Jay <[email protected]> :
Assuming you had no missing data, how would you analyze this? I would
have thought some GMM or stacked approach... I am assuming errors are
correlated across models (one may sacrifice accuracy to improve RT or
vice versa). How many subjects do you lose if you use complete cases?
On Thu, Jun 4, 2009 at 2:56 PM, Verkuilen, Jay <[email protected]> wrote:
> I'm working on a model which is, well not to put too fine a point on it,
> complex:
>
> -Two dependent variables, one binary (accuracy) and the other
> well-approximated by the normal (log-reaction time);
> -Multilevel structure (120 observations per subject, about 300
> subjects);
> -Missing data that would delete entire subjects from the dataset, i.e.,
> level 2 missingess (but which is probably MAR).
>
> Fortunately there is a pretty strong theory about the relationship
> between accuracy and RT and the regressors.
>
> Any two of the three would be workable but the third is just a killer.
> Suggestions? Does it make sense to do MI first and then run -gllamm-?
> Anyone have any experience doing something like this?
>
> JV
>
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
