The bootstrap - a review (Q1965952)

In this review the main ideas of bootstrap techniques are described. At first Efron's naive bootstrap is compared with the jackknife as a tool of bias and variance estimation for smooth statistics of i.i.d. samples. Then bootstrap approximations of the distribution of a statistic and bootstrap confidence intervals are considered. The Edgeworth expansion is used to demonstrate the second order efficiency of bootstrap procedures. Then adaptations of bootstrap to linear regression models, ARMA models and to estimation of prediction errors are described. The alternatives for naive bootstrap (smoothed, symmetrized, Bayesian and wild bootstrap methods) are mentioned. Bootstrap methods of bandwidth selection for kernel density estimation and nonparametric regression are discussed as well as bootstrap for censored data. Results of original simulation studies are presented to illustrate examples of bootstrap methods application. The bibliography contains 195 items.