Robustness Comparisons of Some Classes of Location Parameter Estimators

Asymptotic biases and variances of M-, L- and R-estimators of a location parameter are compared under ε-contamination of the known error distribution F ₀ by an unknown (and possibly asymmetric) distribution. For each ε-contamination neighborhood of F ₀, the corresponding M-, L- and R-estimators which are asymptotically efficient at the least informative distribution are compared under asymmetric ε-contamination. Three scale-invariant versions of the M-estimator are studied: (i) one using the interquartile range as a preliminary estimator of scale: (ii) another using the median absolute deviation as a preliminary estimator of scale; and (iii) simultaneous M-estimation of location and scale by Huber's Proposal 2. A question considered for each case is: when are the maximal asymptotic biases and variances under asymmetric ε-contamination attained by unit point mass contamination at ∞? Numerical results for the case of the ε-contaminated normal distribution show that the L-estimators have generally better performance (for small to moderate values of ε) than all three of the scale-invariant M-estimators studied.     

On the robust estimation in poisson processes with periodic intensities

Under some regularity conditions, it is well known that the maximum likelihood estimator (MLE) is asymptotically normal and efficient. However, if the observation is contaminated, the MLE is not always an appropriate estimator. In this paper, we treat M-estimators and study their asymptotic behavior. By choosing estimation equations, robust M-estimators are presented for phase parameters.     

Robust estimation in the linear model with asymmetric error distributions

In the linear model X^{n × 1} = C^{n × p}θ^{p × 1} + E^{n × 1}, Huber's theory of robust estimation of the regression vector θ^{p × 1} is adapted for two models for the partially specified common distribution F of the i.i.d. components of the error vector E^{n × 1}. In the first model considered, the restriction of F to a set [−a₀, b₀] is a standard normal distribution contaminated, with probability , by an unknown distribution symmetric about 0. In the second model, the restriction of F to [−a₀, b₀] is completely specified (and perhaps asymmetrical). In both models, the distribution of F outside the set [−a₀, b₀] is completely unspecified. For both models, consistent and asymptotically normal M-estimators of θ^{p × 1} are constructed, under mild regularity conditions on the sequence of design matrices {C^{n × p}}. Also, in both models, M-estimators are found which minimize the maximal mean-squared error. The optimal M-estimators have influence curves which vanish off compact sets.     

Toward a unified theory for third R-order iterative methods for operators with unbounded second derivative

In this paper, we provide a semilocal convergence analysis for a family of Newton-like methods, which contains the best-known third-order iterative methods for solving a nonlinear equation F(x)=0 in Banach spaces. It is assumed that the operator F is twice Fréchet differentiable and F^″ satisfies a Lipschitz type condition but it is unbounded. By using majorant sequences, we provide sufficient convergence conditions to obtain cubic semilocal convergence. Results on existence and uniqueness of solutions, and error estimates are also given. Finally, a numerical example is provided.     

Robust estimation ofk-component univariate normal mixtures

The estimating equations derived from minimising aL ₂ distance between the empirical distribution function and the parametric distribution representing a mixture ofk normal distributions with possibly different means and/or different dispersion parameters are given explicitly. The equations are of theM estimator form in which the function is smooth, bounded and has bounded partial derivatives. As a consequence it is shown that there is a solution of the equations which is robust. In particular there exists a weakly continuous, Fréchet differentiable root and hence there is a consistent root of the equations which is asymptotically normal. These estimating equations offer a robust alternative to the maximum likelihood equations, which are known to yield nonrobust estimators.     

A local breakdown property of robust tests in linear regression

The breakdown slope, as a useful summary measure of local stability for estimators and test statistics, has been studied recently by He, Simpson, and Protnoy (1990, J. Amer. Statist. Assoc., 85). It is shown here that all regression estimates based on residuals alone in linear models have zero breakdown slopes in contamination neighborhoods, even though they can have breakdown points as high as one-half. The breakdown functions of tests based on the S-estimation are investigated. It is also shown that the Generalized M-estimators can have better local breakdown robustness. One way to obtain regression estimators with desirable local and global breakdown properties is discussed.     

Homotopic Fréchet distance between curves or, walking your dog in the woods in polynomial time

The Fréchet distance between two curves in the plane is the minimum length of a leash that allows a dog and its owner to walk along their respective curves, from one end to the other, without backtracking. We propose a natural extension of Fréchet distance to more general metric spaces, which requires the leash itself to move continuously over time. For example, for curves in the punctured plane, the leash cannot pass through or jump over the obstacles (“trees”). We describe a polynomial-time algorithm to compute the homotopic Fréchet distance between two given polygonal curves in the plane minus a given set of polygonal obstacles.     

Attainment and (sub)differentiability of the infimal convolution of a function and the square of the norm

Let X be a Banach space whose norm is simultaneously LUR and Gateaux (Fréchet) smooth. Under some assumptions, it is shown that the infimal convolution of a fairly general function on X and the square of the norm is generically strongly attained and hence is Gateaux (Fréchet) differentiable. This contains a result of S. Dutta on distance functions.     

On closed subspaces with Schauder bases in non-archimedean Fréchet spaces

The main purpose of this paper is to prove that a non-archimedean Fréchet space of countable type is normable (respectively nuclear; reflexive; a Montel space) if and only if any its closed subspace with a Schauder basis is normable (respectively nuclear; reflexive; a Montel space). It is also shown that any Schauder basis in a non-normable non-archimedean Fréchet space has a block basic sequence whose closed linear span is nuclear. It follows that any non-normable non-archimedean Fréchet space contains an infinite-dimensional nuclear closed subspace with a Schauder basis. Moreover, it is proved that a non-archimedean Fréchet space E with a Schauder basis contains an infinite-dimensional complemented nuclear closed subspace with a Schauder basis if and only if any Schauder basis in E has a subsequence whose closed linear span is nuclear.     

Fréchet differentiability for a damped sine-Gordon equation

We consider a damped sine-Gordon equation with a variable diffusion coefficient. The goal is to derive necessary conditions for the optimal set of parameters minimizing the objective function J. First, we show that the solution map is continuous under a weak assumption on the topology of the admissible set P. Then the solution map is shown to be weakly Gâteux differentiable on P, implying the Gâteux differentiability of the objective function. Finally we show the Fréchet differentiability of J. The optimal set of parameters is shown to satisfy a bang–bang control law.     

Metrization of Epi-Convergence: An Application to the Strong Consistency of M-Estimators

Strong consistency in the class of M-estimators is examined here as an application of epi-convergence, a functional convergence which is particularly suited for the study of convergence of the functions' minimizing values and arguments. Starting from a 1988 paper by J. Dupaova and R. Wets, which contains a thorough account of the relations between consistency and epi-convergence, a quantitative approach of the same topic is pursued here. Epi-convergence is compared with two definitions introduced in 1980 by one of the authors. The results are merged in order to define a distance between lower semicontinuous functions that is compatible with epi-convergence and bounds the distance between the minimizing arguments. These results applied to the statistical problem allow the definition of a bound of the distance between the estimator and the parameter.     

High Breakdown Estimation for Multiple Populations with Applications to Discriminant Analysis

We consider S-estimators of multivariate location and common dispersion matrix in multiple populations. Instead of averaging the robust estimates of the individual covariance matrices, as used by Todorov, Neykov and Neytchev (1990), the observations are pooled for estimating the common covariance more efficiently. Two such proposals are evaluated by a breakdown point analysis and Monte Carlo simulations. Their applications to the discriminant analysis are also considered.     

Local convergence of some iterative methods for generalized equations

We study generalized equations of the following form:

(render)

0f(x)+g(x)+F(x),

where f is Fréchet differentiable in a neighborhood of a solution x^* of (*) and g is Fréchet differentiable at x^* and where F is a set-valued map acting in Banach spaces. We prove the existence of a sequence (x_k) satisfying

which is super-linearly convergent to a solution of (*). We also present other versions of this iterative procedure that have superlinear and quadratic convergence, respectively.     

Asymptotic distribution theory for general statistical functionals

Summary In this paper an asymptotic theory is developed for a general statistical functional which includesL-estimators,M-estimators, andR-estimators as special cases. It is shown that a proof of the asymptotic normality ofL-estimators,M-estimators, andR-estimators can be based on one important fact, namely, that the inverse c.d.f. is compactly differentiable.     

Tukey's linear sensitivity and order statistics

Tukey (1965,Proc. Nat. Acad. Sci. U.S.A.,53, 127–134) introducedlinear sensitivity as a measure of informativeness in a collection of order statistics. Here we study its general properties and discuss how it is related to the best linear unbiased estimator, Fisher information measure, and asymptotic relative efficiency. Also, we obtain explicit and asymptotic expressions for the linear sensitivity of a collection of consecutive order statistics from a location or from a scale family, and discuss its role in the comparison ofL-estimators. We conclude our discussion with examples from uniform, exponential and normal populations.     

Every non-normable non-archimedean Köthe space has a quotient without the bounded approximation property

It is proved that any non-archimedean non-normable Fréchet space with a Schauder basis and a continuous norm has a quotient without the bounded approximation property. It follows that any infinite-dimensional non-archimedean Fréchet space, which is not isomorphic to any of the following spaces: , has a quotient without a Schauder basis. Clearly, any quotient of c₀ and has a Schauder basis. It is shown a similar result for and     

A universal differentiability set in Banach spaces with separable dual

We show that any non-zero Banach space with a separable dual contains a totally disconnected, closed and bounded subset S of Hausdorff dimension 1 such that every Lipschitz function on the space is Fréchet differentiable somewhere in S.     

A Chebyshev Set and its Distance Function

We prove that in a Banach space X with rotund dual X^* a Chebyshev set C is convex iff the distance function d_C is regular on X\C iff d_C admits the strict and Gâteaux derivatives on X\C which are determined by the subdifferential ∂x−

Full-size image

x" height="11" width="10"> for each xX\C and

Full-size image

x" height="11" width="10">P_C(x)\{cC:x−c=d_C(x)}. If X is a reflexive Banach space with smooth and Kadec norm then C is convex iff it is weakly closed iff P_C is continuous. If the norms of X and X^* are Fréchet differentiable then C is convex iff d_C is Fréchet differentiable on X\C. If also X has a uniformly Gâteaux differentiable norm then C is convex iff the Gâteaux (Fréchet) subdifferential ∂⁻d_C(x) (∂_Fd_C(x)) is nonempty on X\C.     

Differentiability of C0-semigroups with respect to parameters and its application

In this paper we discuss some continuous dependency between a C₀-semigroup and its parameter. We first get the following results: if the infinitesimal generator of a C₀-semigroup is continuous in the generalized sense, strongly (or weakly) Fréchet continuously differentiable, or strongly (or weakly) Gâteaux continuously differentiable with respect to the parameter, respectively, so is the corresponding C₀-semigroup. We then obtain corresponding expressions for the differential operators. Finally, the obtained results are applied to a mixed initial-boundary value problem of a semi-linear parabolic equation, and it is proved that the solution of the equation is Fréchet continuously differentiable with respect to the leading variable coefficient of the equation.