Timothy Dang <[email protected]>:
I would think
bootstrap m=r(mean), cluster(Subject): sum choice
would be better, allowing for correlation within subject.
How do you model learning by subjects over the 12 periods?
Maybe
mean choice, over(time)
is better, ignoring heterogeneity in preferences and learning rates.
I guess it depends on the theoretical model...
But none of these approaches are appropriate for proportions that near
zero or one, since the true probability is constrained to the unit
interval. Better still might be
g w=1
svyset [pw=w]
svy: tab time choice, row ci
On Wed, Mar 19, 2008 at 12:40 PM, Timothy Dang <[email protected]> wrote:
> Hello statalisters-
>
> I need clarification on the -strata- option in -bootstrap-.
>
> The help documentation says: "strata(varlist) specifies the variables
> that identify strata. If this option is specified, bootstrap samples
> are taken independently within each stratum."
>
> I'm not entirely clear on what that means. Obviously, any bootstrap
> samples are taken independently (unless otherwise specified with
> something like -cluster-). Does the above mean that each replication
> has one sample from each stratum?
>
> I'm looking at some economic experimental data. Over 12 periods, a
> bunch of subjects made binary decisions. I want to estimate the
> overall odds of a positive choice, but obviously the choices of one
> subject over all those periods is not independent.
>
> So I'm essentially trying this:
>
> bootstrap m=r(mean), reps(100) level(90) strata(Subject): sum choice
>
> I'm trying to figure out if that's the correct way to cope with the
> inter-dependence of choices for each Subject.
>
> Thanks
>
> -Timothy
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
