Hi Andy,
A few other points that might be helpful:
1. There's really no difference between block and event-related designs, i.e. all designs are block 
designs.  Event-related designs just have really short blocks.  The point to take from this is that 
it's always more accurate to model the duration of each event rather than use a 0-duration 
impulse function.

2. It is possible that when you use impulse functions you don't get a rank deficient matrix but 
when you use finite duration events you do.  If this happens you shouldn't use the impulse 
function model.  Instead you should change your design such that the finite duration model is not 
rank deficient.  

3. Practically speaking, I've found that it's important to try to make the distribution of inter trial 
intervals as uniform as possible, to make the range of values as wide as possible, and to make the 
values continuous rather than multiples of TR.  
cheers,

jack

>>>> [log in to unmask] 08/20/05 1:17 AM >>>
>Hi Andrew,
>
>On Fri, 19 Aug 2005, Andrew Kozel wrote:
>
>> I am just starting to work with FSL and would appreciate some help with
>> the following issue with the model design in FEAT.
>> 
>> In another analysis package, events were modeled as onsets at a point in
>> time and a duration of zero.  As long as there was appropriate jitter in
>> these event designs, they appeared to run without problem. Reading
>> through the documentation, rank deficient designs or co-linearity were
>> discussed primarily with block designs and not event-related designs.
>> 
>> For FILM, looking at the documentation and archived emails, it appears
>> that in the 3 column method of building the design - events are
>> designated by onset, duration, and value of input during the period.  
>> The duration in these event designs that I have seen are greater than
>> zero.  My interest is in modeling BOLD signal correlates of
>> psychological processes.
>> 
>> In FILM, can events be modeled with a duration of zero (i.e. designated
>> as a point in time)?
>
>Yes - the initial temporal resolution of the model (before it is sampled
>at the TR resolution) is 0.05s, but if you set the duration of your events
>to be <0.05 (eg 0) it will make sure that you get at least one spike in
>this high-res model - hence all will be fine.
>
>> Could you then model all events in a paradigm which has no rest or
>> baseline conditions?
>
>Generally, for single-event designs yes, for block designs, no.
>
>> If the duration is greater than zero, how much baseline must be in the
>> model to prevent a rank deficient design?  As an example, for an event
>> that you designate as lasting 1 TR (e.g. 2 sec), you would need 2 TRs
>> (e.g. 4 sec) per event * once again with no baseline condition.
>> 
>> If you designate the duration of the event as greater than zero in FILM,
>> is the chosen haemodynamic response function convolved with the event as
>> if it were a "short" block?  Thus you must have "blocks" of a baseline
>> condition that are not designated as EVs?
>> 
>> Any help or direction for further reading would be greatly appreciated.
>
>Not sure exactly what you're asking here. All events, even "0" events, are 
>first created as a block (as described above, "0" events will be given a 
>0.05s duration) and then convolved with the HRF before sampling at the TR 
>resolution. Does this help answer the question?
>
>Cheers, Steve.
>
>-- 
> Stephen M. Smith  DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726  (fax 222717)
>
> [log in to unmask]  http://www.fmrib.ox.ac.uk/~steve
>===========================================================
=============