[VoxBo] "scale by trial count" question

[Ranjani Prabhakaran]

Hi Dan,

Thanks for your reply.  I just want to make sure I understand.  I'm wondering
whether you would consider the following case an appropriate use of the "scale
by trial count" option:

I've got 16 different trial types, and since I'm also coding incorrect trials -
that makes for 32 trial types in my condition function.  There are some cases
where I've got 12 correct trials (out of 14 total trials) of a particular trial
type (A) and 7 correct trials (out of 14 total trials) for a particular trial
type (B).  Since there are a different number of correct trials for each trial
type, would it make sense to "scale by trial count" to make sure that correct A
trials aren't weighted more when comparing A vs. B?

Thanks so much for your help!

Ranjani


Quoting Daniel Y Kimberg <kimberg at mail.med.upenn.edu>:

> Ranjani Prabhakaran wrote:
> > I remember someone had recently asked about how the "scale by trial count"
> > option in the G design window works and when you would want to use it.  I
> think
> > how it works was explained, but I'm still not clear on when you'd want to
> use
> > this option.
> > Would you want to use it if you've got a different number of trials in each
> > condition in your condition function (for example - if you're separately
> coding
> > correct and incorrect trials)?  If not, under what conditions would you
> want to
> > select this option (instead of going with the default, which is "do not
> > scale")?
>
> Ranjani, hi and sorry I've been slow to answer.  I've been hoping that
> someone else can come up with a good example, because I've never had
> to use this option personally.  But I'll give you a generic reply and
> maybe someone will elaborate.  For some reason I always have a hard
> time thinking up a good use for this feature.
>
> Basically, this offers a way of scaling the betas for different
> conditions differently.  So it's mostly useful if you have two or more
> conditions in addition to the baseline.  Imagine you have a baseline
> and two experimental conditions, A and B, with twice as many TRs of B
> as A, and you plan to contrast your two conditions.  If you don't
> scale by trial count, then your simple 1,-1 contrast will compare the
> betas on even footing.  If you do, then your betas for condition B
> will be doubled, and you'll get t=0 when the mean signal difference
> for condition B is half that of condition A.  This might make sense
> for an experiment on pacing, or something along those lines.
>
> dan
>