[VoxBo] How to run a FIR analysis

[Charan Ranganath]

Dear Dan, if I understand your question correctly, then here is my best 
answer:
Regardless of whether you have a sparse design or overlapping responses 
due to short ITIs, an FIR set can help you model the time course of the 
BOLD response uniquely associated with your trial type of interest. So, 
what you want to do is take your original covariates (say for condition 
1 2), and then modify-->FIR set (or whatever the gdesign widget calls 
it). If you have a 2s TR, then you want to model with at least 9 
covariates in the basis set (i.e., to model a 16s response). So, you 
should end up with the following:
cov1-time1 (modeling signal at t=0s)
cov1-time2 (t=2)
cov1-time3 (t=4)
cov1-time4 (t=6)
cov1-time5 (t=8)
cov1-time6 (t=10)
cov1-time7 (t=12)
cov1-time8 (t=14)
cov1-time9 (t=16)
[my notation probably does not correspond with VoxBo's, but hopefully 
you'll get the idea].

So you will get 9 betas for each condition. Plot 'em in a line graph, 
and voila, you have an estimated time course.

Note that it does NOT matter that your responses will overlap across 
trials. The basis set will model the unique BOLD response that best fits 
the entire set of trials. Therefore, overlapping signal from other trial 
types will be attributed to the covariates for those trials, with the 
rest accounted for by nuisance covariates or ending up in your error term.

I don't think you want to get only a single estimate for a given 
condition, because that defeats the purpose of an FIR set. If you really 
want a single estimate, you'll need to assume an HRF shape, and convolve 
your reference function with that HRF in order to generate the covariate 
to model your responses.

I hope you find this to be helpful. Cheers, CR

Dan Acheson wrote:
> I ran a study with a rapid event-related design using a random jitter of 
> the trial length (minimum 2 TRs, max 12 TRs), intending to use a FIR 
> analysis. I initially ran a FIR with 2 basis functions, but this 
> produced nonsensical results, and there were estimates for each FIR 
> basis function. Two questions: How do I run a FIR and is there a way of 
> getting a single estimate for a given condition, rather than an estimate 
> for each basis function specified?
> 

-- 
Charan Ranganath, Ph.D.
Associate Professor
Center for Neuroscience and Dept. of Psychology
University of California at Davis
1544 Newton Ct.
Davis, CA 95616

phone: 530-757-8750
fax: 530-757-8640

http://DynamicMemoryLab.org