bootstrap, strata

[tpon...@gmail.com]

Such an active group!

The help documentation for the -bootstrap- command says about -strata-
"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): ci choice

I'm trying to figure out if that's the correct way to cope with the
inter-dependence of choices for each Subject.

Any thoughts would be appreciated (even if only letting me know that
you're looking at the group messages).

-Timothy

[Wafa Hakim Orman]

Hey,

The help documentation for the -bootstrap- command says about -strata-
"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.

--> It's for survey data, if the survey used stratified sampling. Not necessary for experimental data.

 

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): ci choice

I'm trying to figure out if that's the correct way to cope with the
inter-dependence of choices for each Subject.

--> Thoughts from the others are welcome, but I think (& a number of experimental papers agree) that panel techniques are a good way to handle this.

http://www.stata.com/support/faqs/stat/xt_boot.html --> hopefully this should do what you're looking for.



Wafa.

--
"So be it."
--Kurt Vonnegut

[Timothy Dang]

Thanks, Wafa

I'm not sure why I got myself so confused, but rather than -strata-,
I'm using -cluster-. I think panel methods are overkill for this
application.

Since my data are binary choices and I want proportions, I'm adjusting
the confidence intervals with -bca- (bias-corrected accelerated
confidence intervals):

quietly bootstrap m=r(mean), bca reps(1000) level(90) cluster(session
Subject): ci MyFee if Type==1 & repeated==1
estat bootstrap, all

Hopefully that does the trick. The results look sensible. (Actually
the results are pretty insensitive to all of this.)

Thanks!

-Timothy

--
------------------------------
Timothy O'Neill Dang / Cretog8
623-587-0532
One monkey don't stop no show.