Dear Joe
> My question is why the temporal derivative does not appear in the
> design matrix with the following model (using SPM99): (simple A/B
> box-car design, 20 scans per condition)
>
> interscan interval = 2
> scans per session = 240
> number of trials: 2 (A, B)
> Stochastic Design = NO
> SOA = Fixed
> SOA for "A" = 40
> Time to first trial for "A" = 0
> SOA for "B" = 40
> Time to first trial for "B" = 20
> Parametric modulation = NONE
> Epoch, Fixed Box-car responses
> Convolve with hrf = NO
> Temporal derivative = YES
> Epoch lengths for both "A" and "B" = 20
> interactions among trials : NO
> user's specified regressors: 0
The derivatives are applied to the basis functions before convolution.
Strictly speaking the box-car has a derivative = 0 over its support and
therefore does not do anything. This is a bug or a feature depending
on your perspective. The derivative is generally used to accomodate
latency differentials but this relies on slowly varying continuous
basis functions. The derivative of a box-car that includes its
discontinuities would be a positive spike at the beginning and a
negative spike at the end. If you really wanted these, they could be
modelled as event-related responses.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|