A gradient Markov chain Monte Carlo algorithm for computing multivariate maximum likelihood estimates and posterior distributions: mixture dose-response assessment

Multivariate probability distributions, such as may be used for mixture dose‐response assessment, are typically highly parameterized and difficult to fit to available data. However, such distributions may be useful in analyzing the large electronic data sets becoming available, such as dose‐response biomarker and genetic information. In this article, a new two‐stage computational approach is introduced for estimating multivariate distributions and addressing parameter uncertainty. The proposed first stage comprises a gradient Markov chain Monte Carlo (GMCMC) technique to find Bayesian posterior mode estimates (PMEs) of parameters, equivalent to maximum likelihood estimates (MLEs) in the absence of subjective information. In the second stage, these estimates are used to initialize a Markov chain Monte Carlo (MCMC) simulation, replacing the conventional burn‐in period to allow convergent simulation of the full joint Bayesian posterior distribution and the corresponding unconditional multivariate distribution (not conditional on uncertain parameter values). When the distribution of parameter uncertainty is such a Bayesian posterior, the unconditional distribution is termed predictive. The method is demonstrated by finding conditional and unconditional versions of the recently proposed emergent dose‐response function (DRF). Results are shown for the five‐parameter common‐mode and seven‐parameter dissimilar‐mode models, based on published data for eight benzene–toluene dose pairs. The common mode conditional DRF is obtained with a 21‐fold reduction in data requirement versus MCMC. Example common‐mode unconditional DRFs are then found using synthetic data, showing a 71% reduction in required data. The approach is further demonstrated for a PCB 126‐PCB 153 mixture. Applicability is analyzed and discussed. Matlab^® computer programs are provided.     

Bayesian Network Model with Application to Smart Power Semiconductor Lifetime Data

In this article, Bayesian networks are used to model semiconductor lifetime data obtained from a cyclic stress test system. The data of interest are a mixture of log‐normal distributions, representing two dominant physical failure mechanisms. Moreover, the data can be censored due to limited test resources. For a better understanding of the complex lifetime behavior, interactions between test settings, geometric designs, material properties, and physical parameters of the semiconductor device are modeled by a Bayesian network. Statistical toolboxes in MATLAB® have been extended and applied to find the best structure of the Bayesian network and to perform parameter learning. Due to censored observations Markov chain Monte Carlo (MCMC) simulations are employed to determine the posterior distributions. For model selection the automatic relevance determination (ARD) algorithm and goodness‐of‐fit criteria such as marginal likelihoods, Bayes factors, posterior predictive density distributions, and sum of squared errors of prediction (SSEP) are applied and evaluated. The results indicate that the application of Bayesian networks to semiconductor reliability provides useful information about the interactions between the significant covariates and serves as a reliable alternative to currently applied methods.     

Sign Restrictions,Structural Vector Autoregressions,and Useful Prior Information

This paper makes the following original contributions to the literature. (i) We develop a simpler analytical characterization and numerical algorithm for Bayesian inference in structural vector autoregressions (VARs) that can be used for models that are overidentified, just‐identified, or underidentified. (ii) We analyze the asymptotic properties of Bayesian inference and show that in the underidentified case, the asymptotic posterior distribution of contemporaneous coefficients in an n‐variable VAR is confined to the set of values that orthogonalize the population variance–covariance matrix of ordinary least squares residuals, with the height of the posterior proportional to the height of the prior at any point within that set. For example, in a bivariate VAR for supply and demand identified solely by sign restrictions, if the population correlation between the VAR residuals is positive, then even if one has available an infinite sample of data, any inference about the demand elasticity is coming exclusively from the prior distribution. (iii) We provide analytical characterizations of the informative prior distributions for impulse‐response functions that are implicit in the traditional sign‐restriction approach to VARs, and we note, as a special case of result (ii), that the influence of these priors does not vanish asymptotically. (iv) We illustrate how Bayesian inference with informative priors can be both a strict generalization and an unambiguous improvement over frequentist inference in just‐identified models. (v) We propose that researchers need to explicitly acknowledge and defend the role of prior beliefs in influencing structural conclusions and we illustrate how this could be done using a simple model of the U.S. labor market.     

Harnessing the Theoretical Foundations of the Exponential and Beta‐Poisson Dose‐Response Models to Quantify Parameter Uncertainty Using Markov Chain Monte Carlo

Dose‐response models are the essential link between exposure assessment and computed risk values in quantitative microbial risk assessment, yet the uncertainty that is inherent to computed risks because the dose‐response model parameters are estimated using limited epidemiological data is rarely quantified. Second‐order risk characterization approaches incorporating uncertainty in dose‐response model parameters can provide more complete information to decisionmakers by separating variability and uncertainty to quantify the uncertainty in computed risks. Therefore, the objective of this work is to develop procedures to sample from posterior distributions describing uncertainty in the parameters of exponential and beta‐Poisson dose‐response models using Bayes's theorem and Markov Chain Monte Carlo (in OpenBUGS). The theoretical origins of the beta‐Poisson dose‐response model are used to identify a decomposed version of the model that enables Bayesian analysis without the need to evaluate Kummer confluent hypergeometric functions. Herein, it is also established that the beta distribution in the beta‐Poisson dose‐response model cannot address variation among individual pathogens, criteria to validate use of the conventional approximation to the beta‐Poisson model are proposed, and simple algorithms to evaluate actual beta‐Poisson probabilities of infection are investigated. The developed MCMC procedures are applied to analysis of a case study data set, and it is demonstrated that an important region of the posterior distribution of the beta‐Poisson dose‐response model parameters is attributable to the absence of low‐dose data. This region includes beta‐Poisson models for which the conventional approximation is especially invalid and in which many beta distributions have an extreme shape with questionable plausibility.     

A Systematic Approach to Determining the Identifiability of Multistage Carcinogenesis Models

Multistage clonal expansion (MSCE) models of carcinogenesis are continuous‐time Markov process models often used to relate cancer incidence to biological mechanism. Identifiability analysis determines what model parameter combinations can, theoretically, be estimated from given data. We use a systematic approach, based on differential algebra methods traditionally used for deterministic ordinary differential equation (ODE) models, to determine identifiable combinations for a generalized subclass of MSCE models with any number of preinitation stages and one clonal expansion. Additionally, we determine the identifiable combinations of the generalized MSCE model with up to four clonal expansion stages, and conjecture the results for any number of clonal expansion stages. The results improve upon previous work in a number of ways and provide a framework to find the identifiable combinations for further variations on the MSCE models. Finally, our approach, which takes advantage of the Kolmogorov backward equations for the probability generating functions of the Markov process, demonstrates that identifiability methods used in engineering and mathematics for systems of ODEs can be applied to continuous‐time Markov processes.     

Bayesian Estimation of Dynamic Discrete Choice Models

We propose a new methodology for structural estimation of infinite horizon dynamic discrete choice models. We combine the dynamic programming (DP) solution algorithm with the Bayesian Markov chain Monte Carlo algorithm into a single algorithm that solves the DP problem and estimates the parameters simultaneously. As a result, the computational burden of estimating a dynamic model becomes comparable to that of a static model. Another feature of our algorithm is that even though the number of grid points on the state variable is small per solution‐estimation iteration, the number of effective grid points increases with the number of estimation iterations. This is how we help ease the “curse of dimensionality.” We simulate and estimate several versions of a simple model of entry and exit to illustrate our methodology. We also prove that under standard conditions, the parameters converge in probability to the true posterior distribution, regardless of the starting values.     

Lognormal Approximations of Fault Tree Uncertainty Distributions

Fault trees are used in reliability modeling to create logical models of fault combinations that can lead to undesirable events. The output of a fault tree analysis (the top event probability) is expressed in terms of the failure probabilities of basic events that are input to the model. Typically, the basic event probabilities are not known exactly, but are modeled as probability distributions: therefore, the top event probability is also represented as an uncertainty distribution. Monte Carlo methods are generally used for evaluating the uncertainty distribution, but such calculations are computationally intensive and do not readily reveal the dominant contributors to the uncertainty. In this article, a closed‐form approximation for the fault tree top event uncertainty distribution is developed, which is applicable when the uncertainties in the basic events of the model are lognormally distributed. The results of the approximate method are compared with results from two sampling‐based methods: namely, the Monte Carlo method and the Wilks method based on order statistics. It is shown that the closed‐form expression can provide a reasonable approximation to results obtained by Monte Carlo sampling, without incurring the computational expense. The Wilks method is found to be a useful means of providing an upper bound for the percentiles of the uncertainty distribution while being computationally inexpensive compared with full Monte Carlo sampling. The lognormal approximation method and Wilks’s method appear attractive, practical alternatives for the evaluation of uncertainty in the output of fault trees and similar multilinear models.     

Methods for Assessing Uncertainty in Fundamental Assumptions and Associated Models for Cancer Risk Assessment

The distributional approach for uncertainty analysis in cancer risk assessment is reviewed and extended. The method considers a combination of bioassay study results, targeted experiments, and expert judgment regarding biological mechanisms to predict a probability distribution for uncertain cancer risks. Probabilities are assigned to alternative model components, including the determination of human carcinogenicity, mode of action, the dosimetry measure for exposure, the mathematical form of the dose‐response relationship, the experimental data set(s) used to fit the relationship, and the formula used for interspecies extrapolation. Alternative software platforms for implementing the method are considered, including Bayesian belief networks (BBNs) that facilitate assignment of prior probabilities, specification of relationships among model components, and identification of all output nodes on the probability tree. The method is demonstrated using the application of Evans, Sielken, and co‐workers for predicting cancer risk from formaldehyde inhalation exposure. Uncertainty distributions are derived for maximum likelihood estimate (MLE) and 95th percentile upper confidence limit (UCL) unit cancer risk estimates, and the effects of resolving selected model uncertainties on these distributions are demonstrated, considering both perfect and partial information for these model components. A method for synthesizing the results of multiple mechanistic studies is introduced, considering the assessed sensitivities and selectivities of the studies for their targeted effects. A highly simplified example is presented illustrating assessment of genotoxicity based on studies of DNA damage response caused by naphthalene and its metabolites. The approach can provide a formal mechanism for synthesizing multiple sources of information using a transparent and replicable weight‐of‐evidence procedure.     

基于贝叶斯神经网络短期负荷预测模型

本文提出了基于贝叶斯神经网络(BNN)短期负荷预测模型。根据气象影响因素和电力负荷的样本数据,针对权向量参数的先验分布分别为正态分布和柯西分布两种情况,应用混合蒙特卡洛(HMC)算法学习了BNN的权向量参数。由HMC算法和Laplace算法学习的贝叶斯神经网络以及BP算法学习的传统神经网络分别对4月 (春)、8月 (夏)、10月 (秋)和1月(冬)每月25天的每个整点时刻的负荷进行了预测。这些神经网络的输入层有11个节点,它们分别与每个整点时刻和的气象因素、上一个整点时刻的气象因素和时间变量相对应,输出层只有一个节点,它与负荷变量对应。试验结果表明HMC算法学习的BNN的预测结果的百分比平均绝对误差( MAPE)和平方根平均误差( RSME )取值远远小于由Laplace 算法学习的BNN和BP算法学习的人工神经网络的 MAPE和RMSE。 而且,HMC算法学习的BNN在测试集和训练集上的预测误差MAPE和RMSE的相差很小。 实验结果充分说明HMC算法学习的BNN具有较高的预测精度和较强的泛化能力。     

Likelihood Inference for Discretely Observed Nonlinear Diffusions

This paper is concerned with the Bayesian estimation of nonlinear stochastic differential equations when observations are discretely sampled. The estimation framework relies on the introduction of latent auxiliary data to complete the missing diffusion between each pair of measurements. Tuned Markov chain Monte Carlo (MCMC) methods based on the Metropolis‐Hastings algorithm, in conjunction with the Euler‐Maruyama discretization scheme, are used to sample the posterior distribution of the latent data and the model parameters. Techniques for computing the likelihood function, the marginal likelihood, and diagnostic measures (all based on the MCMC output) are developed. Examples using simulated and real data are presented and discussed in detail.     

基于时变Markov的DCC-GARCH模型最小风险套期保值研究

考虑Markov状态转移概率的时变特征,在传统DCC-GARCH基础上,提出基于Markov时变转移概率的DCC-GARCH模型（TVTP-DCC-GARCH）研究最小风险套期保值比例的估计方法,并利用两阶段极大似然法对模型参数进行估计。进一步分别从样本内和样本外估计沪深300指数期货和现货的最优套期保值比率,对套期保值的绩效进行检验,并将检验结果分别与Markov转移概率恒定的DCC-GARCH（FTP-DCC-GARCH）、DCC-GARCH、OLS、1：1完全套期保值以及无套期保值的沪深300指数现货的绩效进行对比。实证结果表明,利用基于Markov状态转移的DCC-GARCH模型研究沪深300指数期货的套期保值问题具有一定合理性,且在参数估计中TVTP-DCC-GARCH模型的拟合效果最佳;在套期保值有效性方面,TVTP-DCC-GARCH模型优于其他模型,说明在DCC-GARCH模型中引入时变状态转移概率能够有效提高套期保值组合的绩效。     

Decision-Making with Heterogeneous Sources of Information

In any model the values of estimates for various parameters are obtained from different sources each with its own level of uncertainty. When the probability distributions of the estimates are obtained as opposed to point values only, the measurement uncertainties in the parameter estimates may be addressed. However, the sources used for obtaining the data and the models used to select appropriate distributions are of differing degrees of uncertainty. A hierarchy of different sources of uncertainty based upon one's ability to validate data and models empirically is presented. When model parameters are aggregated with different levels of the hierarchy represented, this implies distortion or degradation in the utility and validity of the models used. Means to identify and deal with such heterogeneous data sources are explored, and a number of approaches to addressing this problem is presented. One approach, using Range/Confidence Estimates coupled with an Information Value Analysis Process, is presented as an example.     

Framework for Microbial Food‐Safety Risk Assessments Amenable to Bayesian Modeling

Regulatory agencies often perform microbial risk assessments to evaluate the change in the number of human illnesses as the result of a new policy that reduces the level of contamination in the food supply. These agencies generally have regulatory authority over the production and retail sectors of the farm‐to‐table continuum. Any predicted change in contamination that results from new policy that regulates production practices occurs many steps prior to consumption of the product. This study proposes a framework for conducting microbial food‐safety risk assessments; this framework can be used to quantitatively assess the annual effects of national regulatory policies. Advantages of the framework are that estimates of human illnesses are consistent with national disease surveillance data (which are usually summarized on an annual basis) and some of the modeling steps that occur between production and consumption can be collapsed or eliminated. The framework leads to probabilistic models that include uncertainty and variability in critical input parameters; these models can be solved using a number of different Bayesian methods. The Bayesian synthesis method performs well for this application and generates posterior distributions of parameters that are relevant to assessing the effect of implementing a new policy. An example, based on Campylobacter and chicken, estimates the annual number of illnesses avoided by a hypothetical policy; this output could be used to assess the economic benefits of a new policy. Empirical validation of the policy effect is also examined by estimating the annual change in the numbers of illnesses observed via disease surveillance systems.     

Bayesian and Frequentist Inference in Partially Identified Models

A large‐sample approximation of the posterior distribution of partially identified structural parameters is derived for models that can be indexed by an identifiable finite‐dimensional reduced‐form parameter vector. It is used to analyze the differences between Bayesian credible sets and frequentist confidence sets. We define a plug‐in estimator of the identified set and show that asymptotically Bayesian highest‐posterior‐density sets exclude parts of the estimated identified set, whereas it is well known that frequentist confidence sets extend beyond the boundaries of the estimated identified set. We recommend reporting estimates of the identified set and information about the conditional prior along with Bayesian credible sets. A numerical illustration for a two‐player entry game is provided.     

Hierarchical Bayesian Modeling of Post-Earthquake Ignition Probabilities Considering Inter-Earthquake Heterogeneity

Post-earthquake fires are high-consequence events with extensive damage potential. They are also low-frequency events, so their nature remains underinvestigated. One difficulty in modeling post-earthquake ignition probabilities is reducing the model uncertainty attributed to the scarce source data. The data scarcity problem has been resolved by pooling the data indiscriminately collected from multiple earthquakes. However, this approach neglects the inter-earthquake heterogeneity in the regional and seasonal characteristics, which is indispensable for risk assessment of future post-earthquake fires. Thus, the present study analyzes the post-earthquake ignition probabilities of five major earthquakes in Japan from 1995 to 2016 (1995 Kobe, 2003 Tokachi-oki, 2004 Niigata–Chuetsu, 2011 Tohoku, and 2016 Kumamoto earthquakes) by a hierarchical Bayesian approach. As the ignition causes of earthquakes share a certain commonality, common prior distributions were assigned to the parameters, and samples were drawn from the target posterior distribution of the parameters by a Markov chain Monte Carlo simulation. The results of the hierarchical model were comparatively analyzed with those of pooled and independent models. Although the pooled and hierarchical models were both robust in comparison with the independent model, the pooled model underestimated the ignition probabilities of earthquakes with few data samples. Among the tested models, the hierarchical model was least affected by the source-to-source variability in the data. The heterogeneity of post-earthquake ignitions with different regional and seasonal characteristics has long been desired in the modeling of post-earthquake ignition probabilities but has not been properly considered in the existing approaches. The presented hierarchical Bayesian approach provides a systematic and rational framework to effectively cope with this problem, which consequently enhances the statistical reliability and stability of estimating post-earthquake ignition probabilities.     

Nonparametric Identification of Finite Mixture Models of Dynamic Discrete Choices

In dynamic discrete choice analysis, controlling for unobserved heterogeneity is an important issue, and finite mixture models provide flexible ways to account for it. This paper studies nonparametric identifiability of type probabilities and type‐specific component distributions in finite mixture models of dynamic discrete choices. We derive sufficient conditions for nonparametric identification for various finite mixture models of dynamic discrete choices used in applied work under different assumptions on the Markov property, stationarity, and type‐invariance in the transition process. Three elements emerge as the important determinants of identification: the time‐dimension of panel data, the number of values the covariates can take, and the heterogeneity of the response of different types to changes in the covariates. For example, in a simple case where the transition function is type‐invariant, a time‐dimension of T = 3 is sufficient for identification, provided that the number of values the covariates can take is no smaller than the number of types and that the changes in the covariates induce sufficiently heterogeneous variations in the choice probabilities across types. Identification is achieved even when state dependence is present if a model is stationary first‐order Markovian and the panel has a moderate time‐dimension (T 6).     

Bayesian Development of a Dose‐Response Model for Aspergillus fumigatus and Invasive Aspergillosis

Invasive aspergillosis (IA) is a major cause of mortality in immunocompromized hosts, most often consecutive to the inhalation of spores of Aspergillus. However, the relationship between Aspergillus concentration in the air and probability of IA is not quantitatively known. In this study, this relationship was examined in a murine model of IA. Immunosuppressed Balb/c mice were exposed for 60 minutes at day 0 to an aerosol of A. fumigatus spores (Af293 strain). At day 10, IA was assessed in mice by quantitative culture of the lungs and galactomannan dosage. Fifteen separate nebulizations with varying spore concentrations were performed. Rates of IA ranged from 0% to 100% according to spore concentrations. The dose‐response relationship between probability of infection and spore exposure was approximated using the exponential model and the more flexible beta‐Poisson model. Prior distributions of the parameters of the models were proposed then updated with data in a Bayesian framework. Both models yielded close median dose‐responses of the posterior distributions for the main parameter of the model, but with different dispersions, either when the exposure dose was the concentration in the nebulized suspension or was the estimated quantity of spores inhaled by a mouse during the experiment. The median quantity of inhaled spores that infected 50% of mice was estimated at 1.8 × 10⁴ and 3.2 × 10⁴ viable spores in the exponential and beta‐Poisson models, respectively. This study provides dose‐response parameters for quantitative assessment of the relationship between airborne exposure to the reference A. fumigatus strain and probability of IA in immunocompromized hosts.     

A Bayes Linear Bayes Method for Estimation of Correlated Event Rates

Typically, full Bayesian estimation of correlated event rates can be computationally challenging since estimators are intractable. When estimation of event rates represents one activity within a larger modeling process, there is an incentive to develop more efficient inference than provided by a full Bayesian model. We develop a new subjective inference method for correlated event rates based on a Bayes linear Bayes model under the assumption that events are generated from a homogeneous Poisson process. To reduce the elicitation burden we introduce homogenization factors to the model and, as an alternative to a subjective prior, an empirical method using the method of moments is developed. Inference under the new method is compared against estimates obtained under a full Bayesian model, which takes a multivariate gamma prior, where the predictive and posterior distributions are derived in terms of well‐known functions. The mathematical properties of both models are presented. A simulation study shows that the Bayes linear Bayes inference method and the full Bayesian model provide equally reliable estimates. An illustrative example, motivated by a problem of estimating correlated event rates across different users in a simple supply chain, shows how ignoring the correlation leads to biased estimation of event rates.     

Modeling the winning seed distribution of the NCAA Division I men׳s basketball tournament

The National Collegiate Athletic Association׳s (NCAA) men׳s Division I college basketball tournament is an annual competition that draws widespread attention in the United States. Predicting the winner of each game is a popular activity undertaken by numerous websites, fans, and more recently, academic researchers. This paper analyzes the 29 tournaments from 1985 to 2013, and presents two models to capture the winning seed distribution (i.e., a probability distribution modeling the winners of each round). The Exponential Model uses the exponential random variable to model the waiting time between a seed׳s successive winnings in a round. The Markov Model uses Markov chains to estimate the winning seed distributions by considering a seed׳s total number of winnings in previous tournaments. The proposed models allow one to estimate the likelihoods of different seed combinations by applying the estimated winning seed distributions, which accurately summarize aggregate performance of the seeds. Moreover, the proposed models show that the winning rate of seeds is not a monotonically decreasing function of the seed number. Results of the proposed models are validated using a chi-squared goodness of fit test and compared to the frequency of observed events.     

Dynamic Pricing to Minimize Maximum Regret

We consider a dynamic pricing problem that involves selling a given inventory of a single product over a short, two‐period selling season. There is insufficient time to replenish inventory during this season, hence sales are made entirely from inventory. The demand for the product is a stochastic, nonincreasing function of price. We assume interval uncertainty for demand, that is, knowledge of upper and lower bounds but not a probability distribution, with no correlation between the two periods. We minimize the maximum total regret over the two periods that results from the pricing decisions. We consider a dynamic model where the decision maker chooses the price for each period contingent on the remaining inventory at the beginning of the period, and a static model where the decision maker chooses the prices for both periods at the beginning of the first period. Both models can be solved by a polynomial time algorithm that solves systems of linear inequalities. Our computational study demonstrates that the prices generated by both our models are insensitive to errors in estimating the demand intervals. Our dynamic model outperforms our static model and two classical approaches that do not use demand probability distributions, when evaluated by maximum regret, average relative regret, variability, and risk measures. Further, our dynamic model generates a total expected revenue which closely approximates that of a maximum expected revenue approach which requires demand probability distributions.