Hi,
On 1 Sep 2006, at 09:42, Isabell Wartenburger wrote:
> Dear experts,
> We have applied one long run and are interested in learning related
> changes
> within this run. There are 3 conditions, events of up to 7.5 seconds,
> depending on the reaction time of the subject. The stimulus
> disappears when
> the button is pressed, so trial duration = reaction time.
>
> So we start with 3 event files (EVs), 3 column format (start,
> duration,
> weighting) for the 3 conditions e.g.
>
> start - duration (=RT) - weighting
> 2 2.3 1
> 14 2.1 1
> 30 4.1 1
> 42 1.3 1
> ...
> 756 2.6 1
>
> RESULT: EV1 (condition1), EV2 (condition2), EV3 (condition3)
>
> (1) to remove reaction time effects I included an additional EV,
> which is
> weighted with reaction times. this EV includes ALL trials and is made
> orthogonal to the other EVs.
>
> start - duration (=RT) - weighting (=RT)
> 2 2.3 2.3
> 6 1.3 1.3
> 14 2.1 2.1
> 22 3.2 3.2
> 30 4.1 4.1
> 49 2.7 2.7
> 53 4.5 4.5
> ...
> 800 2.1 2.1
>
> this EV is modeled but not included in the contrasts (like a
> regressor of no
> interest).
>
> RESULT: EV4 (all trials weighted by reaction times, orthogonal to
> EV1, EV2,
> and EV3)
>
> Is this correct?
It depends on exactly what you want here. If you want to just soak up
variability proportional to reaction time left over after modelling
the mean condition effect then this is fine, yes.
> (2) to account for learning effects over the run I included three
> additional
> EVs, where the weighting is a falling number (= reduction of brain
> activation over time as a result of learning).
>
> start - duration (RT) - weighting (falling number)
> 2 2.3 68
> 14 2.1 67
> 30 4.1 66
> 42 1.3 65
> ...
> 756 2.6 1
>
>
> RESULT:
> EV5 (condition1 weighted by a falling number)
> EV6 (condition2 weighted by a falling number)
> EV7 (condition3 weighted by a falling number).
>
>
> Contrast:
> Effect of Learning (????)
> e.g. learning in condition 1 (EV5 vs. EV1)
> [-1 0 0 0 1 0 0]
>
> This gives highly significant nice results, but I guess they don’t
> present
> the effect of learning. In fact these blobs are identical to negative
> correlations of EV1 alone ([1 0 0 0 0 0 0]). (Independent on
> whether the
> weighting number goes from 68 to 1 or from +34 to –34).
>
> Is it wrong to try to account for intra-run (learning related)
> changes via
> weighting?
> Is it generally better to compare the first third of trials of
> condition 1
> with the last third of trials of condition 1?
Yes, I think this is wrong, but nearly correct. If I understand you
correctly then to test the "falling" effect, you don't want to
contrast this against the mean effect EV1 (for one thing the relative
scaling of EVs 1 and 5 is not comparable) - instead you just want to
orthogonalise EV5 wrt EV1 (ie remove the mean response) and then just
test a contrast of +1 or -1 on EV5 on its own.
Hope this makes sense?
Cheers, Steve.
>
> I was unfortunately not able to find an answer in the manual or the
> mailing
> list.
>
> Thank you so much for your help!!!
> Best, isabell.
------------------------------------------------------------------------
---
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
------------------------------------------------------------------------
---
|