Implementation of resampling for reducing bias and creating a representative nation-wide data-base
by
Evguenia Jilinskaia
Pharmetrics, Inc., Watertown, Massachusetts, USA
Coauthors: Tom Marx, Stanley Norton, Trung Do (Pharmetrics, Inc., Watertown, Massachusetts, USA)

Creating a representative nation-wide Health Care data-base and some ways of reducing the bias in estimates are discussed. The method of multiple resampling and farther projection to the whole nation is proposed. The optimal sample sizes and rates of convergence for three different types of resampling - estimates are compared by simulation and on the national data-base: K-grouped jackknife, ßquare-root delete " jacknife and usual bootstrap. The resampling methodology can essentially reduce the length of confidence intervals for estimation of the mean values of variables of interest, of their order statistics (median, quartiles), and of more complicated test statistics. The results are presented for resampling techniques employing different sample sizes: small, medium and large. The algorithm of changing the sample sizes of the derived subsamples and at the same time increasing (or reducing) the number of bootstrap - iterations is implemented to non-normal cost data which is highly skewed. The comparison of proposed method with existing ones shows our method to be more reliable and giving more accurate estimates.

Date received: October 16, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cafr-59.