Some asymptotics for U-statistics

We consider a one-sample U-statistic with kernel of dimension 2. We obtain its asymptotic bias and skewness and its Edgeworth and Cornish-Fisher type expansions. We also consider in less detail the one sample U-statistic with kernel of arbitrary dimension.     

Multiple change-point estimation with U-statistics

We consider a multiple change-point problem: a finite sequence of independent random variables consists of segments given by a known number of the so-called change-points such that the underlying distribution differs from segment to segment. The task is to estimate these change-points under no further assumptions on the within-segment distributions. In this completely nonparametric framework the proposed estimator is defined as the maximizing point of weighted multivariate U-statistic processes. Under mild moment conditions we prove almost sure convergence and the rate of convergence.     

Two-stage U-statistics for Hypothesis Testing

Abstract.  A U -statistic is not easy to apply or cannot be applied in hypothesis testing when it is degenerate or has an indeterminate degeneracy under the null hypothesis. A class of two-stage U -statistics (TU-statistics) is proposed to remedy these drawbacks. Both the asymptotic distributions under the null and the alternative of TU-statistics are shown to have simple forms. When the degeneracy is indeterminate, the Pitman asymptotic relative efficiency of a TU-statistic dominates that of the incomplete U -statistics. If the kernel is degenerate under the null hypothesis but non-degenerate under the alternative, a TU-statistic is proved to be more powerful than its corresponding U -statistic. Applications to testing independence of paired angles in ecology and marine biology are given. Finally, a simulation study shows that a TU-statistic is more powerful than its corresponding incomplete U -statistic in almost all cases under two settings.     

Minimum variance rectangular designs for U-statistics

We consider a certain class of rectangular designs for incomplete U-statistics based on Latin squares and show it to be optimal with respect to the minimal variance criterion. We also show it to be asymptotically efficient when compared with the corresponding complete statistics, as well as uniformly more efficient than the random subset selection. We provide the necessary and sufficient conditions for the existence of our design and give some examples of applications.     

A james-stein type detour of U-statistics

For a vector of estimable parameters, a modified version of the James-Stein rule (incorporating the idea of preliminary test estimators) is utilized in formulating some estimators based on U-statistics and their jackknifed estimator of dispersion matrix. Asymptotic admissibility properties of the classical U-statistics, their preliminary test version and the proposed estimators are studied.     

The Berry-Esseen bound for Studentized U-statistics

Callaert and Veraverbeke (1981) recently obtained a Berry-Esseen-type bound of order n^–1/2 for Studentized nondegenerate U-statistics of degree two. The condition these authors need to obtain this order bound is the finiteness of the 4.5th absolute moment of the kernel h. In this note it is shown that this assumption can be weakened to that of a finite (4 + ?)th absolute moment of the kernel h, for some ? > 0. Our proof resembles part of Helmers and van Zwet (1982), where an analogous result is obtained for the Student t-statistic. The present note extends this to Studentized U-statistics.     

Testing for changes using permutations of U-statistics

The critical values for various tests based on U-statistics to detect a possible change are obtained through permutations of the observations. We obtain the same approximations for the permutated U-statistics under the no change null hypothesis as well as under the exactly one change alternative. The results are used to show that the simulated critical values are asymptotically valid under the null hypothesis and the tests reject with the probability tending to one under the alternative.     

Bootstrap for U-statistics: a new approach

Bootstrap for nonlinear statistics like U-statistics of dependent data has been studied by several authors. This is typically done by producing a bootstrap version of the sample and plugging it into the statistic. We suggest an alternative approach of getting a bootstrap version of U-statistics. We will show the consistency of the new method and compare its finite sample properties in a simulation study and by applying both methods to financial data.     

Clustering and classification problems in genetics through U-statistics

ABSTRACT

Genetic data are frequently categorical and have complex dependence structures that are not always well understood. For this reason, clustering and classification based on genetic data, while highly relevant, are challenging statistical problems. Here we consider a versatile U-statistics-based approach for non-parametric clustering that allows for an unconventional way of solving these problems. In this paper we propose a statistical test to assess group homogeneity taking into account multiple testing issues and a clustering algorithm based on dissimilarities within and between groups that highly speeds up the homogeneity test. We also propose a test to verify classification significance of a sample in one of two groups. We present Monte Carlo simulations that evaluate size and power of the proposed tests under different scenarios. Finally, the methodology is applied to three different genetic data sets: global human genetic diversity, breast tumour gene expression and Dengue virus serotypes. These applications showcase this statistical framework's ability to answer diverse biological questions in the high dimension low sample size scenario while adapting to the specificities of the different datatypes.     

Unbiased Estimation of Central Moments by using U-statistics

We obtain an estimator of the r th central moment of a distribution, which is unbiased for all distributions for which the first r moments exist. We do this by finding the kernel which allows the r th central moment to be written as a regular statistical functional. The U-statistic associated with this kernel is the unique symmetric unbiased estimator of the r th central moment, and, for each distribution, it has minimum variance among all estimators which are unbiased for all these distributions.     

On Berry-Esseen theorem for some functions of U-statistics

The rate of convergence in the central limit theorem and in the random central limit theorem for some functions of U-statistics are established. The theorems refer to the asymptotic behaviour of the sequence {g(U_n),n≥1}, where g belongs to the class of all differentiable functions g such that g′εL(δ) and U_n is a U-statistics.     

Distributed inference for two-sample U-statistics in massive data analysis

This paper considers distributed inference for two-sample U-statistics under the massive data setting. In order to reduce the computational complexity, this paper proposes distributed two-sample U-statistics and blockwise linear two-sample U-statistics. The blockwise linear two-sample U-statistic, which requires less communication cost, is more computationally efficient especially when the data are stored in different locations. The asymptotic properties of both types of distributed two-sample U-statistics are established. In addition, this paper proposes bootstrap algorithms to approximate the distributions of distributed two-sample U-statistics and blockwise linear two-sample U-statistics for both nondegenerate and degenerate cases. The distributed weighted bootstrap for the distributed two-sample U-statistic is new in the literature. The proposed bootstrap procedures are computationally efficient and are suitable for distributed computing platforms with theoretical guarantees. Extensive numerical studies illustrate that the proposed distributed approaches are feasible and effective.     

Tests for high-dimensional covariance matrices using the theory of U-statistics

Test statistics for sphericity and identity of the covariance matrix are presented, when the data are multivariate normal and the dimension, p, can exceed the sample size, n. Under certain mild conditions mainly on the traces of the unknown covariance matrix, and using the asymptotic theory of U-statistics, the test statistics are shown to follow an approximate normal distribution for large p, also when p?n. The accuracy of the statistics is shown through simulation results, particularly emphasizing the case when p can be much larger than n. A real data set is used to illustrate the application of the proposed test statistics.     

Non-Gaussian limit distributions for U-statistics based on trimmed and Winsorized samples

Conditions ensuring the asymptotic normality of U-statistics based on either trimmed samples or Winsorized samples are well known [P. Janssen, R. Serfling, and N. Veraverbeke, Asymptotic normality of U-statistics based on trimmed samples, J. Statist. Plann. Inference 16 (1987), pp. 63–74; U-statistics on Winsorized and trimmed samples, Statist. Probab. Lett. 9 (1990), pp. 439–447]. However, the class of U-statistics has a much richer family of limiting distributions. This paper complements known results by providing general limit theorems for U-statistics based on trimmed or Winsorized samples where the limiting distribution is given in terms of multiple Ito–Wiener stochastic integrals.     

Model selection consistency of U-statistics with convex loss and weighted lasso penalty

In the paper we consider minimisation of U-statistics with the weighted Lasso penalty and investigate their asymptotic properties in model selection and estimation. We prove that the use of appropriate weights in the penalty leads to the procedure that behaves like the oracle that knows the true model in advance, i.e. it is model selection consistent and estimates nonzero parameters with the standard rate. For the unweighted Lasso penalty, we obtain sufficient and necessary conditions for model selection consistency of estimators. The obtained results strongly based on the convexity of the loss function that is the main assumption of the paper. Our theorems can be applied to the ranking problem as well as generalised regression models. Thus, using U-statistics we can study more complex models (better describing real problems) than usually investigated linear or generalised linear models.     

Invariant co-ordinate selection

Summary.  A general method for exploring multivariate data by comparing different estimates of multivariate scatter is presented. The method is based on the eigenvalue–eigenvector decomposition of one scatter matrix relative to another. In particular, it is shown that the eigenvectors can be used to generate an affine invariant co-ordinate system for the multivariate data. Consequently, we view this method as a method for invariant co-ordinate selection . By plotting the data with respect to this new invariant co-ordinate system, various data structures can be revealed. For example, under certain independent components models, it is shown that the invariant co- ordinates correspond to the independent components. Another example pertains to mixtures of elliptical distributions. In this case, it is shown that a subset of the invariant co-ordinates corresponds to Fisher's linear discriminant subspace, even though the class identifications of the data points are unknown. Some illustrative examples are given.     

Second-order and resampling approximation of finite population U-statistics based on stratified samples

We construct one-term Edgeworth expansions to distributions of U statistics and Studentized U-statistics, based on stratified samples drawn without replacement. Replacing the cumulants defining the expansions by consistent jackknife estimators, we obtain empirical Edgeworth expansions. The expansions provide second-order approximations that improve upon the normal approximation. Theoretical results are illustrated by a simulation study where we compare various approximations to the distribution of the commonly used Gini's mean difference estimator.