On Huber's contaminated model

Huber's contaminated model is a basic model for data with outliers. This paper aims at addressing several fundamental problems about this model. We first study its identifiability properties. Several theorems are presented to determine whether the model is identifiable for various situations. Based on these results, we discuss the problem of estimating the parameters with observations drawn from Huber's contaminated model. A definition of estimation consistency is introduced to handle the general case where the model may be unidentifiable. This consistency is a strong robustness property. After showing that existing estimators cannot be consistent in this sense, we propose a new estimator that possesses the consistency property under mild conditions. Its adaptive version, which can simultaneously possess this consistency property and optimal asymptotic efficiency, is also provided. Numerical examples show that our estimators have better overall performance than existing estimators no matter how many outliers in the data.     

Change-of-variance sensitivities in regression analysis

Summary The change-of-variance function is defined for estimators of regression coefficients. Both an unstandardized and a standardized form of the change-of-variance sensitivity are introduced, and their relation with the corresponding gross-error-sensitivities is investigated. The problems of optimal robustness lead to the Hampel-Krasker and the Krasker-Welsch estimators. At the same time, also the scale parameter has to be estimated robustly. By means of the change-of-variance sensitivity, optimal robust redescending scale estimators are constructed.     

A new infinitesimal approach to robust estimation

Summary The infinitesimal robustness of the asymptotic variance of location M-estimators is investigated by means of the change-of-variance curve (CVC), which bears some resemblance to the influence curve (IC). It is proved that this CVC leads to a more stringent robustness property than the IC and that the Huber estimators are still optimal in this new sense.     

Influence curves of general statistics

The influence curve (JC) of a Fisher-consistent functional was introduced by F. Hampel and plays a central role in the search for robust estimators. An extension of this notion to non-Fisher-consistent functionals is proposed in order to investigate the infinitesimal robustness of more general statistics, e.g. those used in hypothesis testing. This new definition inherits many useful properties, including some on asymptotic efficiency. Functionals in two variables, arising from two-sample statistics, are treated too. Connections with Hodges-Lehmann shift estimators are discovered. One- and two-sample rank statistics illustrate the theory.     

On the dispersion matrix of variance estimators of the sample mean in the analysis of simulation output

A classical problem of stochastic simulation is how to estimate the variance of point estimators, the prototype problem being the sample mean from a steady-state autocorrelated process. A variety of estimators for the variance of the sample mean have been proposed, all designed to provide robustness to violations of assumptions, small variance, and reasonable computing requirements. Evaluation and comparison of such estimators depend on the ability to calculate their variances.A numerical approach is developed here to calculate the dispersion matrix of a set of estimators expressible as quadratic forms of the data. The approach separates the analysis of the estimator type from the analysis of the data type. The analysis for overlapping-batch-means estimators is developed, as is the analysis for steady-state first-order autoregressive and moving-average data. Closed-form expressions for overlapping-batch-means estimators and independently distributed data are obtained.     

Estimating population proportions in the presence of missing data

This paper discusses the estimation of a population proportion in the presence of missing data and using auxiliary information at the estimation stage. A general class of estimators, which make efficient use of the available information, are proposed. Some theoretical properties of the proposed estimators are analyzed, and they allow us to find the optimal value for the proposed class in the sense of minimal variance. The optimal estimator is thus more efficient than the customary estimator. Results derived from a simulation study indicate that the proposed optimal estimator gives desirable results in comparison to alternative estimators.     

A Posteriori Error Estimators for Elliptic Equations with Discontinuous Coefficients

We consider linear elliptic equations with discontinuous coefficients in two and three space dimensions with varying boundary conditions. The problem is discretized with linear finite elements. An adaptive procedure based on a posteriori error estimators for the treatment of singularities is proposed. Within the class of quasi-monotonically distributed coefficients we derive a posteriori error estimators with bounds that are independent of the variation of the coefficients. In numerical test cases we confirm the robustness of the error estimators and observe that on adaptively refined meshes the reduction of the error is optimal with respect to the number of unknowns.     

Optimal variance estimation based on lagged second-order difference in nonparametric regression

Differenced estimators of variance bypass the estimation of regression function and thus are simple to calculate. However, there exist two problems: most differenced estimators do not achieve the asymptotic optimal rate for the mean square error; for finite samples the estimation bias is also important and not further considered. In this paper, we estimate the variance as the intercept in a linear regression with the lagged Gasser-type variance estimator as dependent variable. For the equidistant design, our estimator is not only \(n^{1/2}\)-consistent and asymptotically normal, but also achieves the optimal bound in terms of estimation variance with less asymptotic bias. Simulation studies show that our estimator has less mean square error than some existing differenced estimators, especially in the cases of immense oscillation of regression function and small-sized sample.     

Microstructure noise in the continuous case: The pre-averaging approach

This paper presents a generalized pre-averaging approach for estimating the integrated volatility, in the presence of noise. This approach also provides consistent estimators of other powers of volatility — in particular, it gives feasible ways to consistently estimate the asymptotic variance of the estimator of the integrated volatility. We show that our approach, which possesses an intuitive transparency, can generate rate optimal estimators (with convergence rate n^−1/4$n^{-1/4}$).     

Ordinary and penalized minimum power-divergence estimators in two-way contingency tables

Basu and Basu (Statistica Sinica 8:841–860, 1998) have proposed an empty cell penalty for the minimum power-divergence estimators which can lead to improvements in the small sample properties of these estimators. In this paper, we study the small and moderate sample performances of the ordinary and penalized minimum power-divergence estimators in terms of efficiency and robustness for the log-linear models in two-way contingency tables under the assumptions of multinomial sampling. Calculations made by enumerating all possible sample combinations show that the penalized estimators are competitive with the ordinary estimators for the moderate samples and definitely better for the smallest sample considered for both efficiency and robustness under the considered models. The results also reveal that the bigger the main effects the more need for penalization.     

Finite sample tail behavior of multivariate location estimators

A finite sample performance measure of multivariate location estimators is introduced based on “tail behavior”. The tail performance of multivariate “monotone” location estimators and the halfspace depth based “non-monotone” location estimators including the Tukey halfspace median and multivariate L-estimators is investigated. The connections among the finite sample performance measure, the finite sample breakdown point, and the halfspace depth are revealed. It turns out that estimators with high breakdown point or halfspace depth have “appealing” tail performance. The tail performance of the halfspace median is very appealing and also robust against underlying population distributions, while the tail performance of the sample mean is very sensitive to underlying population distributions. These findings provide new insights into the notions of the halfspace depth and breakdown point and identify the important role of tail behavior as a quantitative measure of robustness in the multivariate location setting.     

Dual divergence estimators and tests: Robustness results

The class of dual ?-divergence estimators (introduced in Broniatowski and Keziou (2009) [5]) is explored with respect to robustness through the influence function approach. For scale and location models, this class is investigated in terms of robustness and asymptotic relative efficiency. Some hypothesis tests based on dual divergence criteria are proposed and their robustness properties are studied. The empirical performances of these estimators and tests are illustrated by Monte Carlo simulation for both non-contaminated and contaminated data.     

Efficiencies of tests and estimators forp-order autoregressive processes when the error distribution is nonnormal

Summary We considerpth order autoregressive time series where the shocks need not be normal. By employing the concept of contiguity, we obtain the sysmptotic power for tests of hypothesis concerning the autoregressive parameters. Our approach allows consideration of the double exponential and other thicker-tailed distributions for the shocks. We derive a new result in the contiguity framework that leads directly to an expression for the Pitman efficiencies of tests as well as estimators. The numerical values of the efficiencies suggest a lack of robustness for the normal theory least squares estimators when the shock distribution is thick tailed or an outlier prone mixed normal. An important alternative test statistic is proposed that competes with the normal theory tests. This research was supported by the Office of Naval Research under Grant No. N00014-78-C-0722 and by the Army Research Office.     

Stochastic leadtimes in a one-warehouse,N-retailer inventory system with the warehouse not carrying stock

This study extends upon a multi-echelon inventory model developed by Graves, introducing in the one-warehouse, N-retailer case—as Graves suggested—stochastic leadtimes between the warehouse and the retail sites in place of the original deterministic leadtimes. Effects of stochastic leadtimes on required base stock levels at the retail sites in the case where the warehouse carries no stock (e.g., serves as a cross-dock point) were investigated analytically. Two alternative treatments of stochastic leadtime distributions were considered. Using as a baseline Graves’ computational study under deterministic leadtimes, results of the current study suggest that it may be better to use the deterministic model with an accurately estimated mean leadtime than a stochastic model with a poorly estimated mean leadtime.     

Adaptation under probabilistic error for estimating linear functionals

The problem of estimating linear functionals based on Gaussian observations is considered. Probabilistic error is used as a measure of accuracy and attention is focused on the construction of adaptive estimators which are simultaneously near optimal under probabilistic error over a collection of convex parameter spaces. In contrast to mean squared error it is shown that fully rate optimal adaptive estimators can be constructed for probabilistic error. A general construction of such estimators is provided and examples are given to illustrate the general theory.     

共同均值矩阵的线性估计的泛容许性

本文给出了多元线性模型中共同均值矩阵可估函数的线性估计的泛容许性定义,并得到了共同均值矩阵可估函数的线性估计分别在齐次和非齐次线性估计类中的泛容许性特征。     

On improved estimation of normal precision matrix and discriminant coefficients

The problem of estimating the precision matrix of a multivariate normal distribution model is considered with respect to a quadratic loss function. A number of covariance estimators originally intended for a variety of loss functions are adapted so as to obtain alternative estimators of the precision matrix. It is shown that the alternative estimators have analytically smaller risks than the unbiased estimator of the precision matrix. Through numerical studies of risk values, it is shown that the new estimators have substantial reduction in risk. In addition, we consider the problem of the estimation of discriminant coefficients, which arises in linear discriminant analysis when Fisher's linear discriminant function is viewed as the posterior log-odds under the assumption that two classes differ in mean but have a common covariance matrix. The above method is also adapted for this problem in order to obtain improved estimators of the discriminant coefficients under the quadratic loss function. Furthermore, a numerical study is undertaken to compare the properties of a collection of alternatives to the “unbiased” estimator of the discriminant coefficients.     

Nonparametric orthogonal series estimators of regression: A class attaining the optimal convergence rate in L2

In this note a class of nonparametric orthogonal series type estimators for regression function fitting is considered. Sufficient conditions are given for the estimators to attain the optimal convergence rate in the mean integrated square error sense. Using results from the theory of numerical integration, examples of estimators are given, for which the above mentioned conditions hold.     

Asymptotic theory for robust principal components

The asymptotic distribution of the eigenvalues and eigenvectors of the robust scatter matrix proposed by R. Maronna in 1976 is given when the observations are from an ellipsoidal distribution. The elements of each characteristic vector are the coefficients of a robustified version of principal components. We give a definition for the asymptotic efficiency of these estimators and we evaluate their influence curve. The problem of maximizing the efficiency under a bound on the influence curve is solved. Numerically, we calibrate the optimal estimators under the multivariate normal distribution and we evaluate their sensitivity.     

Coefficients of determination for least absolute deviation analysis

The least-absolute deviation or l₁ analysis of a linear model is an important alternative to the classical least squares analysis from the point of view of efficiency for longer-tailed error distributions and robustness to the presence of outliers. In this paper two coefficients of determination are proposed for the least-absolute deviation analysis. It is shown that they have desirable properties as measures of multiple association. Both fixed and random predictor variable cases are considered. In the case of random predictor variables, the sample coefficients of determination are shown to be consistent estimators of appropriate population parameters.