An overview of asymptotic properties of [L.sub.p] regression under general classes of error distributions.

From: Journal of the American Statistical Association | Date: June 1, 2005| Author: | Copyright information

We survey the asymptotic properties of regression [L.sub.p] estimators under general classes of error distributions. It is found that the asymptotic distributions of [L.sub.p] estimators depend crucially on p and the shape of the error distribution near the origin. A number of important features arise as a result, among which are (a) use of a small p may yield accelerated convergence rates for [L.sub.p] estimators under certain classes of error distributions; (b) for p

KEY WORDS: Convergence rate; Gaussian process; [L.sub.p] estimator; m out of n bootstrap; Regression.

1. INTRODUCTION

...

<2, 2p + [zeta]><2, 2p + [zeta]><0, for some [[zeta].sup.+], [[zeta].sup.-]><1 and [F'.sub.U](0)><1 and p><0 or><0 and [[gamma].sub.0]><1 and [zeta]><1 and [zeta]><1. Figure 7 summarizes the results, which show that the smallest MSE is again achieved at p around 1.5 in all cases. Larger values of p hamper the accuracy due to their sensitivity to outliers. We observe that the MSE decreases slightly when p decreases below 1, and the trend is unaffected by outliers. In all cases, [L.sub.1] estimates are the least accurate, reflecting their slow convergence rate, [n.sup.1/(2[zeta])] for [zeta]><1 but slower than [n.sup.1/2] when [zeta]><1/2. Setting p [member of] (1/2, 1] actually gives an optimal convergence rate in the sense that it is comparable to that of any asymptotically efficient estimator of [[beta].sub.0]. Thus there are circumstances under which use of a small p not only guards against outliers or heavy-tailed errors, but also provides a means of constructing estimators that yield optimal convergence rates, especially when [F.sub.U] is heavy-tailed or varies regularly at a fast rate at 0. Such ratewise optimality decidedly warrants an in-depth investigation into the asymptotic properties and practical implementation of [L.sub.p] estimation for small p. Our simulation study provides empirical evidence that both robustness (against outliers), and efficiency can be achieved simultaneously if the error distribution has heavy tails and a central shape that favors using a small p for faster convergence. Using a large p (><[r.sub.n]||[beta] - [[beta].sub.0]|| [less than or equal to] [2.sup.j]}. For fixed [eta], M><1 and q>

Related newspaper, magazine, and trade journal articles from HighBeam Research

(Including press releases, facts, information, and biographies)

Distinctive modulatory effects of five human auxiliary [beta]^sub 2^ subunit splice variants on L-type calcium channel gating
; ABSTRACT Sequence analysis of the human genome permitted cloning of five Ca^sup 2+^-channel [beta]^sub 2^ splice variants ([beta]^sub 2a^-[beta]^sub 2e^) that differed only in their proximal amino-termini. The functional consequences of such [beta...
The unbinding of ATP from F^sub 1^-ATPase
; ...are located in the [alpha]^sub 3^[beta]^sub 3^ hexamer. The [gamma]-shaft connecting [alpha]^sub 3^[beta]^sub 3^ to c^sub 10-14^ is an antiparallel...which penetrates the [alpha]^sub 3^[beta]^sub 3^ hexamer of alternating [alpha...
Secretory low molecular weight phospholipase A^sub 2^ plays important roles in cell elongation and shoot gravitropism in Arabidopsis
; ...weight phospholipase A^sub 2^ (AtsPLA^sub 2^[beta]) was isolated. Phenotype analyses...showed that overexpression of AtsPLA^sub 2^[beta] promotes cell elongation, resulting...interference-mediated silencing of AtsPLA^sub 2^[beta] expression retards cell elongation...
16,17-Dihydro Gibberellin A^sub 5^ Competitively Inhibits a Recombinant Arabidopsis GA 3[beta]-Hydroxylase Encoded by the GA4 Gene
; ...protein was able to 3[beta]-hydroxylate the 3-deoxy GAs, GA^sub 9^ and GA^sub 20^, to...1999), that GA^sub 1^, GA^sub 4^, and other 3[beta]-hydroxylated GAs are...17-dichloromethano-dihydro GA^sub 5^, to inhibit GA 3[beta]-hydroxylase activity...
In Vivo Evaluation of [beta]^sub 1^-AR Density
; ...design a high-affinity selective [beta]^sub 1^-adrenoceptor (AR) radioligand...in vivo SPECT imaging of cardiac [beta]^sub 1^-AR in humans. Although published...biopsies suggest that myocardial [beta]^sub 1^-AR density is reduced in patients...
Quantification of Cleaved [beta]^sub 2^-Microglobulin in Serum from Patients Undergoing Chronic Hemodialysis
; ...Patients on chronic hemodialysis are prone to develop amyloid deposits of misfolded β^sub 2^-microglobulin (β^sub 2^M) in osteoarticular tissues. β^sub 2^M with various deletions/truncations and chemical modifications has been found together...
Biodistribution and Pharmacokinetics of the [alpha]^sub v^[beta]^sub 3^-Selective Tracer ^sup 18^F-Galacto-RGD in Cancer Patients
; ...18^F-Galacto-RGD has been developed for PET of α^sub v^β^sub 3^ integrin expression, a receptor involved in, for example...that ^sup 18^F-Galacto-RGD allows visualization of α^sub v^β^sub 3^ expression in tumors with high contrast...
cis-Regulatory Elements for Mesophyll-Specific Gene Expression in the C^sub 4^ Plant Flaveria trinervia, the Promoter of the C^sub 4^ Phosphoenolpyruvate Carboxylase Gene(W)
; ...isoform of phosphoenolpyruvate carboxylase (ppcA1) of the C^sub 4^ dicot Flaveria trinervia. To elucidate and understand the anatomy of the C^sub 4^ ppcA1 promoter, detailed promoter/reporter gene studies were performed in the closely related C^sub 4^ species F. bidentis, revealing that the C^sub ...
Enzyme-linked immunosorbent assay for [beta]^sub 2^-glycoprotein I quantitation: the importance of variability in the plastic support
; [beta]^sub 2^-glycoprotein I ([beta]2GPI) is a poorly...monoclonal anti-human [beta]2GPI (Chemicon International...buffer (15.0 mmol/L Na^sub 2^CO^sub 3^, 34.9...4^, 1.5 mmol/L KH^sub 2^PO^sub 4^ [pH 7.2], 0.05...against purified [[beta]2GPI (Crystal Chem...
Transient solution of the M/M/[C.sub.1] queue with additional [C.sub.2] servers for longer queues and balking.(Report)
; ...to] n < [c.sub.1] (2) [P'.sub.n](t) = -([beta][lambda] + [c.sub.1][[mu].sub.1]) [P.sub.n](t...t), n = [c.sub.1] (3) [P'.sub.n](t) = -([beta][lambda] + [c.sub.1][[mu].sub...lambda]z(1 + 2[beta])) (1 - [[beta].sub.0a]) - ...