McEntee:
>
>> A few questions/points:
>>
>> 1. How did you design the experiment: block or event-related?
>>
> Mixed: It is a bit more complicated: The paradigm I'm talking about  
> is some kind of wheel of fortune game: 1. Starting with an epoch  
> with variable duration, where a choice for a bet is made. 2. After  
> 1-3 seconds an anticipation phase, where the wheel of fortune  
> spinns for 10 seconds. 3 The wheel of fortune stops and the outcome  
> is displayed for 3 seconds  4. Variable blank screen 8-10 s.
>
> I modelled: the choice(event) , anticipation phase(block 10s),  
> outcome(block 3s).
>
>> 2. Are you sure that you jittered the scan acquisition time? (it's my
>> understanding that TR is fixed, but this may be incorrect ...)
>>
> I meant by jittered, the onsets of stimuli are not synchroneous  
> with the onsets of the TR. Sorry, I did write that in a confusing way.
>
>> 3. If it is a block design, jittering any of the events within (or  
>> between)
>> a trial does not apply, as you cannot derive information regarding  
>> any
>> specific event during each trial (i.e., to do that you would need an
>> event-related design, in which case ...
>>
>> 4. When constructing a trial for an event-related design, keep in  
>> mind that
>> the BOLD response is, on average, about 20 s, with a latency of  
>> about 8 s.
>> Therefore, in order to obtain BOLD response information at a  
>> particular time
>> for a given duration during each trial (e.g., stimulus onset,  
>> response), the
>> design must allow time for any change in BOLD response to  
>> "recover", i.e.,
>> return to baseline. If the BOLD response is not allowed to recover,
>> questions regarding a specific event within each trial is not  
>> theoretically
>> possible; BOLD response may even disappear after high-pass  
>> filtering. The
>> purpose of jittering (typically applied to the inter-trial  
>> interval and/or
>> inter-stimulus interval, for example) is to de-convolve stimulus  
>> onset with
>> HRF and overcome the correlation between regressors, therefore  
>> increasing
>> the validity that the change in BOLD response is due to the  
>> experimental
>> events.
>>
>
>
>> To answer your question, if you  have taken the above into account  
>> in an
>> event-related paradigm, and created a complete model of the  
>> experiment at
>> the first (subject) level of analysis (i.e., the model must  
>> account for the
>> duration of every event during each trial, as opposed to only  
>> those that you
>> are "interested" in) you can ask questions of the model by defining
>> contrasts. For instance, I'll assume you modeled the experiment by  
>> defining
>> the conditions Anticipation, Outcome and Rest, in that order. The  
>> following
>> contrasts would produce:
>>
> I think I have taken this into account. But my big mistake was,  
> that I did not vary the duration of the anticipation phase. Like  
> this the regressor of the immediatly ensuing outcome phase is  
> always corelated to the regressor of the anticipation phase. As far  
> as I understand, a contrast of the outcome phase against the rest  
> ( e.g.: 0 1 0 below) can not be interpreted, because it is unclear,  
> how much of this "activity" is due to neural activity in the  
> anticipation phase and how much is due to the outcome phase...
>
>
>> name: Anticipation
>> weight: 1 0 0
>> all changes in BOLD response during Anticipation
>>
>> name: Outcome
>> weight: 0 1 0
>> all changes in BOLD response during Outcome
>>
>> name: Outcome minus Anticipation
>> weight: 1 -1 0
>> changes in activation during Outcome only, excluding BOLD response  
>> changes
>> present in both conditions
>>
>> name: Anticipation minus Outcome
>> weight: -1 1 0
>> changes in activation during Anticipation only, excluding BOLD  
>> response
>> changes present in both conditions
>
> I will try this. But I'm not sure, if this really can separate  
> activity between the two phases...
>
>> Creating masks using contrasts and applying them to other  
>> contrasts is
>> another approach that is used to isolate BOLD response- perhaps  
>> someone else
>> can contribute their expertise in this area, if applicable.
>>
> Would that be something like this?
> To detect activity unique to the outcome phase:
> Create a contrast for the anticipation against all other regressors  
> (your 1 0 0), then exclusively masking the outcome contrast (0 1 0) ?
>
>
>
>> I hope this is helpful-
>>
> Thanks again for your suggestions and nice ideas!
>
> Kind regards
>
> Andreas
>
>
>> Julie
>>
>> Julie E. McEntee, M.A., C.C.R.P.
>> Senior Research Program Coordinator
>> Department of Psychiatry and Behavioral Sciences
>> Division of Psychiatric Neuroimaging
>> Johns Hopkins University School of Medicine
>> 600 N. Wolfe St./ Phipps 300 (office: room 317)
>> Baltimore, MD 21287
>> Phone: 410-502-0468
>> Fax: 410-614-3676
>>
>> -----Original Message-----
>> From: SPM (Statistical Parametric Mapping)  
>> [mailto:[log in to unmask]] On
>> Behalf Of Andreas Pedroni
>> Sent: Tuesday, July 01, 2008 3:36 AM
>> To: [log in to unmask]
>> Subject: [SPM] influence of non-jittered block-event
>>
>> Dear SMP list!
>>
>> I have made an fmri experiment with an anticipation phase and an  
>> immediatly
>> following outcome phase. The scan aqcuision time is jittered in  
>> time, but
>> the durations of the two phases are (regrettably) fix (10 sec for
>> anticipation, 3 sec for outcome phase).
>> To me it makes sense, that the neural activity elicited in the  
>> anticipation
>> phase could bleed into the activity recorded in the outcome phase.  
>> Thus an
>> interpretation of the outcome phase is limited. But can the neural  
>> activity
>> in the outcome phase also influence the preceeding activity in the  
>> first
>> (anticipation) phase?
>>
>> Thanks for any suggestions!
>>
>> Cheers
>>
>> Andreas
>