Statistical Methodology最新文献

In medical diagnostics, the ROC curve is the graph of sensitivity against 1-specificity as the diagnostic threshold runs through all possible values. The ROC curve and its associated summary indices are very useful for the evaluation of the discriminatory ability of biomarkers/diagnostic tests with continuous measurements. Among all summary indices, the area under the ROC curve (AUC) is the most popular diagnostic accuracy index, which has been extensively used by researchers for biomarker evaluation and selection. Sometimes, taking the actual measurements of a biomarker is difficult and expensive, whereas ranking them without actual measurements can be easy. In such cases, ranked set sampling based on judgment order statistics would provide more representative samples yielding more accurate estimation. In this study, Gaussian kernel is utilized to obtain a nonparametric estimate of the AUC. Asymptotic properties of the AUC estimates are derived based on the theory of U-statistics. Intensive simulation is conducted to compare the estimates using ranked set samples versus simple random samples. The simulation and theoretical derivation indicate that ranked set sampling is generally preferred with smaller variances and mean squared errors (MSE). The proposed method is illustrated via a real data analysis.

Latent class analysis is used to group categorical data into classes via a probability model. Model selection criteria then judge how well the model fits the data. When addressing incomplete data, the current methodology restricts the imputation to a single, pre-specified number of classes. We seek to develop an entropy-based model selection criterion that does not restrict the imputation to one number of clusters. Simulations show the new criterion performing well against the current standards of AIC and BIC, while a family studies application demonstrates how the criterion provides more detailed and useful results than AIC and BIC.

Designing cluster trials depends on the knowledge of the intracluster correlation coefficient. To overcome the issue of parameter dependence, Bayesian designs are proposed for two level models with and without covariates. These designs minimize the variance of the treatment contrast under certain cost constraints. A pseudo Bayesian design approach is advocated that integrates and averages the objective function over a prior distribution of the intracluster correlation coefficient. Theoretical results on the Bayesian criterion are noted when the intracluster correlation follows a uniform distribution. Two data sets based on educational surveys conducted in schools are used to illustrate the proposed methodology.

In this paper we compare the domains of attraction of limit laws of intermediate order statistics under power normalization with those of limit laws of intermediate order statistics under linear normalization. As a result of this comparison, we obtain necessary and sufficient conditions for a univariate distribution function to belong to the domain of attraction for each of the possible limit laws of intermediate order statistics under power normalization.

In this paper, we introduce a stationary first-order integer-valued autoregressive process with geometric–Poisson marginals. The new process allows negative values for the series. Several properties of the process are established. The unknown parameters of the model are estimated using the Yule–Walker method and the asymptotic properties of the estimator are considered. Some numerical results of the estimators are presented with a brief discussion. Possible application of the process is discussed through a real data example.

For estimating a lower restricted parametric function in the framework of Marchand and Strawderman (2006), we show how $(1-\alpha )\times 100\%$ Bayesian credible intervals can be constructed so that the frequentist probability of coverage is no less than $1-\frac{3\alpha }{2}$. As in Marchand and Strawderman (2013), the findings are achieved through the specification of the spending function of the Bayes credible interval and apply to an “equal-tails” modification of the HPD procedure among others. Our results require a logconcave assumption for the distribution of a pivot, and apply to estimating a lower bounded normal mean with known variance, and to further examples include lower bounded scale parameters from Gamma, Weibull, and Fisher distributions, with the latter also applicable to random effects analysis of variance.

The aim of this paper is to propose procedures that test statistical hypotheses locally, that is, assess the validity of a model in a specific domain of the data. In this context, the one and two sample problems will be discussed. The proposed tests are based on local divergences which are defined in such a way as to quantify the divergence between probability distributions locally, in a specific area of the joint domain of the underlined models. The theoretical results are exemplified using simulations and two real datasets.

Integer-valued time series analysis offers various applications in biomedical, financial, and environmental research. However, existing works usually assume no or constant over-dispersion. In this paper, we propose a new model for time series of counts, the autoregressive conditional negative binomial model that has a time-varying conditional autoregressive mean function and heteroskedasticity. The location and scale parameters of the negative binomial distribution are flexible in the proposed set-up, inducing dynamic over-dispersion. We adopt Bayesian methods with a Markov chain Monte Carlo sampling scheme to estimate model parameters and utilize deviance information criterion for model comparison. We conduct simulations to investigate the estimation performance of this sampling scheme for the proposed negative binomial model. To demonstrate the proposed approach in modelling time-varying over-dispersion, we consider two types of criminal incidents recorded by New South Wales (NSW) Police Force in Australia. We also fit the autoregressive conditional Poisson model to these two datasets. Our results demonstrate that the proposed negative binomial model is preferable to the Poisson model.

In this paper, the estimation of the stress–strength reliability $Pr(X>Y)$ based on upper record values is considered when $X$ and $Y$ are independent random variables from a two-parameter bathtub-shaped lifetime distribution with the same shape but different scale parameters. The maximum likelihood estimator (MLE), the approximate Bayes estimator and the exact confidence intervals of stress–strength reliability are obtained when the shape parameter is known. When the shape parameter is unknown, we obtain the MLE, the asymptotic confidence interval and some bootstrap confidence intervals of stress–strength reliability. In this case, we also apply the Gibbs sampling technique to study the Bayesian estimation of stress–strength reliability and the corresponding credible interval. A Monte Carlo simulation study is conducted to investigate and compare the performance of the different proposed methods in this paper. Finally, analysis of a real data set is presented for illustrative purposes.

The odds ratio is the predominant measure of association in $2\times 2$ contingency tables, which, for inferential purposes, is usually considered on the log-scale. Under an information theoretic set-up, it is connected to the Kullback–Leibler divergence. Considering a generalized family of divergences, the $\varphi$ divergence, alternative association measures are derived for $2\times 2$ contingency tables. Their properties are studied and asymptotic inference is developed. For some members of this family, the estimated association measures remain finite in the presence of a sampling zero while for a subset of these members the estimators of these measures have finite variance as well. Special attention is given to the power divergence, which is a parametric family. The role of its parameter $\lambda$, in terms of the asymptotic confidence intervals’ coverage probability and average relative length, is further discussed. In special probability table structures, for which the performance of the asymptotic confidence intervals for the classical log odds ratio is poor, the measure corresponding to $\lambda =1/3$ is suggested as an alternative.