Logistic回归模型的统计诊断

统计诊断的主要任务就是通过诊断统计量检测已知观测数据在用既定模型拟合时的合理性,主要是找出数据当中的异常点或强影响点。本文主要研究Logostic回归模型的诊断统计量和诊断统计图。用牛顿迭代法给出Logistic回归模型的极大似然估计值,根据扰动模型得到传统的诊断统计量,结合残差、杠杆值和系数变化三者构造新的诊断统计量,绘制新的诊断统计图,通过模拟研究说明新的诊断统计量的有效性,最后用一个实际案例说明新的诊断方法的应用并进一步验证其优越性。     

One-Step Estimator Paths for Concave Regularization

The statistics literature of the past 15 years has established many favorable properties for sparse diminishing-bias regularization: techniques that can roughly be understood as providing estimation under penalty functions spanning the range of concavity between ?₀ and ?₁ norms. However, lasso ?₁-regularized estimation remains the standard tool for industrial Big Data applications because of its minimal computational cost and the presence of easy-to-apply rules for penalty selection. In response, this article proposes a simple new algorithm framework that requires no more computation than a lasso path: the path of one-step estimators (POSE) does ?₁ penalized regression estimation on a grid of decreasing penalties, but adapts coefficient-specific weights to decrease as a function of the coefficient estimated in the previous path step. This provides sparse diminishing-bias regularization at no extra cost over the fastest lasso algorithms. Moreover, our gamma lasso implementation of POSE is accompanied by a reliable heuristic for the fit degrees of freedom, so that standard information criteria can be applied in penalty selection. We also provide novel results on the distance between weighted-?₁ and ?₀ penalized predictors; this allows us to build intuition about POSE and other diminishing-bias regularization schemes. The methods and results are illustrated in extensive simulations and in application of logistic regression to evaluating the performance of hockey players. Supplementary materials for this article are available online.     

Odd length for even hyperoctahedral groups and signed generating functions

We define a new statistic on the even hyperoctahedral groups which is a natural analogue of the odd length statistic recently defined and studied on Coxeter groups of types $A$ and $B$. We compute the signed (by length) generating function of this statistic over the whole group and over its maximal and some other quotients and show that it always factors nicely. We also present some conjectures.     

Elastic properties of model random three-dimensional open-cell solids

Most cellular solids are random materials, while practically all theoretical structure-property relations are for periodic models. To generate theoretical results for random models the finite element method (FEM) was used to study the elastic properties of open-cell solids. We have computed the density (ρ) and microstructure dependence of the Young's modulus (E) and Poisson's ratio (ν) for four different isotropic random models. The models were based on Voronoi tessellations, level-cut Gaussian random fields, and nearest neighbour node-bond rules. These models were chosen to broadly represent the structure of foamed solids and other (non-foamed) cellular materials. At low densities, the Young's modulus can be described by the relation E∝ρⁿ. The exponent n and constant of proportionality depend on microstructure. We find 1.3<n<3, indicating a more complex dependence than indicated by periodic cell theories, which predict n=1 or 2. The observed variance in the exponent was found to be consistent with experimental data. At low densities we found that ν≈0.25 for three of the four models studied. In contrast, the Voronoi tessellation, which is a common model of foams, became approximately incompressible (ν≈0.5). This behaviour is not commonly observed experimentally. Our studies showed the result was robust to polydispersity and that a relatively large number (15%) of the bonds must be broken to significantly reduce the low-density Poission's ratio to ν≈0.33.     

Use of a tree‐structured hierarchical model for estimation of location and uncertainty in multivariate spatial data

Analysis and modeling of spatial data are of considerable interest in many applications. However, the prediction of geographical features from a set of chemical measurements on a set of geographically distinct samples has never been explored. We report a new, tree‐structured hierarchical model for the estimation of geographical location of spatially distributed samples from their chemical measurements. The tree‐structured hierarchical modeling used in this study involves a set of geographic regions stored in a hierarchical tree structure, with each nonterminal node representing a classifier and each terminal node representing a regression model. Once the tree‐structured model is constructed, given a sample with only chemical measurements available, the predicted regional location of the sample is gradually restricted as it is passed through a series of classification steps. The geographic location of the sample can be predicted using a regression model within the terminal subregion. We show that the tree‐structured modeling approach provides reasonable estimates of geographical region and geographic location for surface water samples taken across the entire USA. Further, the location uncertainty, an estimate of a probability that a test sample could be located within a pre‐estimated, joint prediction interval that is much smaller than the terminal subregion, can also be assessed. Copyright © 2014 John Wiley & Sons, Ltd.     

Determining equilibrium constants for dimerization reactions from molecular dynamics simulations

With today's available computer power, free energy calculations from equilibrium molecular dynamics simulations "via counting" become feasible for an increasing number of reactions. An example is the dimerization reaction of transmembrane alpha-helices. If an extended simulation of the two helices covers sufficiently many dimerization and dissociation events, their binding free energy is readily derived from the fraction of time during which the two helices are observed in dimeric form. Exactly how the correct value for the free energy is to be calculated, however, is unclear, and indeed several different and contradictory approaches have been used. In particular, results obtained via Boltzmann statistics differ from those determined via the law of mass action. Here, we develop a theory that resolves this discrepancy. We show that for simulation systems containing two molecules, the dimerization free energy is given by a formula of the form ΔG ∝ ln(P(1) /P(0) ). Our theory is also applicable to high concentrations that typically have to be used in molecular dynamics simulations to keep the simulation system small, where the textbook dilute approximations fail. It also covers simulations with an arbitrary number of monomers and dimers and provides rigorous error estimates. Comparison with test simulations of a simple Lennard Jones system with various particle numbers as well as with reference free energy values obtained from radial distribution functions show full agreement for both binding free energies and dimerization statistics.     

Multivariate statistical identification of human bladder carcinomas using ambient ionization imaging mass spectrometry

Diagnosis of human bladder cancer in untreated tissue sections is achieved by using imaging data from desorption electrospray ionization mass spectrometry (DESI-MS) combined with multivariate statistical analysis. We use the distinctive DESI-MS glycerophospholipid (GP) mass spectral profiles to visually characterize and formally classify twenty pairs (40 tissue samples) of human cancerous and adjacent normal bladder tissue samples. The individual ion images derived from the acquired profiles correlate with standard histological hematoxylin and eosin (H&E)-stained serial sections. The profiles allow us to classify the disease status of the tissue samples with high accuracy as judged by reference histological data. To achieve this, the data from the twenty pairs were divided into a training set and a validation set. Spectra from the tumor and normal regions of each of the tissue sections in the training set were used for orthogonal projection to latent structures (O-PLS) treated partial least-square discriminate analysis (PLS-DA). This predictive model was then validated by using the validation set and showed a 5% error rate for classification and a misclassification rate of 12%. It was also used to create synthetic images of the tissue sections showing pixel-by-pixel disease classification of the tissue and these data agreed well with the independent classification that uses histological data by a certified pathologist. This represents the first application of multivariate statistical methods for classification by ambient ionization although these methods have been applied previously to other MS imaging methods. The results are encouraging in terms of the development of a method that could be utilized in a clinical setting through visualization and diagnosis of intact tissue.     

A Laplace transform method for order statistics from nonidentical random variables and its application in Phase-type distribution

In this paper, we derive a method for obtaining the Laplace transform of order statistics (o.s.) arising from general independent nonidentically distributed random variables (r.v.’s). A survey of the most important properties, applications and the o.s. of a Phase-type (PH) distribution are also presented. Two illustrative examples are provided.     

一种基于混合策略的自适应多输出高斯过程响应面法

针对工程结构数值模拟常常是矢量与函数(矢量函数)混合输出的特点,提出一种基于混合策略的含索引变量的矢量高斯过程响应面法,同时提出一种设计点自适应加密方法进一步提高效率。算例显示:本文方法能够优先拾取预测误差最大的点作为加密点;真实误差全部都落在最大可能误差范围之内,最大误差1.4%~1.6%远小于预设误差限5%,验证了本文方法的有效性。