Large Sample Theory Research Articles

Meta-analysis framework for exact inferences with application to the analysis of rare events.

The usefulness of meta-analysis has been recognized in the evaluation of drug safety, as a single trial usually yields few adverse events and offers limited information. For rare events, conventional meta-analysis methods may yield an invalid inference, as they often rely on large sample theories and require empirical corrections for zero events. These problems motivate research in developing exact methods, including Tian et al.'s method of combining confidence intervals (2009, Biostatistics, 10, 275-281) and Liu et al.'s method of combining p-value functions (2014, JASA, 109, 1450-1465). This article shows that these two exact methods can be unified under the framework of combining confidence distributions (CDs). Furthermore, we show that the CD method generalizes Tian et al.'s method in several aspects. Given that the CD framework also subsumes the Mantel-Haenszel and Peto methods, we conclude that the CD method offers a general framework for meta-analysis of rare events. We illustrate the CD framework using two real data sets collected for the safety analysis of diabetes drugs.

Sample Size Calculation of Exact Tests for the Weak Causal Null Hypothesis in Randomized Trials with a Binary Outcome

The main purpose in many randomized trials is to make an inference about the average causal effect of a treatment. Therefore, on a binary outcome, the null hypothesis for the hypothesis test should be that the causal risks are equal in the two groups. This null hypothesis is referred to as the weak causal null hypothesis. Nevertheless, at present, hypothesis tests applied in actual randomized trials are not for this null hypothesis; Fisher’s exact test is a test for the sharp causal null hypothesis that the causal effect of treatment is the same for all subjects. In general, the rejection of the sharp causal null hypothesis does not mean that the weak causal null hypothesis is rejected. Recently, Chiba developed new exact tests for the weak causal null hypothesis: a conditional exact test, which requires that a marginal total is fixed, and an unconditional exact test, which does not require that a marginal total is fixed and depends rather on the ratio of random assignment. To apply these exact tests in actual randomized trials, it is inevitable that the sample size calculation must be performed during the study design. In this paper, we present a sample size calculation procedure for these exact tests. Given the sample size, the procedure can derive the exact test power, because it examines all the patterns that can be obtained as observed data under the alternative hypothesis without large sample theories and any assumptions.

Small Sample Kernel Association Tests for Human Genetic and Microbiome Association Studies.

Kernel machine based association tests (KAT) have been increasingly used in testing the association between an outcome and a set of biological measurements due to its power to combine multiple weak signals of complex relationship with the outcome through the specification of a relevant kernel. Human genetic and microbiome association studies are two important applications of KAT. However, the classic KAT framework relies on large sample theory, and conservativeness has been observed for small sample studies, especially for microbiome association studies. The common approach for addressing the small sample problem relies on computationally intensive resampling methods. Here, we derive an exact test for KAT with continuous traits, which resolve the small sample conservatism of KAT without the need for resampling. The exact test has significantly improved power to detect association for microbiome studies. For binary traits, we propose a similar approximate test, and we show that the approximate test is very powerful for a wide range of kernels including common variant- and microbiome-based kernels, and the approximate test controls the type I error well for these kernels. In contrast, the sequence kernel association tests have slightly inflated genomic inflation factors after small sample adjustment. Extensive simulations and application to a real microbiome association study are used to demonstrate the utility of our method.

Asymptotic distribution of the maximum interpoint distance in a sample of random vectors with a spherically symmetric distribution

Extreme value theory is part and parcel of any study of order statistics in one dimension. Our aim here is to consider such large sample theory for the maximum distance to the origin, and the related maximum “interpoint distance,” in multidimensions. We show that for a family of spherically symmetric distributions, these statistics have a Gumbel-type limit, generalizing several existing results. We also discuss the other two types of limit laws and suggest some open problems. This work complements our earlier study on the minimum interpoint distance.

Some small sample properties of the EWOC design in dose escalation trials

Model-based approaches to dose escalation trials such as CRM, EWOC and their variants are becoming increasingly popular in Phase I oncology trials. Although the literature includes a number of reasons why the practitioner may want to consider using a model-based design over a more traditional 3+3 design, there are still a number of reasons why they are not universally adopted. Theoretical reasons have often been provided for their superiority over 3+3 designs and these have generally relied on large sample theory, whereas Phase I trial sizes are usually considerably lower. Furthermore, when information external to the planned trial is available and may be used to construct informative prior distributions, the small sample properties of model-based designs are less well understood. By considering a planned trial for which only a discrete set of doses are available, we use simulation to show some small sample properties of the EWOC design (Babb et al, 1998). When the model recommends a dose between the available doses, we evaluate the strategy of always choosing the dose below that recommended by the model versus the closest available dose (even if that is higher). We also characterise two different parameterisations of the Bayesian 2-parameter logistic model that underpins the EWOC design and make some recommendations that could increase the acceptability of the EWOC design to those more comfortable with implementing a 3+3 design.

Adjusting for small-sample non-normality of design event estimators under a generalized Pareto distribution

Summary The generalized Pareto distribution (GPD) is a widely used frequency model for fitting extremes in hydrology, especially to fit exceedances over a threshold in the peaks-over-threshold (POT) modeling of floods or other extreme hydrological phenomena. A key goal in fitting frequency distributions to data is to allow the estimation of distribution quantiles, which in hydrology are often used as “design events”. The maximum likelihood (ML) method is a recommended method for fitting the GPD to data. To provide a measure of the statistical error involved in the estimation of design events, confidence intervals for quantiles (CIQs) have to be calculated. Hydrologists have traditionally used large-sample theory to construct such CIQs, but it is shown in the present study that this leads to inaccurate results for quantiles in the right-tail of a GPD. An improvement is therefore proposed for these classically obtained CIQs under a GPD model fitted by ML. The conventional and proposed approaches are compared through Monte Carlo (MC) simulation, and the resulting recommendations are put to use in a hydrological application.

Asymptotic theory for density ridges

The large sample theory of estimators for density modes is well understood. In this paper we consider density ridges, which are a higher-dimensional extension of modes. Modes correspond to zero-dimensional, local high-density regions in point clouds. Density ridges correspond to $s$-dimensional, local high-density regions in point clouds. We establish three main results. First we show that under appropriate regularity conditions, the local variation of the estimated ridge can be approximated by an empirical process. Second, we show that the distribution of the estimated ridge converges to a Gaussian process. Third, we establish that the bootstrap leads to valid confidence sets for density ridges.

Robust Parameter Design for Quality and Reliability Issues Based on Accelerated Degradation Measurements

Manufacturing quality and lifetime testing conditions may affect product reliability measurements. The literature for the design of experiments (DOE) and robust product optimization considering both quality and reliability issues is scarce. This article develops a model to include both manufacturing variables and accelerated degradation test (ADT) conditions. A simple algorithm provides calculations of the maximum likelihood estimates (MLEs) of these model parameters and percentile lifetimes. Variances of these estimates are derived based on large sample theory. Our DOE plans focus on deciding replication sizes and proportions of the test-units allocated at three stress levels for various manufacturing and ADT conditions. This work also explores robust parameter design (RPD) optimizations for selected controllable manufacturing variables to achieve the longest product lifetime and smallest variation in lifetime distributions.

Test Statistics for Volume under the ROC Surface and Hypervolume under the ROC Manifold

The area under the ROC curve can be represented by both Mann-Whitney and Wilcoxon rank sum statistics. Consider an ROC surface and manifold equal to three dimensions or more. This paper finds that the volume under the ROC surface (VUS) and the hypervolume under the ROC manifold (HUM) could be derived as functions of both conditional Mann-Whitney statistics and conditional Wilcoxon rank sum statistics. The nullhypothesis equal to three distribution functions or more are identical can be tested using VUS and HUM statistics based on the asymptotic large sample theory of Wilcoxon rank sum statistics. Illustrative examples with three and four random samples show that two approaches give the same VUS and . The equivalence of several distribution functions is also tested with VUS and in terms of conditional Wilcoxon rank sum statistics.

Estimating sampling error of evolutionary statistics based on genetic covariance matrices using maximum likelihood.

We explore the estimation of uncertainty in evolutionary parameters using a recently devised approach for resampling entire additive genetic variance-covariance matrices (G). Large-sample theory shows that maximum-likelihood estimates (including restricted maximum likelihood, REML) asymptotically have a multivariate normal distribution, with covariance matrix derived from the inverse of the information matrix, and mean equal to the estimated G. This suggests that sampling estimates of G from this distribution can be used to assess the variability of estimates of G, and of functions of G. We refer to this as the REML-MVN method. This has been implemented in the mixed-model program WOMBAT. Estimates of sampling variances from REML-MVN were compared to those from the parametric bootstrap and from a Bayesian Markov chain Monte Carlo (MCMC) approach (implemented in the R package MCMCglmm). We apply each approach to evolvability statistics previously estimated for a large, 20-dimensional data set for Drosophila wings. REML-MVN and MCMC sampling variances are close to those estimated with the parametric bootstrap. Both slightly underestimate the error in the best-estimated aspects of the G matrix. REML analysis supports the previous conclusion that the G matrix for this population is full rank. REML-MVN is computationally very efficient, making it an attractive alternative to both data resampling and MCMC approaches to assessing confidence in parameters of evolutionary interest.

Two-stage local Walsh average estimation of generalized varying coefficient models

In this paper we investigate the robust estimation of generalized varying coefficient models in which the unknown regression coefficients may change with different explanatory variables. Based on the B-spline series approximation and Walsh-average technique we develop an initial estimator for the unknown regression coefficient functions. By virtue of the initial estimator, the generalized varying coefficient model is reduced to a univariate nonparametric regression model. Then combining the local linear smooth and Walsh average technique we further propose a two-stage local linear Walsh-average estimator for the unknown regression coefficient functions. Under mild assumptions, we establish the large sample theory of the proposed estimators by utilizing the results of U-statistics and shows that the two-stage local linear Walsh-average estimator own an oracle property, namely the asymptotic normality of the two-stage local linear Walsh-average estimator of each coefficient function is not affected by other unknown coefficient functions. Extensive simulation studies are conducted to assess the finite sample performance, and a real example is analyzed to illustrate the proposed method.

Visualizing and Testing the Multivariate Linear Regression Model

Recent results make the multivariate linear regression model much easier to use. This model has $m \geq 2$ response variables. Results by Kakizawa (2009) and Su and Cook (2012) can be used to explain the large sample theory of the least squares estimator and of the widely used Wilks' $\Lambda$, Pillai's trace, and Hotelling Lawley trace test statistics. Kakizawa (2009) shows that these statistics have the same limiting distribution. This paper reviews these results and gives two theorems to show that the Hotelling Lawley test generalizes the usual partial $F$ test for $m = 1$ response variable to $m \geq 1$ response variables. Plots for visualizing the model are also given, and can be used to check goodness and lack of fit, to check for outliers and influential cases, and to check whether the error distribution is multivariate normal or from some other elliptically contoured distribution.

Assessing Characteristic Scales Using Wavelets

Summary Characteristic scale is a notion that pervades the geophysical sciences, but it has no widely accepted precise definition. Motivated by the facts that the wavelet transform decomposes a time series into coefficients that are associated with different scales and that the variance of these coefficients (the so-called wavelet variance) decomposes the variance of the time series across scales, the paper proposes a definition for characteristic scale in terms of peaks in plots of the wavelet variance versus scale. After presenting basic theory for wavelet-based characteristic scales, a natural estimator for these scales is considered. Large sample theory for this estimator permits the construction of confidence intervals for a true unknown characteristic scale. Computer experiments are presented that demonstrate the efficacy of the large sample theory for finite sample sizes. Characteristic scale estimates are calculated for medium multiyear Arctic sea ice, global temperature records, coherent structures in river flows and the Madden–Julian oscillation in an atmospheric time series.

Detecting changes in cross-sectional dependence in multivariate time series

Classical and more recent tests for detecting distributional changes in multivariate time series often lack power against alternatives that involve changes in the cross-sectional dependence structure. To be able to detect such changes better, a test is introduced based on a recently studied variant of the sequential empirical copula process. In contrast to earlier attempts, ranks are computed with respect to relevant subsamples, with beneficial consequences for the sensitivity of the test. For the computation of p-values we propose a multiplier resampling scheme that takes the serial dependence into account. The large-sample theory for the test statistic and the resampling scheme is developed. The finite-sample performance of the procedure is assessed by Monte Carlo simulations. Two case studies involving time series of financial returns are presented as well.

Saddlepoint Approximations of the Distribution of the Person Parameter in the Two Parameter Logistic Model

Large sample theory states the asymptotic normality of the maximum likelihood estimator of the person parameter in the two parameter logistic (2PL) model. In short tests, however, the assumption of normality can be grossly wrong. As a consequence, intended coverage rates may be exceeded and confidence intervals are revealed to be overly conservative. Methods belonging to the higher-order-theory, more specifically saddlepoint approximations, are a convenient way to deal with small-sample problems. Confidence bounds obtained by these means hold the approximate confidence level for a broad range of the person parameter. Moreover, an approximation to the exact distribution permits to compute median unbiased estimates (MUE) that are as likely to overestimate as to underestimate the true person parameter. Additionally, in small samples, these MUE are less mean-biased than the often-used maximum likelihood estimator.

Two Useful Reformulations of the Hazard Ratio

The hazard ratio is a standard summary for comparing survival curves yet hazard ratios are often difficult for scientists and clinicians to interpret. Insight into the interpretation of hazard ratios is obtained by relating hazard ratios to the maximum difference and an average difference between survival probabilities. These reformulations of the hazard ratio are useful in classroom discussions of survival analysis and when discussing analyses with scientists and clinicians. Large-sample distribution theory is provided for these reformulations of the hazard ratio. Two examples are used to illustrate the ideas.

Powerful Parametric Tests Based on Sum‐Functions of Spacings

ABSTRACTAssume that we have a sequence of n independent and identically distributed random variables with a continuous distribution function F, which is specified up to a few unknown parameters. In this paper, tests based on sum‐functions of sample spacings are proposed, and large sample theory of the tests are presented under simple null hypotheses as well as under close alternatives. Tests, which are optimal within this class, are constructed, and it is noted that these tests have properties that closely parallel those of the likelihood ratio test in regular parametric models. Some examples are given, which show that the proposed tests work also in situations where the likelihood ratio test breaks down. Extensions to more general hypotheses are discussed.

Estimating kurtosis and confidence intervals for the variance under nonnormality

Exact confidence intervals for variances rely on normal distribution assumptions. Alternatively, large-sample confidence intervals for the variance can be attained if one estimates the kurtosis of the underlying distribution. The method used to estimate the kurtosis has a direct impact on the performance of the interval and thus the quality of statistical inferences. In this paper the author considers a number of kurtosis estimators combined with large-sample theory to construct approximate confidence intervals for the variance. In addition, a nonparametric bootstrap resampling procedure is used to build bootstrap confidence intervals for the variance. Simulated coverage probabilities using different confidence interval methods are computed for a variety of sample sizes and distributions. A modification to a conventional estimator of the kurtosis, in conjunction with adjustments to the mean and variance of the asymptotic distribution of a function of the sample variance, improves the resulting coverage values for leptokurtically distributed populations.

Asymptotic refinements of a misspecification-robust bootstrap for generalized method of moments estimators

I propose a nonparametric iid bootstrap that achieves asymptotic refinements for t tests and confidence intervals based on GMM estimators even when the model is misspecified. In addition, my bootstrap does not require recentering the moment function, which has been considered as critical for GMM. Regardless of model misspecification, the proposed bootstrap achieves the same sharp magnitude of refinements as the conventional bootstrap methods which establish asymptotic refinements by recentering in the absence of misspecification. The key idea is to link the misspecified bootstrap moment condition to the large sample theory of GMM under misspecification of Hall and Inoue (2003). Two examples are provided: combining data sets and invalid instrumental variables.

Asymptotic Refinements of a Misspecification-Robust Bootstrap for Generalized Method of Moments Estimators

I propose a nonparametric iid bootstrap that achieves asymptotic refinements for t tests and confidence intervals based on the generalized method of moments (GMM) estimators even when the model is misspecified. In addition, my bootstrap does not require recentering the bootstrap moment function, which has been considered as critical for GMM. Regardless of model misspecification, the proposed bootstrap achieves the same sharp magnitude of refinements as the conventional bootstrap methods which establish asymptotic refinements by recentering in the absence of misspecification. The key idea is to link the misspecified bootstrap moment condition to the large sample theory of GMM under misspecification of Hall and Inoue (2003, Journal of Econometrics 114, 361-394). Examples of possibly misspecified moment condition models with Monte Carlo simulation results are provided: (i) Combining data sets, and (ii) invalid instrumental variables.