Sorry not to be clear. The standard approach in FSL does not account for the amplitude bias described by Vince's paper - it assumes that the td does not affect the amplitude.
Vince does not propose performing an F-test across the evs, but to test a nonlinear contrast of the two pes, sign(b1)*sqrt(b1^2+b2^2), that reflects the overall amplitude.
The core FEAT routines are designed for linear contrasts, so you cannot use them to generate the contrasts and perform the stats. The nonlinear contrasts could be easily generated with fslmaths, however.
Eugene
I'm sorry for this if it's obvious to everyone else, but I'm quite
confused now.
Vince -- reading your response makes it sound like you *don't* want to
test just the main EVs at the higher level, because you lose the
information that your temporal derivative was explaining. But you
started your reply to Eugene with "that's right", and I read Eugene's
response as saying that you *do* just test the main EVs at the higher
level.
What did I miss?
Thanks, and again, sorry for the confusion,
Todd
On Tue, Mar 10, 2009 at 3:51 PM, Vince Calhoun <[log in to unmask]> wrote:
> that's right...the gist of our paper is that you can compute nonlinear
> contrast at the first level which is a combination of both the main EV and
> it's derivative to determine an estimate of the fitted amplitude. similarly,
> henson computed the ratio of the two to determine an estimate of the
> latency...then these aspects (which are confounded if you only test the main
> EV) are decoupled and can be tested on a group of subjects at a higher
> level.
>
> VDC
>
> ________________________________
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> Of Eugene Duff
> Sent: Tuesday, March 10, 2009 1:18 PM
> To: [log in to unmask]
> Subject: Re: [FSL] compare models
>
> No, usually contrasts defined only using the main EV(s) are assessed in
> higher level analyses. In the expected situation, the temporal derivatives
> will temporally shift the response and not contribute to the amplitude of
> the response.
>
> Cheers,
>
> Eugene
>
> 2009/3/10 Todd Thompson <[log in to unmask]>
>>
>> I haven't yet fully digested the Calhoun 2004 Neuroimage paper which
>> talks about using temporal derivatives to improve your model fit, but
>> if my skimming is accurate, the gist seems to be that you want to run
>> an f-test between the EV and the temporal derivative of the EV, and
>> use the result of that f-test as the input to your second level model,
>> rather than using the parameter estimate of the EV.
>>
>> Is that the general approach FSL users take? Is that easily doable in
>> FSL? Or have I totally mis-skimmed the paper?
>>
>> Thanks,
>> Todd
>>
>>
>> On Tue, Mar 10, 2009 at 4:26 AM, Steve Smith <[log in to unmask]> wrote:
>> > Hi,
>> >
>> > This makes sense, yes. If you are happy that the temporal derivative is
>> > accounting for lag changes then you could simply compare models by look
>> > at
>> > the zstat on the EV of interest for each model - or as you say look at
>> > second-level stats for further confirmation.
>> >
>> > Cheers.
>> >
>> >
>> > On 9 Mar 2009, at 15:30, Vishwadeep Ahluwalia wrote:
>> >
>> >> Thanks Steve! once i run the 1st level analysis using different models
>> >> , i
>> >> was planning on
>> >> comparing the models by comparing the PE values extracted from an ROI
>> >> after running the
>> >> group analysis. Is this a valid approach? or is there a better way to
>> >> do
>> >> this. Also, i'm
>> >> slightly confused about whether i should be using a temporal derivative
>> >> at
>> >> the 1st level if i
>> >> want to compare various models. Please correct me if i'm wrong:
>> >> Temporal
>> >> derivative
>> >> would absorb any differences in Onset latencies and so the model fit
>> >> (and
>> >> hence PE)would
>> >> then be dependent only on height and shape of the response. So would
>> >> you
>> >> recommend the
>> >> use of temp. derivatives considering i'm comparing models?
>> >> Thanks!
>> >> -Vishwadeep Ahluwalia
>> >>
>> >
>> >
>> >
>> > ---------------------------------------------------------------------------
>> > 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
>> >
>> > ---------------------------------------------------------------------------
>> >
>
>
>
> --
>
> Eugene Duff
>
> FMRIB Centre,
> University of Oxford
> John Radcliffe Hospital, Headington OX3 9DU Oxford UK
>
> Ph: +44 (0) 1865 222 739 Fax: +44 (0) 1865 222 717
>
> --
>