Lectures on some aspects of the bootstrap

DOI10.1007/BFB0092619MaRDI QIDQ4373859zbMATH OpenFDO

Publication date 25 January 1998

Published in Lecture Notes in Mathematics (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1007/BFB0092619

Nonparametric statistical resampling methods (62G09) Asymptotic properties of nonparametric inference (62G20) Order statistics; empirical distribution functions (62G30)

Review text

This is a nice lecture on the first-order consistency of the bootstrap \((b.)\) in the simple but important case of the mean. Chapter 1 is devoted to the one-dimensional case (it is simple to derive general finite-dimensional settings), while in Chapter 2 the \(b.\) of empirical processes, that is the \(b.\) of the mean in infinite dimensions, is considered. In Chapter 1, Efron's \(b.\) is investigated. Let \(X:= \{X_i\}^\infty_{i=1}\) be i.i.d. r.v., \(\overline {X_n}:= \sum_{i =1}^n X_i/n\). The \(b.\) variables are defined by \[ X^*_{n,i}: =\sum^n_{j=1} X_j \cdot I_{U_i} \in (( j-1)/n,j/n], \quad i=1, \dots, n,\;n\geq 1, \] where \(\{U_i\}\) are i.i.d. uniform on \([0,1]\), independent of \(X\). Denote by \({\mathcal L}^*\) the conditional law given the sample \(X\). In Section 1.1, the convergence \[ {\mathcal L}^* \left({1\over \sqrt n} \sum^n_{i=1} (X^*_{n,i} -\overline X_n) \right) @>d>> N(0,\sigma^2)\tag{1} \] a.s. or in probability is shown, the same for the studentized version of (1), and the limits of validity of (1) are explored. A more general, `exchangeable \(b.\)', is considered in Section 1.2. Section 1.3 is devoted to the mixing observations. The next section is devoted to the \(b.\) of \(U\)-statistics. In this case the regular Efron's \(b.\) doesn't work, and the procedure is adapted so as to mimic the main features of the original statistics. In Section 1.5 the `\(m\) out of \(n\) \(b.\) without replacement' is presented, that is the \(b.\) sample is obtained by resampling \(m_n\) data from \(X_1, \dots, X_n\) without replacement, and with \(m_n/n \to\infty\). This procedure is consistent in great generality. In Chapter 2 it is proved that the \(clt\) for empirical processes, in the Vapnik-Červonenkis-Dudley general setting, can always be bootstrapped. This automatically gives the \(b.\) of many limit theorems. It is also shown that for a more restricted but still quite general set of classes of functions, basically any sensible model based \(b.\) works as well. Some applications (the \(b.\) of the median and of \(M\)-estimators) are considered. These lectures are almost self-contained, all the auxiliary inequalities and background from empirical process theory are given with proofs. For the entire collection see [Zbl 0871.00040].

Recommendations

Cited in

(15)

This page was built for publication: Lectures on some aspects of the bootstrap

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4373859)