Re: st: Bootstrap strata

[Laura Gibbons <gibbonsl@u.washington.edu>]

Yes, in my example it's quite tricky, because the reason one person has 
more values than another probably is informative.  We are trying to 
capture this in the predictive model, but of course will not do so 
completely.

This is one of several modeling approaches we will be comparing, all with 
deficiencies of one sort or another.

thanks,

Laura


On Tue, 4 Jun 2013, Steve Samuels wrote:

> I should also point out that your procedure assumes that the number
> of observations on an individual and, perhaps, the order of individual
> observations are uninformative about the study outcomes ("perhaps",
> because you might already have included sequence information among
> the predictors).
>
> S.
>
> Laura:
>
>
> The coefficients reported after -bootstrap- will be the coefficients for
> from -regress- on the full sample. You want the mean  of the coefficients
> from the -bootstrap- samples. You can get these with:
>
> . estat bootstrap, all
> . matrix list e(b_bs)
>
>
> Throwing out data can be inefficient. As you have nested observations,
> you might try -xtmixed- with robust standard errors and compare the
> results.
>
> Steve
>
> On Jun 4, 2013, at 12:53 PM, Laura Gibbons wrote:
>
> > In case someone else has this problem in the future, here is what I found out:
> >
> > By default, the -strata- option for -bootstrap- would perform
> > sampling with replacement within each strata; but the default number
> > of observations is the number of observations present in the dataset.
> >
> > If a person has 7 observations, the sample will consist of
> > 7 observations for that person; the difference with the original
> > dataset is that it will be a sample with replacement, i.e., repeated
> > observations might be present.
> >
> > For my purposes, since I want my samples per individual have size 1, I need to specify that as follows:
> >
> > bootstrap, strata(id) size(1): regress...
>
> On Fri, 31 May 2013, Laura Gibbons wrote:
>
> > I have data with 1 to 7 observations/person.  I want to randomly sample one observation/person in some sort of a bootstrapping process to get an appropriate error term for, say, x.
> >
> > I can set this up myself using a loop around the -bsample- command [bsample 1, strata(id)], posting the estimated coefficient for x each sample and taking the std dev of the mean of x for my 1000 times through the loop.
> >
> > Is there a way to do this directly in -bootstrap-?  Here is what the manual says about the strata option there:
> >
> >  strata(varlist) specifies the variables that identify strata.  If this
> >      option is specified, bootstrap samples are taken independently within
> >      each stratum.
> >
> > It is not clear to me what that means.  Does it mean it will sample once from each stratum?  I think not, based on my results, but I honestly can't tell.
> >
> > Many thanks,
> >
> > Laura
> > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > Laura E. Gibbons, PhD
> > General Internal Medicine, University of Washington
> > Box 359780, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104
> > phone: 206-744-1842, fax: 206-744-9917,
> > Office address: 401 Broadway, Suite 5122.6
> > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/faqs/resources/statalist-faq/
> > *   http://www.ats.ucla.edu/stat/stata/
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Laura E. Gibbons, PhD
> General Internal Medicine, University of Washington
> Box 359780, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104
> phone: 206-744-1842, fax: 206-744-9917,
> Office address: 401 Broadway, Suite 5122.6
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Laura E. Gibbons, PhD
General Internal Medicine, University of Washington
Box 359780, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104
phone: 206-744-1842, fax: 206-744-9917,
Office address: 401 Broadway, Suite 5122.6
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/