Hi Todd,
The industry standard is Option #2. The correct analysis, however, depends
on the goal.
If you assume a cognitive or neural model, and you want to know which areas
in the brain have activity consistent with that model, then Option #1 is
correct. It has a straightforward physiological interpretation i.e.
decision-related neurons fire for the duration of the decision and RT is
your best estimate of that duration.
If you don't care about the neural or cognitive validity of the model, but
are interested in constructing a statistical model for predictive purposes,
then Option #2 is correct. It has more degrees of freedom and thus is more
flexible in fitting the data. Of course, though it makes a better
predictive model, the results are not physiologically interpretable because
changes in duration are arbitrarily being transformed into changes of intensity.
Hope this helps.
jack
On Mon, 1 Jun 2009 17:47:44 -0400, Todd Thompson <[log in to unmask]> wrote:
>Hi, all. I've been having a discussion on how best to include reaction
>time in my analyses, and I'd like to see if there's a consensus
>opinion on the "right" way to do it. My experiment is a simple flanker
>experiment with some tweaks, with reaction times ranging from 350ms
>to, say, 750 ms.
>
>Option 1: instead of modeling events as impulses, model them with a
>duration of whatever the RT was on that trial. Intuitively, this seems
>to have the advantage of improving the fit of your model to your
>actual data, and thus decreasing your residual and improving your
>stats.
>
>Option 2: model events as impulses, but also include a covariate that
>is the (possibly de-meaned) RT. It seems like the idea here is that
>the RT covariate will soak up some of the unexplained variance, and
>that it will absorb more on the longer duration trials than the
>shorter duration trials. The intuition is that then your EV of
>interest will then be more homogenous across trials.
>
>Option 3: both? (No one has actually suggested this one, but I thought
>I'd throw it out there.)
>
>I know this is a basic question, but I can't quite figure out if
>there's an "industry-standard" answer or not...
>
>Thanks!
>Todd
|