Stratified Bootstrap question

Dear Tim,

Thank you very much for your information. I will try to play with S+ as
you suggested. At the same time, I would like to share our idea with you
about the stratified bootstrapping for my scenario. I am not sure if it is
correct. I am playing with it now.

We created a new dataset containing clinic and patient id within clinic,
then stratified boot() function was used to bootstrap
the newly-created dataset. Based on the indices of the bootstrap result,
since patient id is unique, we found the patient ids from the new dataset,
then found the corresponding dataset to fit a mixed model from the
original dataset using patient ids.

That sounds reasonable.  That is what the S+Resample library does
internally.

I am trying to run the program now, but it takes longer than what I
expected. 500 times takes more than 3 hours and it is still running. I
will see if this is working properly.

This may be normal.  Fitting mixed models is iterative, unlike
simple linear regression for which there is a closed-form solution.
So running many replications can take a while.

It might help if you specify starting values for the fixed-effects
coefficients.  Run the model for the original data, and extract
the fixed-effects coefficients.  Then specify those as starting
values; this could make the bootstrap replicates run faster.

Thank you very much for your input,
Qian



On 30 Mar 2005, Tim Hesterberg wrote:

Dear Qian,

You might try the S+Resample library, which has built-in support
for both sampling by subject and stratified sampling.

If you are a student, there is a free student version of S+.

See
www.insightful.com/downloads/libraries	(S+Resample)
www.insightful.com/Hesterberg/bootstrap	(has link to the student version)

For the missing values, consider the S+Missing library,
which offers multiple imputation.  With S+, do
	library(missing)

Tim Hesterberg

P.S.  The combination of sampling by subject and stratified sampling
was terribly messy to program.  If I'd known in advance how messy, I
never would have done it :-(  But it is done now.

Dear R users,

I have a question regarding stratified bootstrap question and how to implement
it using boot() in R's boot package.

My dataset is a longitudinal dataset (3 measurements per person at year
1, 4 and 7) composed of multiple clinic centers and multiple participants
within each clinic. It has missing values.

I want to do a bootstrap to find the standard errors and confidence
intervals for my variance components. My model is a mixed model with
random clinic and random participant within clinic.

I thought two methods to do bootstrap:
(1) bootstrap data; however, I have problem specifying the second
parameter for my statistic function, shall I use indices, weight or
frequency and how shall I relate to my dataset.
(2) bootstrap residuals; however, the dataset has multiple measurements
and missing values. I am wondering how to construct a new data frame
containing the residuals and fitted values.

Any ideas will be highly appreciated!
Sincerely yours,
Qian

========================================================
| Tim Hesterberg       Research Scientist              |
| timh at insightful.com  Insightful Corp.                |
| (206)802-2319        1700 Westlake Ave. N, Suite 500 |
| (206)283-8691 (fax)  Seattle, WA 98109-3044, U.S.A.  |
|                      www.insightful.com/Hesterberg   |
========================================================
Download the S+Resample library from www.insightful.com/downloads/libraries

***************************************
Qian An
Division of Biostatistics
University of Minnesota
(phone) 612-626-2263
(fax) 612-626-8892
Email: qiana at biostat.umn.edu
***************************************