Dear Ian,


> hi all,
> we have an event-related design fMRI study in which the conditions
> of the spm analysis have been constructed retroactively based
> upon the nature of the subjects' recognition responses tested after
> scanning.
>
> when i inspect the design matrix for several subjects, spm appears to be
>
> deciding in advance not to sample certain time points on a session by
> session basis. that is, if i explore the design matrix and pick a
> particular
> condition, the lower left hand plot, which shows the synthetic hrf and
> time derivative, indicates that different peristimulus times will be
> sampled
> across the different sessions for that condition. for example, there
> may
> be red asterisks along the x-axis out to 20 seconds in scan one, whereas
>
> it only goes out to 15 seconds in scan two.

>
> am i interpreting this plot correctly? if so, is it that the program
> is omitting time points because they are somehow inefficient or
> statistically
> biased and does this mean that some of the data do not make into the
> Y.mad
> file regardless of the overall F threshold?
>

No, not really. The peristimulus time for a given event is simply defined
as the time that has passed since the last presentation of that particular
event type. Let us say that you have a single event type, and that you
present it every 20 seconds. Your maximum time from any sampling point to
the previous presentation will be 20 seconds, after that the PST will refer
to the next event and will be quite short. Hence, in that case the plot
would show you times out to 20 seconds.
No assume you have two event types, a fixed SOA of 5 seconds and a complete
randomisation of event types. Lets further assume that in one session there
is a stretch of 4 consecutive presentations of event type 2 (this being the
longest in that session). Then the longest time that will pass between two
conscutive presentations of event type 1 is 25 seconds, and the plot will
show times to up to 25 seconds. Lets further assume in the next session it
happens that there are 8 consecutive presentations of event type 2. Then
the longest time that will pass between two conscutive presentations of
event type 1 is 45 seconds, and the plot will show times to up to 45
seconds. SPM has used all points in both cases, but in one case the longest
PST is 25 seconds, and in the other 45 seconds.

You should also realise that in the modelling a given time point can be
e.g. 4 seconds after presentation of event type 1, 9 seconds after an event
of type 2 and 14 seconds after another event of type 2. It will contribute
to the final solution through all those parts that it simultaneously plays.
It is only in the plotting that one must decide what the PST is relative,
and then it is always relative the latest presentation of a given event
type.

Good luck Jesper




%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%