THE COVERAGE PROBABILITY OF CONFIDENCE INTERVALS IN 2r FACTORIAL EXPERIMENTS AFTER PRELIMINARY HYPOTHESIS TESTING

We consider a 2^r factorial experiment with at least two replicates. Our aim is to find a confidence interval for θ, a specified linear combination of the regression parameters (for the model written as a regression, with factor levels coded as ?1 and 1). We suppose that preliminary hypothesis tests are carried out sequentially, beginning with the rth‐order interaction. After these preliminary hypothesis tests, a confidence interval for θ with nominal coverage 1 ?α is constructed under the assumption that the selected model had been given to us a priori. We describe a new efficient Monte Carlo method, which employs conditioning for variance reduction, for estimating the minimum coverage probability of the resulting confidence interval. The application of this method is demonstrated in the context of a 2³ factorial experiment with two replicates and a particular contrast θ of interest. The preliminary hypothesis tests consist of the following two‐step procedure. We first test the null hypothesis that the third‐order interaction is zero against the alternative hypothesis that it is non‐zero. If this null hypothesis is accepted, we assume that this interaction is zero and proceed to the second step; otherwise, we stop. In the second step, for each of the second‐order interactions we test the null hypothesis that the interaction is zero against the alternative hypothesis that it is non‐zero. If this null hypothesis is accepted, we assume that this interaction is zero. The resulting confidence interval, with nominal coverage probability 0.95, has a minimum coverage probability that is, to a good approximation, 0.464. This shows that this confidence interval is completely inadequate.     

A simulation study of estimators for generalized linear measurement error models

The purpose of this paper is to examine the properties of several bias-corrected estimators for generalized linear measurement error models, along with the naive estimator, in some special settings. In particular, we consider logistic regression, poisson regression and exponential-gamma models where the covariates are subject to measurement error. Monte Carlo experiments are conducted to compare the relative performance of the estimators in terms of several criteria. The results indicate that the naive estimator of slope is biased towards zero by a factor increasing with the magnitude of slope and measurement error as well as the sample size. It is found that none of the biased-corrected estimators always outperforms the others, and that their small sample properties typically depend on the underlying model assumptions.     

Density modification-based reliability sensitivity analysis

Sensitivity analysis (SA) of a numerical model, for instance simulating physical phenomena, is useful to quantify the influence of the inputs on the model responses. This paper proposes a new sensitivity index, based upon the modification of the probability density function (pdf) of the random inputs, when the quantity of interest is a failure probability (probability that a model output exceeds a given threshold). An input is considered influential if the input pdf modification leads to a broad change in the failure probability. These sensitivity indices can be computed using the sole set of simulations that has already been used to estimate the failure probability, thus limiting the number of calls to the numerical model. In the case of a Monte Carlo sample, asymptotical properties of the indices are derived. Based on Kullback–Leibler divergence, several types of input perturbations are introduced. The relevance of this new SA method is analysed through three case studies.     

A sequential split-and-conquer approach for the analysis of big dependent data in computer experiments

Massive correlated data with many inputs are often generated from computer experiments to study complex systems. The Gaussian process (GP) model is a widely used tool for the analysis of computer experiments. Although GPs provide a simple and effective approximation to computer experiments, two critical issues remain unresolved. One is the computational issue in GP estimation and prediction where intensive manipulations of a large correlation matrix are required. For a large sample size and with a large number of variables, this task is often unstable or infeasible. The other issue is how to improve the naive plug-in predictive distribution which is known to underestimate the uncertainty. In this article, we introduce a unified framework that can tackle both issues simultaneously. It consists of a sequential split-and-conquer procedure, an information combining technique using confidence distributions (CD), and a frequentist predictive distribution based on the combined CD. It is shown that the proposed method maintains the same asymptotic efficiency as the conventional likelihood inference under mild conditions, but dramatically reduces the computation in both estimation and prediction. The predictive distribution contains comprehensive information for inference and provides a better quantification of predictive uncertainty as compared with the plug-in approach. Simulations are conducted to compare the estimation and prediction accuracy with some existing methods, and the computational advantage of the proposed method is also illustrated. The proposed method is demonstrated by a real data example based on tens of thousands of computer experiments generated from a computational fluid dynamic simulator.     

Optimal design for discrete choice experiments

In this paper we derive locally optimal designs for discrete choice experiments. As in Kanninen (2002) we consider a multinomial logistic model, which contains various qualitative attributes as well as a quantitative one, which may range over a sufficiently large interval. The derived optimal designs improve upon those given in the literature, but have the feature that every choice set contains alternatives, which coincide in all but the quantitative attributes. The multinomial logistic model will then lead to a response behavior, which is apparently unrealistic.     

Sequential tuning of complex computer models

We propose a method that uses a sequential design instead of a space filling design for estimating tuning parameters of a complex computer model. The goal is to bring the computer model output closer to the real system output. The method fits separate Gaussian process (GP) models to the available data from the physical experiment and the computer experiment and minimizes the discrepancy between the predictions from the GP models to obtain estimates of the tuning parameters. A criterion based on the discrepancy between the predictions from the two GP models and the standard error of prediction for the computer experiment output is then used to obtain a design point for the next run of the computer experiment. The tuning parameters are re-estimated using the augmented data set. The steps are repeated until the budget for the computer experiment data is exhausted. Simulation studies show that the proposed method performs better in bringing a computer model closer to the real system than methods that use a space filling design.     

Testing homogeneity in a heteroscedastic contaminated normal mixture

Large-scale simultaneous hypothesis testing appears in many areas. A well-known inference method is to control the false discovery rate. One popular approach is to model the z-scores derived from the individual t-tests and then use this model to control the false discovery rate. We propose a heteroscedastic contaminated normal mixture to describe the distribution of z-scores and design an EM-test for testing homogeneity in this class of mixture models. The proposed EM-test can be used to investigate whether a collection of z-scores has arisen from a single normal distribution or whether a heteroscedastic contaminated normal mixture is more appropriate. We show that the EM-test statistic has a shifted mixture of chi-squared limiting distribution. Simulation results show that the proposed testing procedure has accurate type-I error and significantly larger power than its competitors under a variety of model specifications. A real-data example is analysed to exemplify the application of the proposed method.     

The parametric sensitivity of dimethylsulfide flux in the southern ocean

A screening method proposed by Morris has been applied to a model of the oceanic production of dimethylsulfide (DMS), an important sulphur-containing gas which plays an important role in climate regulation. The model is usually refered as the GMSK (Gabric-Murray-Stone-Kohl) model and it is described by Gabricet at. (1993,1996). The aim of the experiment is to quantify the relative importance of some pre-selected input factors in determining the predicted value of a model state variable: the DMS annual flux.

The method is based on an experimental plan, composed of individually randomised one-factor-at..a-time designs, with the purpose of collecting random samples from the distribution of “elementary effects”, denoted by F, associated with each input parameter.

A large (absolute) measure of central tendency for F indicates an input with an overall influence on the output. A large standard deviation indicates an input whose influence is highly dependent on the values of the other inputs -i.e., one involved in interactions or whose effect is nonlinear. In particular, estimates of the means and standard deviations of these distributions will be used as indicators of which inputs should be considered important. Five of the total number of factors involved in the GMSK model have been examined. The selected factors are: the maximum nitrate uptake by phytoplankton (k23 in the model), the algal S:N (sulphur to nitrogen) ratio (gamma in the model), the parameter involved in the link phytoplankton-protozoa (k3), the initial mixed layer nitrogen concentration (DIN), and the sea surface temperature (SST). The experimental results show that none of the parameters studied has both mean and standard deviation close to zero, that is they all have a direct or indirect effect on the output, confirming that the DMS flux prediction is indeed sensitive to these parameters.     

ALL-BIAS DESIGNS FOR POLYNOMIAL SPLINE REGRESSION MODELS

Polynomial spline regression models of low degree have proved useful in modeling responses from designed experiments in science and engineering when simple polynomial models are inadequate. Where there is uncertainty in the number and location of the knots, or breakpoints, of the spline, then designs that minimize the systematic errors resulting from model misspecification may be appropriate. This paper gives a method for constructing such all‐bias designs for a single variable spline when the distinct knots in the assumed and true models come from some specified set. A class of designs is defined in terms of the inter‐knot intervals and sufficient conditions are obtained for a design within this class to be all‐bias under linear, quadratic and cubic spline models. An example of the construction of all‐bias designs is given.     

The Mean Squared Error of the Generalized Ridge Regression Estimator and the Orientation of β

Newhouse and Oman (1971) identified the orientations with respect to the eigenvectors of X'X of the true coefficient vector of the linear regression model for which the ordinary ridge regression estimator performs best and performs worse when mean squared error is the measure of performance. In this paper the corresponding result is derived for generalized ridge regression for two risk functions: mean squared error and mean squared error of prediction.     

A Threshold Dose-Response Model with Random Effects in Teratological Experiments

Abstract

Teratological experiments are controlled dose-response studies in which impregnated animals are randomly assigned to various exposure levels of a toxic substance. Subsequently, both continuous and discrete responses are recorded on the litters of fetuses that these animals produce. Discrete responses are usually binary in nature, such as the presence or absence of some fetal anomaly. This clustered binary data usually exhibits over-dispersion (or under-dispersion), which can be interpreted as either variation between litter response probabilities or intralitter correlation. To model the correlation and/or variation, the beta-binomial distribution has been assumed for the number of positive fetal responses within a litter. Although the mean of the beta-binomial model has been linked to dose-response functions, in terms of measuring over-dispersion, it may be a restrictive method in modeling data from teratological studies. Also for certain toxins, a threshold effect has been observed in the dose-response pattern of the data. We propose to incorporate a random effect into a general threshold dose-response model to account for the variation in responses, while at the same time estimating the threshold effect. We fit this model to a well-known data set in the field of teratology. Simulation studies are performed to assess the validity of the random effects threshold model in these types of studies.     

Minimax D-optimal designs of contingent valuation experiments: willingness to pay for environmentally friendly clothes

This paper demonstrates how to plan a contingent valuation experiment to assess the value of ecologically produced clothes. First, an appropriate statistical model (the trinomial spike model) that describes the probability that a randomly selected individual will accept any positive bid, and if so, will accept the bid A, is defined. Secondly, an optimization criterion that is a function of the variances of the parameter estimators is chosen. However, the variances of the parameter estimators in this model depend on the true parameter values. Pilot study data are therefore used to obtain estimates of the parameter values and a locally optimal design is found. Because this design is only optimal given that the estimated parameter values are correct, a design that minimizes the maximum of the criterion function over a plausable parameter region (i.e. a minimax design) is then found.     

A-optimal designs for optimum mixture in an additive quadratic mixture model

In the analysis of experiments with mixtures, quadratic models have been widely used. The optimum designs for the estimation of optimum mixing proportions in a quadratic mixture model have been studied by Pal and Mandal [Optimum designs for optimum mixtures. Statist Probab Lett. 2006;76:1369–1379] and Mandal et al. [Optimum mixture designs: a pseudo-Bayesian approach. J Ind Soc Agric Stat. 2008;62(2):174–182; Optimum mixture designs under constraints on mixing components. Statist Appl. 2008;6(1&2) (New Series): 189–205], using a pseudo-Bayesian approach. In this paper, a similar approach has been employed to obtain the A-optimal designs for the estimation of optimum proportions in an additive quadratic mixture model, proposed by Darroch and Waller [Additivity and interaction in three-component experiments with mixture. Biometrika. 1985;72:153–163], when the number of components is 3, 4 and 5. It has been shown that the vertices of the simplex are necessarily the support points of the optimum design, and the other support points include barycentres of depth at most 2.     

基于BP神经网络的信息商品价格预测模型

从信息商品价格的特点和BP神经网络的功能出发,借助样本训练集和改进的BP算法提出了基于BP神经网络的信息商品价格预测模型,并对模型的泛化精度进行了讨论,最后进行了仿真实验.     

Bayesian Analysis of Two-Level Fractional Factorial Experiments with Non-Normal Responses

An intractable issue on screening experiments is to identify significant effects and to select the best model when the number of factors is large, especially for fractional factorial experiments with non-normal responses. In such cases, a three-stage Bayesian approach based on generalized linear models (GLMs) is proposed to identify which effects should be included in the target model where the principles of effect sparsity, hierarchy, and heredity are successfully considered. Three simulation experiments with non-normal responses which follow Poisson, binomial, and gamma distributions are presented to illustrate the performance of the proposed approach.     

农户对夏玉米品种性状属性的支付意愿研究——基于选择实验的分析

基于河南和山东农户调研数据,采用选择实验方法,应用随机参数模型与潜在类别模型分析农户对夏玉米品种的生育期、与干旱相关的产量特征、籽粒品质及穗位整齐度性状属性的偏好与支付意愿。研究表明,农户对玉米品种性状偏好具有异质性,根据这种异质性可以将农户分为穗齐偏好农户和产量稳定偏好农户,但追求产量目标的农户少于追求轻简目标的农户;农户对玉米品种的穗位整齐度的支付意愿最高,其次为玉米的产量特征和生育期,对玉米籽粒的品质属性支付意愿最低。因此,培育耐旱稳产型、适宜机械化操作的穗齐型且生育期较短的"轻简型"玉米品种是中国玉米育种的主要方向。     

Solving Noisy ICA Using Multivariate Wavelet Denoising with an Application to Noisy Latent Variables Regression

A novel approach to solve the independent component analysis (ICA) model in the presence of noise is proposed. We use wavelets as natural denoising tools to solve the noisy ICA model. To do this, we use a multivariate wavelet denoising algorithm allowing spatial and temporal dependency. We propose also using a statistical approach, named nested design of experiments, to select the parameters such as wavelet family and thresholding type. This technique helps us to select more convenient combination of the parameters. This approach could be extended to many other problems in which one needs to choose parameters between many choices. The performance of the proposed method is illustrated on the simulated data and promising results are obtained. Also, the suggested method applied in latent variables regression in the presence of noise on real data. The good results confirm the ability of multivariate wavelet denoising to solving noisy ICA.     

Optimum Mixture Design for the Detection of Synergistic Effects

In mixture experiments, optimal designs for the estimation of parameters, both linear and non-linear, have been discussed by several authors. Optimal designs for the estimation of a subset of parameters have also been investigated. However, designs for testing the effects of certain factors and interactions have been studied only in the context of response surface models. In this article, we attempt to find the optimum design for testing the presence of synergistic effects in a mixture model. The classical F-test has been considered and the optimum design has been obtained so as to maximize the power of the test. It is observed that the barycenters are necessarily the support points of the trace-optimal design.     

Inference for a Simple Step-Stress Model with Type-I Censoring and Lognormally Distributed Lifetimes

Accelerated life-testing (ALT) is a very useful technique for examining the reliability of highly reliable products. It allows the experimenter to obtain failure data more quickly at increased stress levels than under normal operating conditions. A step-stress model is one special class of ALT, and in this article we consider a simple step-stress model under the cumulative exposure model with lognormally distributed lifetimes in the presence of Type-I censoring. We then discuss inferential methods for the unknown parameters of the model by the maximum likelihood estimation method. Some numerical methods, such as the Newton–Raphson and quasi-Newton methods, are discussed for solving the corresponding non-linear likelihood equations. Next, we discuss the construction of confidence intervals for the unknown parameters based on (i) the asymptotic normality of the maximum likelihood estimators (MLEs), and (ii) parametric bootstrap resampling technique. A Monte Carlo simulation study is carried out to examine the performance of these methods of inference. Finally, a numerical example is presented in order to illustrate all the methods of inference developed here.     

A Diagnostic Test for the Multinomial Logit Model

I propose a Lagrange multiplier test for the multinomial logit model against the dogit model (Gaudry and Dagenais 1979) as the alternative hypothesis. In view of the well-known drawback of the restrictive property of independence from irrelevant alternatives implied by the multinomial logit model, a specification test has much to recommend it. Finite sample properties of the test are studied using a Monte Carlo experiment, and the test's power against the nested multinomial logit model and the multinomial probit model is investigated. The test is found to be sensitive to the values of the regression parameters of the linear random utility function.