Yes, there is big time individual heterogeneity.
Separately there are two sets of equations with the same regressors and individual heterogeneity for each. The model is not recursive however as it is not obvious that RT preceeds accuracy or vice versa. I need to pin my colleague down on exactly what her hypotheses are.
For model building I think I will deal with the missing data by running MI and simply averaging these datasets. Then after getting some candidates worked out go back and do MI for real. That knocks the missing data problem out for the moment. The notion of running -gllamm- on a bunch of MI replications is so not appealing....
-----Original Message-----
From: "Austin Nichols" <[email protected]>
To: [email protected]
Sent: 6/4/2009 6:15 PM
Subject: Re: st: suggestions on a model
Jay <[email protected]>:
Do you allow individual heterogeneity in the RT/accuracy tradeoff in
response to a changing covariate? I am imagining a subject pressing a
button in a multiple choice exam with very easy questions while being
fed alcohol intraveneously, so their RT and accuracy are declining
over time, but people will react differently, some very slowly and
deliberately getting right answers and some quickly getting wrong
answers (anyone else remember the WKRP episode?)
Can you describe the output for complete case subsample (150) and the
whole sample of 300 excluding the problem covariate? I think I don't
understand your model, and that comparison would help...
On Thu, Jun 4, 2009 at 5:07 PM, Verkuilen, Jay <[email protected]> wrote:
> Austin Nichols wrote:
> 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?
>
> Yes, there is an RT/accuracy tradeoff and also a substantial amount of
> individual heterogeneity. The model would be a path model, essentially,
> as all regressors are observed. They are either tests of cognitive
> function, perception, etc., or age. There are also variables that change
> within subject depending on the stimuli they received. Complete data...
> 150 subjects out of 300, or thereabouts. Missingness means that an
> entire subject has to be excluded because the only missing data are for
> tests of cognitive function.
>
> As I said, any two of the three problems in the model are manageable (if
> complex).
>
> 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/
*
* 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/
