Statistics Seminars: Accelerating the Effects of Bootstrap Iteration
27 April 2001 00:00 in CM221"Bootstrap techniques are empirical methods, based on resampling from a given dataset, for the assessment of errors and related quantities in problems of statistical estimation. The error of a bootstrap method may be reduced by iteration: using the bootstrap itself to estimate the error of the bootstrap estimator, and so recalibrate the calculation. In theory, successive iterations can be used to yield successive reductions in the order of the error. In practice, the iterated bootstrap requires a computationally expensive Monte Carlo simulation involving nested levels of bootstrap sampling from the sample data. In this talk we present a device, involving non-conventional weighted bootstrapping, which accelerates the theoretical effects of iteration. Its practical effectiveness is explored in an empirical study.
[This is joint work with Stephen Lee, The University of Hong Kong.]"
room CM103 at 2.15 - 3 pm
Contact email@example.com for more information