Computational Statistics最新文献

Many multiple test procedures, which control the false discovery rate, have been developed to identify some cases (e.g. genes) showing statistically significant difference between two different groups. However, a common issue encountered in some practical data sets is the presence of highly spiky null distributions. Existing methods struggle to control type I error in such cases due to the “inflated false positives," but this problem has not been addressed in previous literature. Our team recently encountered this issue while analyzing SET4 gene deletion data and proposed modeling the null distribution using a scale mixture normal distribution. However, the use of this approach is limited due to strong assumptions on the spiky peak. In this paper, we present a novel multiple test procedure that can be applied to any type of spiky peak data, including situations with no spiky peak or with one or two spiky peaks. Our approach involves truncating the central statistics around 0, which primarily contribute to the null spike, as well as the two tails that may be contaminated by alternative distributions. We refer to this method as the “double truncation method." After applying double truncation, we estimate the null density using the doubly truncated maximum likelihood estimator. We demonstrate numerically that our proposed method effectively controls the false discovery rate at the desired level using simulated data. Furthermore, we apply our method to two real data sets, namely the SET protein data and peony data.

Kernel estimators of density function and hazard rate function are very important in nonparametric statistics. The paper aims to investigate the uniformly strong representations and the rates of uniformly strong consistency for kernel smoothing density and hazard rate function estimation with censored widely orthant dependent data based on the Kaplan–Meier estimator. Under some mild conditions, the rates of the remainder term and strong consistency are shown to be (Obig (sqrt{log (ng(n))/big (nb_{n}^{2}big )}big )~a.s.) and (Obig (sqrt{log (ng(n))/big (nb_{n}^{2}big )}big )+Obig (b_{n}^{2}big )~a.s.), respectively, where g(n) are the dominating coefficients of widely orthant dependent random variables. Some numerical simulations and a real data analysis are also presented to confirm the theoretical results based on finite sample performances.

In this paper we propose a new method for probabilistic forecasting of electricity prices. It is based on averaging point forecasts from different models combined with expectile regression. We show that deriving the predicted distribution in terms of expectiles, might be in some cases advantageous to the commonly used quantiles. We apply the proposed method to the day-ahead electricity prices from the German market and compare its accuracy with the Quantile Regression Averaging method and quantile- as well as expectile-based historical simulation. The obtained results indicate that using the expectile regression improves the accuracy of the probabilistic forecasts of electricity prices, but a variance stabilizing transformation should be applied prior to modelling.

The projection predictive variable selection is a decision-theoretically justified Bayesian variable selection approach achieving an outstanding trade-off between predictive performance and sparsity. Its projection problem is not easy to solve in general because it is based on the Kullback–Leibler divergence from a restricted posterior predictive distribution of the so-called reference model to the parameter-conditional predictive distribution of a candidate model. Previous work showed how this projection problem can be solved for response families employed in generalized linear models and how an approximate latent-space approach can be used for many other response families. Here, we present an exact projection method for all response families with discrete and finite support, called the augmented-data projection. A simulation study for an ordinal response family shows that the proposed method performs better than or similarly to the previously proposed approximate latent-space projection. The cost of the slightly better performance of the augmented-data projection is a substantial increase in runtime. Thus, if the augmented-data projection’s runtime is too high, we recommend the latent projection in the early phase of the model-building workflow and the augmented-data projection for final results. The ordinal response family from our simulation study is supported by both projection methods, but we also include a real-world cancer subtyping example with a nominal response family, a case that is not supported by the latent projection.

Several zero-augmented models exist for estimation involving outcomes with large numbers of zero. Two of such models for handling count endpoints are zero-inflated and hurdle regression models. In this article, we apply the extreme ranked set sampling (ERSS) scheme in estimation using zero-inflated and hurdle regression models. We provide theoretical derivations showing superiority of ERSS compared to simple random sampling (SRS) using these zero-augmented models. A simulation study is also conducted to compare the efficiency of ERSS to SRS and lastly, we illustrate applications with real data sets.

The scatter halfspace depth (sHD) is an extension of the location halfspace (also called Tukey) depth that is applicable in the nonparametric analysis of scatter. Using sHD, it is possible to define minimax optimal robust scatter estimators for multivariate data. The problem of exact computation of sHD for data of dimension (d ge 2) has, however, not been addressed in the literature. We develop an exact algorithm for the computation of sHD in any dimension d and implement it efficiently for any dimension (d ge 1). Since the exact computation of sHD is slow especially for higher dimensions, we also propose two fast approximate algorithms. All our programs are freely available in the R package scatterdepth.

We consider mixtures of longitudinal trajectories, where one trajectory contains measurements over time of the variable of interest for one individual and each individual belongs to one cluster. The number of clusters as well as individual cluster memberships are unknown and must be inferred. We propose an original Bayesian clustering framework that allows us to obtain an exact finite-sample model selection criterion for selecting the number of clusters. Our finite-sample approach is more flexible and parsimonious than asymptotic alternatives such as Bayesian information criterion or integrated classification likelihood criterion in the choice of the number of clusters. Moreover, our approach has other desirable qualities: (i) it keeps the computational effort of the clustering algorithm under control and (ii) it generalizes to several families of regression mixture models, from linear to purely non-parametric. We test our method on simulated datasets as well as on a real world dataset from the Alzheimer’s disease neuroimaging initative database.

Finite mixture of Gaussians are often used to classify two- (units and variables) or three- (units, variables and occasions) way data. However, two issues arise: model complexity and capturing the true cluster structure. Indeed, a large number of variables and/or occasions implies a large number of model parameters; while the existence of noise variables (and/or occasions) could mask the true cluster structure. The approach adopted in the present paper is to reduce the number of model parameters by identifying a sub-space containing the information needed to classify the observations. This should also help in identifying noise variables and/or occasions. The maximum likelihood model estimation is carried out through an EM-like algorithm. The effectiveness of the proposal is assessed through a simulation study and an application to real data.

Longitudinal studies have been conducted in various fields, including medicine, economics and the social sciences. In this paper, we focus on longitudinal ordinal data. Since the longitudinal data are collected over time, repeated outcomes within each subject may be serially correlated. To address both the within-subjects serial correlation and the specific variance between subjects, we propose a Bayesian cumulative probit random effects model for the analysis of longitudinal ordinal data. The hypersphere decomposition approach is employed to overcome the positive definiteness constraint and high-dimensionality of the correlation matrix. Additionally, we present a hybrid Gibbs/Metropolis-Hastings algorithm to efficiently generate cutoff points from truncated normal distributions, thereby expediting the convergence of the Markov Chain Monte Carlo (MCMC) algorithm. The performance and robustness of our proposed methodology under misspecified correlation matrices are demonstrated through simulation studies under complete data, missing completely at random (MCAR), and missing at random (MAR). We apply the proposed approach to analyze two sets of actual ordinal data: the arthritis dataset and the lung cancer dataset. To facilitate the implementation of our method, we have developed BayesRGMM, an open-source R package available on CRAN, accompanied by comprehensive documentation and source code accessible at https://github.com/kuojunglee/BayesRGMM/.

The main objective of this paper is to discuss selected computational aspects of robust estimation in the linear model with the emphasis on R-estimators. We focus on numerical algorithms and computational efficiency rather than on statistical properties. In addition, we formulate some algorithmic properties that a “good” method for R-estimators is expected to satisfy and show how to satisfy them using the currently available algorithms. We illustrate both good and bad properties of the existing algorithms. We propose two-stage methods to minimize the effect of the bad properties. Finally we justify a challenge for new approaches based on interior-point methods in optimization.