Thank you for taking the time to read and answer my questions, and sorry
for any confusing wording.
I would not have thought of modeling post-event time points separately
if this hadn't been mentioned in mixed-design papers like Visscher's.
That being said, I am not sure how SPM5's hemodynamic lag is different
(or the same) as modeling post-event times explicitly. If it is indeed
the same as including event+1, event+2, event+3 etc as additional
regressors, that would be an easier solution than what I think I need to
My baseline blocks consist of an active baseline (odd-number detection).
Again based on papers using mixed design and modeling their (fixation
cross) baselines separately (see also Mitchell et al., Psychol Sci.
2007 Apr;18(4):292-7.), I was wondering why one would want to do that if
SPM does it implicitly.
Thanks a lot,
Michael T Rubens wrote:
> On Tue, Mar 24, 2009 at 12:52 PM, Esther Fujiwara <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
> Hi (sorry for any double posting),
> I have a few questions about analysing a mixed block/event-related
> design in
> the same model. If I understand Visscher et al. correctly (NeuroImage 19
> (2003) 1694–1708), blocks and events are modeled with different
> (gamma function vs. separate delta regressors for each time point
> the event, covering 14-20 s post-event).
> My design has three blocks (ABC) and 2 event-types (1,2), presented
> in each
> block, i.e., a 2x3 design like this: 1A,1B,1C,2A,2B,2C. Task blocks
> are 60
> seconds long, with 30s baseline blocks. Events (within task blocks) are
> mixed with null events.
> My questions:
> 1) If I want to model events and blocks in the same model using
> SPM5, do I
> need to use different functions for events and blocks? How/where do I do
> this in SPM5?
> The difference between the event and block (epoch) is the duration that
> you specify. For your events, specify the duration as 0. For the blocks
> specify the duration as the actual duration of the block.
> 2) How do I model the post-event time frames explicitly?
> Not sure I quite understand you here, but the convolution of your
> boxcar/stick with your hrf accounts for hemodynamic lag.
> 3) What is the difference between modeling a 'condition' and
> Again, not crystal on your question here. A regressor is one column in
> your design matrix. If you have multiple sessions of the same condition,
> you may choose to model them with separate regressors in order for your
> intercept term to properly account for baseline differences, then you
> can combine the resulting parameter estimates in later analyses.
> And do I need to model baseline blocks and/or null trials explicitly?
> What does your baseline block consist of? In short, if you do not model
> these, which you probably shouldn't, then neural activity during these
> time periods will essentially be accounted for by the intercept term
> (intrinsic baseline). Later if you do a contrast of  for the
> condition, it will subtract the intercept from your condition specific
> betas, along with the rest of the unaccounted variance.
> Research Associate
> Gazzaley Lab
> Department of Neurology
> University of California, San Francisco
Esther Fujiwara, Ph.D.
Department of Psychiatry
University of Alberta
3087 Research Transition Facility
Canada T6G 2V2
Phone: 780 492-6524
Fax: 780 492-6841