Inconsistency of bootstrap: The Grenander estimator

Abstract

In this paper, we investigate the (in)-consistency of different bootstrap methods for constructing confidence intervals in the class of estimators that converge at rate $n^{1/3}$. The Grenander estimator, the nonparametric maximum likelihood estimator of an unknown nonincreasing density function $f$ on $[0,\infty)$, is a prototypical example. We focus on this example and explore different approaches to constructing bootstrap confidence intervals for $f(t_0)$, where $t_0\in(0,\infty)$ is an interior point. We find that the bootstrap estimate, when generating bootstrap samples from the empirical distribution function $\mathbb{F}_n$ or its least concave majorant $\tilde{F}_n$, does not have any weak limit in probability. We provide a set of sufficient conditions for the consistency of any bootstrap method in this example and show that bootstrapping from a smoothed version of $\tilde{F}_n$ leads to strongly consistent estimators. The $m$ out of $n$ bootstrap method is also shown to be consistent while generating samples from $\mathbb{F}_n$ and $\tilde{F}_n$.Comment: Published in at http://dx.doi.org/10.1214/09-AOS777 the Annals of Statistics (http://www.imstat.org/aos/) by the Institute of Mathematical Statistics (http://www.imstat.org

A general approach to categorizing a continuous scale according to an ordinal outcome

“…Other modifications of the conventional bootstrap method, such as m out of n bootstrap and smooth bootstrap, have also been investigated in various estimation settings with cubic root convergence (Lee and Pun, 2006; Léger and MacGibbon, 2006; Sen et al, 2010; Sen and Xu, 2015, among others). These methods can be adapted to make inference about ${\overrightarrow{d}}_0$ under stronger assumptions.…”

“…Estimate the conditional density functions, f X|Y=l ( l = 1,2,..., L ), by the following smooth and symmetric kernel based estimators, $${\hat{f}}_{X\mid Y=l}\left(x\right)=\frac{1}{2n_l{h}_l}true[\sum _{k=1}^{n_l}Ktrue(\frac{x-{\tilde{X}}_{l,k}}{h_l}true)+\sum _{k=1}^{n_l}Ktrue(\frac{x+{\tilde{X}}_{l,k}-2{\widehat{\theta }}_l}{h_l}true)true],$$ where K( · ) is the Gaussian kernel function, h l is a smoothing parameter standing for bandwidth, ${\hat{\theta }}_l$ is the sample median of X given Y = l . We proceed bandwidth selection as follows: Following Sen et al's (2010) rule of thumb, we start with an initial bandwidth, $h_0=0.9A{n}^{-1∕n}$, with $A=\min \{s,IQR∕1.34\}$, where s and IQR are the sample standard deviation and inter-quartile of $\{{\tilde{X}}_{l,k},k=1,2,\dots ,n_l\}$.We evaluate a sequence of candidate bandwidth values in the neighborhood of h 0 , say h 0 –0.1, h 0 –0.05, h 0 , h 0 +.05, h 0 + 0.1, based on the integrated least square cross-validation criterion (Sheather, 2004), …”

Bootstrapping a change-point Cox model for survival data

“…However, when these bootstrap procedures are applied to the change-point model (1.2), they yield invalid confidence intervals (CIs) for ζ 0 (see Section 4). The failure of the usual bootstrap methods in non-standard problems has been documented in the literature; see Abrevaya and Huang (2005) and Sen, Banerjee and Woodroofe (2010) for situations giving rise to n −1/3 asymptotics; also see Bose and Chatterjee (2001) and Cheng and Huang (2010) for M -estimation problems. The performance of different bootstrap methods for the Cox model (1.2) has not been investigated in the literature.…”

Nonparametric estimation of time-to-event distribution based on recall data in observational studies

“…Sen et al (2010) have argued that the usual bootstrap is not be guaranteed to be consistent. The variances of (19) and (22) may be estimated through m out of n bootstrapping of Bickel et al (1997) with selection of m as in Bickel and Sakov (2008), so that consistency is ensured.…”