The weighted bootstrap (Q1346642): Difference between revisions

Let \(X_1, \dots, X_n\) be a sample from the distribution \(P\) and \(P_n= n^{-1} \sum_{i=1}^n \delta_{X_i}\) the empirical distribution. Denote by \(P^*_{n,m_n}= (m_n)^{-1} \sum_{i=1}^{m_n} \delta_{X^*_i}\) the bootstrap empirical distribution based on a bootstrap sample \(X^*_1, \dots, X^*_n\) from \(P_n\). \textit{B. Efron}'s [Ann. Stat. 7, 1-26 (1979; Zbl 0406.62024)] notice that \(P^*_{n,m_n} \overset {d}{=} \sum_{i=1}^n W_{i,n} \delta_{X_i}\), where \((m_n W_{1,n}, \dots, m_n W_{n,n})\) is a random vector with multinomial \((m_n, 1/n, \dots, 1/n)\) distribution, was a starting point of intensive investigations of the so called weighted bootstrap. The weights \(W_{i,n}\) may be choosen in different ways, e.g. they can be exchangeable random variables. This monograph is devoted to the generalized bootstrap. The authors answer two questions: 1) How well does the generalized bootstrap work? 2) What are the differences between all the different weighting schemes? The main text is devided into three chapters: I. Asymptotic theory for the generalized bootstrap of statistical differentiable functionals. II. How to choose the weights. III. Some special forms of the weighted bootstrap. The detailed proofs are put in separate three chapters (IV-- VI). Also some auxiliary results are placed in appendices.