Dear Sirs:
I have a panel data set consisting of 3600 subjects each of whom was
asked to rate 11 health states. The health states were selected
randomly from a pool of 41 states. Ratings were converted to a scale
ranging from -1 to 1. I am interested in using a random-effects model
to estimate the health state ratings from a set of regressors. I would
like to know how nonparametric bootstrapping could be used to derive
confidence limits for the parameters. I am very familiar with
bootstrapping in other contexts, but have never used it with panel
data. Though there appear to be a number of articles discussing the
application of the bootstrap to time series (including Efron and
Tibshirani), I have seen none that discuss use of nonparametric
bootstrapping with panel data.
My question is as follows: Would I need to use block (or clustered)
bootstrapping with my data? I am treating subject as a random effect.
If grouped bootstrapping was called for, then I presume that one would
use Stata's cluster option as follows: bs "..." "...", reps(...)
cluster(subject).
Any suggestions would be greatly appreciated.
Best Regards,
James Shaw
Graduate Research Associate
College of Pharmacy
The University of Arizona
[email protected]
begin:vcard
n:Shaw;James
tel;cell:520-631-3158
tel;fax:520-626-3386
tel;home:520-731-3337
tel;work:520-626-4452
x-mozilla-html:FALSE
org:Center for Health Outcomes and PharmacoEconomic Research;The University of Arizona College of Pharmacy
adr:;;P.O. Box 210207;Tucson;AZ;85721;USA
version:2.1
email;internet:[email protected]
title:Research Associate
fn:James W. Shaw
end:vcard
