Permutation Tests for Multivariate Location Problems

The paper presents some permutation test procedures for multivariate location. The tests are based on projected univariate versions of multivariate data. For one-sample cases, the tests are affine invariant and strictly distribution-free for the symmetric null distribution with elliptical direction and their permutation counterparts are conditionally distribution-free when the underlying null distribution of the sample is angularly symmetric. For multi-sample cases, the tests are also affine invariant and permutation counterparts of the tests are conditionally distribution-free for any null distribution with certain continuity. Hence all of the tests in this paper are exactly valid. Furthermore, the equivalence, in the large sample sense, between the tests and their permutation counterparts are established. The power behavior of the tests and of their permutation counterparts under local alternative are investigated. A simulation study shows the tests to perform well compared with some existing tests in the literature, particularly when the underlying null distribution is symmetric whether light-tailed or heavy-tailed. For revealing the influence of data sparseness on the effect of the test, some simulations with different dimensions are also performed.     

Conditional tests for elliptical symmetry

In this paper, we suggest the conditional test procedures for testing elliptical symmetry of multivariate distribution. The conditional tests are exactly valid if the symmetric center and the shape matrix are given and are asymptotically valid if they are unknowns to be estimated. The equivalence, in the large sample sense, between the conditional tests and their unconditional counterparts is established. The power behavior of the tests under global as well as local alternatives is investigated theoretically. A small simulation study is performed.     

A Monte Carlo comparison of studentized bootstrap and permutation tests for heteroscedastic two-sample problems

The present Monte Carlo study compares bootstrap and permutation tests for semiparametric heteroscedastic two-sample testing problems of Behrens-Fisher type. The underlying functionals to be tested are (a) the difference of the means and (b) the Wilcoxon functionalP(Y < X) which is invariant under strictly increasing transformations. The consideration leads to semiparametric modifications of Welch type tests for the Behrens-Fisher model and an extended two-sample Wilcoxon test which also works under some null hypothesis with non-exchangeable distributions. The present Monte Carlo study confirms the high quality of studentized permutation tests at finite sample size. They are typically better than tests with asymptotic critical values and for many situations and they are also better than two-sample bootstrap tests when their type I error probabilities are compared.     

Profiting from correlations: Adjusted estimators for categorical data

To take sample biases and skewness in the observations into account, practitioners frequently weight their observations according to some marginal distribution. The present paper demonstrates that such weighting can indeed improve the estimation. Studying contingency tables, estimators for marginal distributions are proposed under the assumption that another marginal is known. It is shown that the weighted estimators have a strictly smaller asymptotic variance whenever the two marginals are correlated. The finite sample performance is illustrated in a simulation study. As an application to traffic accident data the method allows for correcting a well‐known bias in the observed injury severity distribution.     

Applications of conditional power function of two-sample permutation test

Permutation or randomization test is a nonparametric test in which the null distribution (distribution under the null hypothesis of no relationship or no effect) of the test statistic is attained by calculating the values of the test statistic overall permutations (or by considering a large number of random permutation) of the observed dataset. The power of permutation test evaluated based on the observed dataset is called conditional power. In this paper, the conditional power of permutation tests is reviewed. The use of the conditional power function for sample size estimation is investigated. Moreover, reproducibility and generalizability probabilities are defined. The use of these probabilities for sample size adjustment is shown. Finally, an illustration example is used.     

Bootstrap and permutation tests in ANOVA for directional data

The problem of testing the null hypothesis of a common direction across several populations defined on the hypersphere arises frequently when we deal with directional data. We may consider the Analysis of Variance (ANOVA) for testing such hypotheses. However, for the Watson distribution, a commonly used distribution for modeling axial data, the ANOVA test is only valid for large concentrations. So we suggest to use alternative tests, such as bootstrap and permutation tests in ANOVA. Then, we investigate the performance of these tests for data from Watson populations defined on the hypersphere.     

Permutation tests using least distance estimator in the multivariate regression model

This paper proposes two permutation tests based on the least distance estimator in a multivariate regression model. One is a type of t test statistic using the bootstrap method, and the other is a type of F test statistic using the sum of distances between observed and predicted values under the full and reduced models. We conducted a simulation study to compare the power of the proposed permutation tests with that of the parametric tests based on the least squares estimator for three types of hypotheses in several error distributions. The results indicate that the power of the proposed permutation tests is greater than that of the parametric tests when the error distribution is skewed like the Wishart distribution, has a heavy tail like the Cauchy distribution, or has outliers.     

A nonparametric version of Wilks’ lambda—Asymptotic results and small sample approximations

We propose a nonparametric version of Wilks’ lambda (the multivariate likelihood ratio test) and investigate its asymptotic properties under the two different scenarios of either large sample size or large number of samples. For unbalanced samples, a weighted and an unweighted variant are introduced. The unweighted variant of the proposed test appears to be novel also in the normal-theory context.The theoretical results are supplemented by a simulation study with parameter settings that are motivated by clinical and agricultural data, considering in particular the performance for small sample sizes, small number of samples, and varying dimensions. Inference methods based on the asymptotic sampling distribution and a small sample approximation are compared to permutation tests and to other parametric and nonparametric procedures. Application of the proposed method is illustrated by examples.     

Tests of fit for the Rayleigh distribution based on the empirical Laplace transform

In this paper a class of goodness-of-fit tests for the Rayleigh distribution is proposed. The tests are based on a weighted integral involving the empirical Laplace transform. The consistency of the tests as well as their asymptotic distribution under the null hypothesis are investigated. As the decay of the weight function tends to infinity the test statistics approach limit values. In a particular case the resulting limit statistic is related to the first nonzero component of Neyman’s smooth test for this distribution. The new tests are compared with other omnibus tests for the Rayleigh distribution.     

Heteroscedasticity checks for regression models

For checking on heteroscedasticity in regression models, a unified approach is proposed to constructing test statistics in parametric and nonparametric regression models. For nonparametric regression, the test is not affected sensitively by the choice of smoothing parameters which are involved in estimation of the nonparametric regression function. The limiting null distribution of the test statistic remains the same in a wide range of the smoothing parameters. When the covariate is one-dimensional, the tests are, under some conditions, asymptotically distribution-free. In the high-dimensional cases, the validity of bootstrap approximations is investigated. It is shown that a variant of the wild bootstrap is consistent while the classical bootstrap is not in the general case, but is applicable if some extra assumption on conditional variance of the squared error is imposed. A simulation study is performed to provide evidence of how the tests work and compare with tests that have appeared in the literature. The approach may readily be extended to handle partial linear, and linear autoregressive models.     

The excess-mass ellipsoid

The excess-mass ellipsoid is the ellipsoid that maximizes the difference between its probability content and a constant multiple of its volume, over all ellipsoids. When an empirical distribution determines the probability content, the sample excess-mass ellipsoid is a random set that can be used in contour estimation and tests for multimodality. Algorithms for computing the ellipsoid are provided, as well as comparative simulations. The asymptotic distribution of the parameters for the sample excess-mass ellipsoid are derived. It is found that a n^1/3 normalization of the center of the ellipsoid and lengths of its axes converge in distribution to the maximizer of a Gaussian process with quadratic drift. The generalization of ellipsoids to convex sets is discussed.     

Goodness-of-Fit Tests for the Cauchy Distribution Based on the Empirical Characteristic Function

Let X ₁,...,X _n be independent observations on a random variable X. This paper considers a class of omnibus procedures for testing the hypothesis that the unknown distribution of X belongs to the family of Cauchy laws. The test statistics are weighted integrals of the squared modulus of the difference between the empirical characteristic function of the suitably standardized data and the characteristic function of the standard Cauchy distribution. A large-scale simulation study shows that the new tests compare favorably with the classical goodness-of-fit tests for the Cauchy distribution, based on the empirical distribution function. For small sample sizes and short-tailed alternatives, the uniformly most powerful invariant test of Cauchy versus normal beats all other tests under discussion.     

Nonparametric methods for unbalanced multivariate data and many factor levels

We propose different nonparametric tests for multivariate data and derive their asymptotic distribution for unbalanced designs in which the number of factor levels tends to infinity (large a, small n_i case). Quasi gratis, some new parametric multivariate tests suitable for the large a asymptotic case are also obtained. Finite sample performances are investigated and compared in a simulation study. The nonparametric tests are based on separate rankings for the different variables. In the presence of outliers, the proposed nonparametric methods have better power than their parametric counterparts. Application of the new tests is demonstrated using data from plant pathology.     

Large sample prediction intervals for future geometric mean a comparative study

This paper provides simulation comparisons among the performance of 11 possible prediction intervals for the geometric.mean of a Pareto distribution with parameters (αB, ). Six different procedures were used to obtain these intervals , namely; true inter -val , pivotal interval , maximum likelihood estimation interval, centrallimit teorem interval, variance stabilizing interval and a mixture of the above intervals . Some of these intervals are valid if the observed sample size m,are large , others are valid if both, n and the future sample size m, are large. Some of these intervals require a knowledge of α or B, while others do not. The simulation validation and efficiency study shows that intervals depending on the MLE's are the best. The second best intervalsare those obtained through pivotal methods or variance stabilization transformation. The third group of intervals is that which depends on the central limit theorem when λ is known. There are two intervals which proved to be unacceptable under any criterion.     

A study of resampling methods for testing equality of means in paired population

The goal of this study is to compare the resampling methods including bootstrap and permutation tests against classical methods for paired samples. A simulation study was conducted to see the performance of both parametric and nonparametric methods under various assumptions such as non-normal populations and small or large sample sizes. The results of the simulation study are with respect to type I error and power of the test.     

Best replies and adaptive learning

This paper discusses the relationships between learning processes based on the iterated elimination of strictly dominated strategies and their myopic and more naive counterparts. The concept of a monotone game, of which games with strategic complementarities are a subclass, is introduced. Then it is shown that convergence under best reply dynamics and dominance solvability are equivalent for all two-player (and some many-player) games in this class.     

On locally optimal nonparametric tests

Summary There is an abundancy of problems in which no parametric model realistically describes the situation and in which, accordingly, we have to resort to nonparametric methods. As the numerical problems connected with nonparametric tests are becoming less and less important, rank tests, permutation tests and the like are becoming more and more part of the standard armatory of applied statisticians. The lack of tabulated critical values, for instance, should no longer be a serious objection against the use of permutation tests in practice; cf. Edgington (1987).The rationale underlying permutation and rank tests has been outlined in quite a number of text books and papers; cf. Fraser (1957), Lehmann (1959), Hájek-Sidák (1967) or Witting (1970). Roughly speaking, permutation tests are constructibel if the data can be condensed by means of a sufficient and complete statistic allowing for the proper kind of conditioning. Rank tests arise if the underlying problem is invariant with respect to (w.r.t.) a large group of transformations which leads to a maximal invariant statistic consisting of (signed) ranks.Most practical nonparametric problems, however, are too complex to be tractable by just one of those approaches. Many of them, however, can be handled by a combination of both techniques. In this paper we outline the logic underlying that combined reduction method and apply it to construct locally most powerful tests. Moreover, we discuss what we label Hoeffding's transfer problem, i.e. the uniformity aspect of locally most powerful tests with respect to the starting point at the boundary. We are concentrating on the discussion of the nonparametric two-sample location and scale problem. Further important problems are mentioned in Section III.This is a written account of an invited lecture delivered by the third author on occasion of the 14th Symposium über Operations Research, Ulm, September 6–8, 1989.     

Asymptotic Optimality of Certain Multihypothesis Sequential Tests: Non‐i.i.d. Case

It is known that certain combinations of one‐sided sequential probability ratio tests are asymptotically optimal (relative to the expected sample size) for problems involving a finite number of possible distributions when probabilities of errors tend to zero and observations are independent and identically distributed according to one of the underlying distributions. The objective of this paper is to show that two specific constructions of sequential tests asymptotically minimize not only the expected time of observation but also any positive moment of the stopping time distribution under fairly general conditions for a finite number of simple hypotheses. This result appears to be true for general statistical models which include correlated and non‐homogeneous processes observed either in discrete or continuous time. For statistical problems with nuisance parameters, we consider invariant sequential tests and show that the same result is valid for this case. Finally, we apply general results to the solution of several particular problems such as a multi‐sample slippage problem for correlated Gaussian processes and for statistical models with nuisance parameters. This revised version was published online in June 2006 with corrections to the Cover Date.     

Testing the no-treatment effect based on a possibly misspecified accelerated failure time model

The accelerated failure time model is a useful alternative to the Cox proportional hazard model. We investigate whether or not a misspecified accelerated failure time model provides a valid test of the no-treatment effect in randomized clinical trials. We show that the minimum dispersion statistic based on rank regression by Wei et al. (1990) must be modified in order to conduct valid tests under misspecification, whereas the resampling-based methods by Jin et al. (2003) are valid without any modification. Numerical studies are conducted to examine the small sample behavior of the modified minimum dispersion statistic and the resampling-based method. Finally, an illustration is given with a dataset from a clinical trial.