Bootstrap adaptive estimation: The trimmed-mean example

We consider the problem of choosing among a class of possible estimators by selecting the estimator with the smallest bootstrap estimate of finite sample variance. This is an alternative to using cross-validation to choose an estimator adaptively. The problem of a confidence interval based on such an adaptive estimator is considered. We illustrate the ideas by applying the method to the problem of choosing the trimming proportion of an adaptive trimmed mean. It is shown that a bootstrap adaptive trimmed mean is asymptotically normal with an asymptotic variance equal to the smallest among trimmed means. The asymptotic coverage probability of a bootstrap confidence interval based on such adaptive estimators is shown to have the nominal level. The intervals based on the asymptotic normality of the estimator share the same asymptotic result, but have poor small-sample properties compared to the bootstrap intervals. A small-sample simulation demonstrates that bootstrap adaptive trimmed means adapt themselves rather well even for samples of size 10.     

Statistics in the High School Curriculum

Recent work, spearheaded by Charles Dunnett (1980a), leads to the conclusion that the Tukey-Kramer (TK) method (popularly known as “Kramer's Method”) is the recommended multiple comparisons procedure for the simultaneous estimation of all pairwise differences of means in an imbalanced one-way ANOVA design with homogeneous variances. Nine other multiple comparisons methods are compared to each other and to the TK method using the criteria of conservativeness, narrowness of confidence intervals, robustness, and ease of use. The degree of superiority of the TK method over these methods, especially over the popular Bonferroni method, is sufficient to warrant recommending its use. Because of the lack of robustness of the TK method in heterogeneous variance cases, other methods are recommended.     

Asymptotically valid single-stage multiple-comparison procedures

We establish general conditions for the asymptotic validity of single-stage multiple-comparison procedures (MCPs) under the following general framework. There is a finite number of independent alternatives to compare, where each alternative can represent, e.g., a population, treatment, system or stochastic process. Associated with each alternative is an unknown parameter to be estimated, and the goal is to compare the alternatives in terms of the parameters. We establish the MCPs’ asymptotic validity, which occurs as the sample size of each alternative grows large, under two assumptions. First, for each alternative, the estimator of its parameter satisfies a central limit theorem (CLT). Second, we have a consistent estimator of the variance parameter appearing in the CLT. Our framework encompasses comparing means (or other moments) of independent (not necessarily normal) populations, functions of means, quantiles, steady-state means of stochastic processes, and optimal solutions of stochastic approximation by the Kiefer–Wolfowitz algorithm. The MCPs we consider are multiple comparisons with the best, all pairwise comparisons, all contrasts, and all linear combinations, and they allow for unknown and unequal variance parameters and unequal sample sizes across alternatives.     

Determination of sample size for selecting the smallest of k poisson population means

A procedure for selecting a Poisson population with smallest mean is considered using an indifference zone approach. The objective is to determine the smallest sample size n required from k ≥ 2 populations in order to attain the desired probability of correct selection. Since the means procedure is not consistent with respect to the difference or ratio alone, two distance measures are used simultaneously to overcome the difficulty in obtaining the smallest probability of correct selection that is greater than some specified limit. The constants required to determine n are computed and tabulated. The asymptotic results are derived using a normal approximation. A comparison with the exact results indicates that the proposed approximation works well. Only in the extreme cases small increases in n are observed. An example of industrial accident data is used to illustrate this procedure.     

Robust estimation methods for exponential data: a monte-carlo comparison

We compare the performance of seven robust estimators for the parameter of an exponential distribution. These include the debiased median and two optimally-weighted one-sided trimmed means. We also introduce four new estimators: the Transform, Bayes, Scaled and Bicube estimators. We make the Monte Carlo comparisons for three sample sizes and six situations. We evaluate the comparisons in terms of a new performance measure, Mean Absolute Differential Error (MADE), and a premium/protection interpretation of MADE. We organize the comparisons to enhance statistical power by making maximal use of common random deviates. The Transform estimator provides the best performance as judged by MADE. The singly-trimmed mean and Transform method define the efficient frontier of premium/protection.     

Pairwise-Difference Rank Estimation of the Transformation Model

This article considers pairwise-difference rank estimators of the coefficient vector in a transformation model. These estimators, like other existing rank estimators, require no subjective bandwidth choice. Monte Carlo simulations, numerical asymptotic efficiency comparisons, and two empirical applications suggest that the proposed estimators perform well in comparison with existing semiparametric estimators.     

A distribution-free approach to qdantal response assays

A nonparametric method is developed to estimate the minimum dosage level required to induce a given response rate in an experiment. The only assumption used about the response rate is that it is a nondecreasing function with respect to the dosage level. Let n_isubjects be independently tested at dosage level x_ix₁x₂x_k. This paper presents methodology for the estimation of the smallest i such that the response probability at x_i is no less than a required level p. A comparison with well-known nonparametric methods shows that this method is better in some cases. A design of minimum required sample size for a given accuracy is also developed.     

Tables for pairwise multiple comparisons using shaffer's modifiedsequentially-rejective procedure

A sequentially rejective (SR) testing procedure introduced by Holm (1979) and modified (MSR) by Shaffer (1986) is considered for testing all pairwise mean comparisons.For such comparisons, both the SR and MSR methods require that the observed test statistics be ordered and compared, each in turn, to appropriate percentiles on Student's t distribution.For the MSR method these percentiles are based on the maximum number of true null hypotheses remaining at each stage of the sequential procedure, given prior significance at previous stages, A function is developed for determining this number from the number of means being tested and the stage of the test.For a test of all pairwise comparisons, the logical implications which follow the rejection of a null hypothesis renders the MSR procedure uniformly more powerful than the SR procedure.Tables of percentiles for comparing K means, 3 < K < 6, using the MSR method are presented.These tables use Sidak's (1967) multiplicative inequality and simplify the use of t he MSR procedure.Several modifications to the MSR are suggested as a means of further increasing the power for testing the pairwise comparisons.General use of the MSR and the corresponding function for testing other parameters besides the mean is discussed.     

A Bayesian Hierarchical Model for Multiple Comparisons in Mixed Models

We propose a Bayesian hierarchical model for multiple comparisons in mixed models where the repeated measures on subjects are described with the subject random effects. The model facilitates inferences in parameterizing the successive differences of the population means, and for them, we choose independent prior distributions that are mixtures of a normal distribution and a discrete distribution with its entire mass at zero. For the other parameters, we choose conjugate or vague priors. The performance of the proposed hierarchical model is investigated in the simulated and two real data sets, and the results illustrate that the proposed hierarchical model can effectively conduct a global test and pairwise comparisons using the posterior probability that any two means are equal. A simulation study is performed to analyze the type I error rate, the familywise error rate, and the test power. The Gibbs sampler procedure is used to estimate the parameters and to calculate the posterior probabilities.     

Two-sided generalized Topp and Leone (TS-GTL) distributions

Over 50 years ago, in a 1955 issue of JASA, a paper on a bounded continuous distribution by Topp and Leone [C.W. Topp and F.C. Leone, A family of J-shaped frequency functions, J. Am. Stat. Assoc. 50(269) (1955), pp. 209–219] appeared (the subject was dormant for over 40 years but recently the family was resurrected). Here, we shall investigate the so-called Two-Sided Generalized Topp and Leone (TS-GTL) distributions. This family of distributions is constructed by extending the Generalized Two-Sided Power (GTSP) family to a new two-sided framework of distributions, where the first (second) branch arises from the distribution of the largest (smallest) order statistic. The TS-GTL distribution is generated from this framework by sampling from a slope (reflected slope) distribution for the first (second) branch. The resulting five-parameter TS-GTL family of distributions turns out to be flexible, encompassing the uniform, triangular, GTSP and two-sided slope distributions into a single family. In addition, the probability density functions may have bimodal shapes or admitting shapes with a jump discontinuity at the ‘threshold’ parameter. We will discuss some properties of the TS-GTL family and describe a maximum likelihood estimation (MLE) procedure. A numerical example of the MLE procedure is provided by means of a bimodal Galaxy M87 data set concerning V–I color indices of 80 globular clusters. A comparison with a Gaussian mixture fit is presented.     

Multiple comparisons with ‘best’ for multivariate normal populations

Comparisons of multivariate normal populations are made using a mul-tivariate approach (instead of reducing the problem to a univariate one). A rather negative finding is that, for comparisons with the ‘best’ of each variate, repeated univariate comparisons appear to be almost as efficient as multivariate comparisons, at least for the bivariate case and, under certain circumstances, for higher dimensional cases. Investigations are done on comparisons with the ‘MAX-best’ population (that one having the largest maximum of the marginal means), the ‘MIN-best’ (having the largest minimum) and the ‘O-best’ (being closest to largest in all marginal means). Detailed results are given for the bivariate normal with extensions indicated for the multivariate.     

Bounds on minimum mean squared error in ridge regression

The minimum MSE (mean squared error) of ridge regression coefficient estimates (for a given set of eigenvalues and variance) is a function of the transformed coefficient vector. In this paper, the authors prove that the minimum MSE is bounded, for a given coefficient vector length, by the two cases corresponding to the signal completely contained in the component associated with the smallest or largest eigenvalue. The implication for evaluating proposed estimators of the ridge regression biasing parameter is discussed.     

Theory of nonparametric statistics for rounded-off data with applications

This article gives a unified account of nonparametric statistics, covering testing, estimating, multiple comparisons, analysis of variance and regression, for rounded-off data, with ties handled by the average scores method. The theory is illustrated by means of numerous applications to the Biomedical, Engineering and Behavioral Sciences.     

Automatic selection of the proper family for simultaneous confidence intervals

Statisticians often employ simultaneous confidence intervals to reduce the likelihood of their drawing false conclusions when they must make a number of comparisons. To do this properly, it is necessary to consider the family of comparisons over which simultaneous confidence must be assured. Sometimes it is not clear what family of comparisons is appropriate. We describe how computer software can monitor the types of contrasts a user examines, and select the smallest family of contrasts that is likely to be of interest. We also describe how to calculate simultaneous confidence intervals for these families using a hybrid of the Bonferroni and Scheffé methods. Our method is especially suitable for problems with discrete and continuous predictors.     

On the Skew-Normal-Cauchy Distribution

A fixed effects one-way layout model of analysis of variance is considered where the variances are taken to be possibly unequal. Conservative single-stage procedures based on Banerjee’s method for the solution of the Behrens-Fisher problem are proposed for the following multiple comparisons problems: 1) all pairwise comparisons with a control population mean, and 2) all pairwise comparisons and all linear contrasts among the means. Since these procedures are likely to be very conservative in practice, approximate procedures based on Welch’s method for the solution of the Behrens-Fisher problem are suggested as alternatives. Monte Carlo studies indicate that the latter are much less conservative and hence may be better in practice. Both these sets of procedures need only the tables of the Student’s t-distribution for their application and are very simple to use. Exact two-stage procedures are proposed for the following multiple comparisons problems: 1) all pairwise comparisons and all linear contrasts among the means, and 2) all linear combinations of the means.     

A Starship Estimation Method for the Generalized λ Distributions

A development of the 'starship' method (Owen, 1988), a computer intensive estimation method, is presented for two forms of generalized λ distributions (gλd). The method can be used for the full parameter space and is flexible, allowing choice of both the form of the generalized λ distribution and of the nature of fit required. Some examples of its use in fitting data and approximating distributions are given. Some simulation studies explore the sampling distribution of the parameter estimates produced by this method for selected values of the parameters and consider comparisons with two other methods, for one of the gλd distributional forms, not previously so investigated. In the forms and parameter regions available to the other methods, it is demonstrated that the starship compares favourably. Although the differences between the methods, where available, tend to disappear with largersamples, the parameter coverage, flexibility and adaptability of the starship method make it attractive. However, the paper also demonstrates that care is needed when fitting and using such quantile-defined distributional families that are rich in shape, but have complex properties.     

Bounds and approximations for pairwise comparisons of independent multinormal means

Properties and relationships of some commonly used probability bounds, along with other recently developed bounds and approximations, are evaluated for their performance with pairwise comparisons. The comparisons are of independent sample means obtained from normal random variables with a common variance. Computational methods are presented and numerical results are used to further evaluate the performance of the bounds.     

Second-order approximations for a multivariate analog of Behrens-Fisher problem through three-stage procedure

Abstract

In this article, we have proposed a three-stage procedure for the estimation of the difference of the means of two multivariate normal populations having unknown and unequal variances. Point as well as confidence region estimation is done for the same. Here, we have used the concept of classical Behrens-Fisher problem. Second-order approximations are obtained in both the cases, i.e., point estimation and confidence region estimation.     

Bayesian approach for meta - analysis of controlled clinical trials

A new method to perform meta - analysis of controlled clinical trials with binary response variable is developed using a Bayesian approach. It consists of three parts: (1) For each trial, the risk difference (the proportion of successes in the treated group minus the proportion of successes in the control group) is estimated; (2) The homogeneity of the risk difference among the different trials is tested; and (3) The hypothesis - the effect of the treatment for the homogeneous pool of trials is greater than or equal to a given fixed constant - is tested. The performance of the Bayesian procedure to test the homogeneity of the risk difference among trials is compared with the chi - square test proposed by DerSimonian and Laird (Controlled Clinical Trials 7, 177-188, 1986) by means of pseudo - random simulation. The conclusion was that the Bayes test is more reliable, either in its exact or asymptotic versions, since it makes fewer decision errors than the chi-square test. As an illustration, the Bayesian method is applied to data of chemotherapeutic prophylaxis of superficial bladder cancer.     

A comparison of some robust,adaptive, and partially adaptive estimators of regression models

Numerous estimation techniques for regression models have been proposed. These procedures differ in how sample information is used in the estimation procedure. The efficiency of least squares (OLS) estimators implicity assumes normally distributed residuals and is very sensitive to departures from normality, particularly to "outliers" and thick-tailed distributions. Lead absolute deviation (LAD) estimators are less sensitive to outliers and are optimal for laplace random disturbances, but not for normal errors. This paper reports monte carlo comparisons of OLS,LAD, two robust estimators discussed by huber, three partially adaptiveestimators, newey's generalized method of moments estimator, and an adaptive maximum likelihood estimator based on a normal kernal studied by manski. This paper is the first to compare the relative performance of some adaptive robust estimators (partially adaptive and adaptive procedures) with some common nonadaptive robust estimators. The partially adaptive estimators are based on three flxible parametric distributions for the errors. These include the power exponential (Box-Tiao) and generalized t distributions, as well as a distribution for the errors, which is not necessarily symmetric. The adaptive procedures are "fully iterative" rather than one step estimators. The adaptive estimators have desirable large sample properties, but these properties do not necessarily carry over to the small sample case.

The monte carlo comparisons of the alternative estimators are based on four different specifications for the error distribution: a normal, a mixture of normals (or variance-contaminated normal), a bimodal mixture of normals, and a lognormal. Five hundred samples of 50 are used. The adaptive and partially adaptive estimators perform very well relative to the other estimation procedures considered, and preliminary results suggest that in some important cases they can perform much better than OLS with 50 to 80% reductions in standard errors.