Hi Andreas,
> do you have to demean the third column? I.e. wouldn't
> 5 0.1 0.66
> 30 0.1 0.33
> 50 0.1 1
> also be ok?
hmm, I think you are right that it would not make any difference for
the modulatory regressor. But I think it would make a difference for
the categorical regressor, i.e. you would lose sensitivity for the
first (categorical) regressor if the second regressor isn't orthogonal
to the first.
> Also - I'm not sure if it makes sense for reaction times - wouldn't
> that primarily shift the onset?
There are of course all sorts of ways you could model these data. You
could have one event type for the cue and another for the response,
the two being separated by the reaction time. You could have a single
regressor with onsets given by the cue and durations given by the
reaction time etc etc. I guess you will just have to chose the model
that makes it easiest to ask the questions you are interested in.
Good luck Jesper
> Cheers-
> Andreas
>
>
> ________________________________________
> Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag
> von Jesper Andersson [[log in to unmask]]
> Gesendet: Dienstag, 30. Juni 2009 16:40
> An: [log in to unmask]
> Betreff: Re: [FSL] first level ancova
>
> Dear Vitaly,
>
>> i have an analysis approach question:
>>
>> event -related design with behavior measure (e.g. reaction time)
>>
>> what do you think about analyzing the data on the single subject
>> level as an
>> ancova, where 1 EV has 1's for all events, zeros elsewhere, and
>> another EV
>> has the demeaned behavior measure at each event, zeros elsewhere
>>
>> versus
>>
>> binning events into fast events or slow events, separate EVs, and
>> then
>> performing a contrast at the first level to pass up.
>>
>> so option A is the discrete approach, option B is a binary approach.
>
> I would suggest a third approach. If you look at http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#stats
> you'll find a description of the "3 column format" for specifying
> event-related designs. The third column can be used to put in a
> parametric modulatory effect. You would then make one EV that model
> the average effect of these events and another one that models the
> modulatory effect.
>
> Let us say you have 3 events that ocurred a 5 30 and 50 seconds. Let
> us further say reaction times were 2 1 and 3 seconds respectively. For
> your modulatory EV you would then specify
>
> 5 0.1 0
> 30 0.1 -1
> 50 0.1 1
>
> where the second column gives you "eventy events" and the third column
> is the mean-corrected reaction times.
>
> You can then use the contrast [0 1] (assuming that the first column is
> the vanilla events) to take the dependence on reaction time to the 2nd
> level.
>
> Note that this assumes that the reaction times have a effect on the
> amplitude of the response, and not on the duration. In principle one
> could equally well assume the reverse, and unfortunately it is very
> hard to test since a changes in amplitude and duration will result in
> very similar models for event related designs.
>
> Good luck Jesper
|