Showing papers by "James O. Berger published in 2005"
TL;DR: This work focuses on Bayesian model selection for the variable selection problem in large model spaces, and the issue of choice of prior distributions for the visited models is also important.
Abstract: We focus on Bayesian model selection for the variable selection problem in large model spaces. The challenge is to search the huge model space adequately, while accurately approximating model posterior probabilities for the visited models. The issue of choice of prior distributions for the visited models is also important.
61 citations
TL;DR: For exchangeable hierarchical multivariate normal models, it is first determined when a standard class of hierarchical priors results in proper or improper posteriors, and which elements of this class lead to admissible estimators of the mean under quadratic loss.
Abstract: Hierarchical modeling is wonderful and here to stay, but hyperparameter priors are often chosen in a casual fashion. Unfortunately, as the number of hyperparameters grows, the effects of casual choices can multiply, leading to considerably inferior performance. As an extreme, but not uncommon, example use of the wrong hyperparameter priors can even lead to impropriety of the posterior. For exchangeable hierarchical multivariate normal models, we first determine when a standard class of hierarchical priors results in proper or improper posteriors. We next determine which elements of this class lead to admissible estimators of the mean under quadratic loss; such considerations provide one useful guideline for choice among hierarchical priors. Finally, computational issues with the resulting posterior distributions are addressed.
56 citations
TL;DR: It is shown that it is possible to incorporate uncertainty in model inputs into analyses of traffic microsimulators such as CORSIM, and that incorporating this uncertainty can significantly change the variability of engineering simulations performed with CORS IM.
Abstract: CORSIM, a microsimulator for vehicular traffic, is being studied with respect to its ability to successfully model and predict behavior of traffic in a 36-block section of Chicago. Inputs to the simulator include information about street configuration, driver behavior, traffic light timing, turning probabilities at each intersection, and distributions of traffic ingress into the system. Data are available concerning the turning proportions in the actual neighborhood, as well as counts of vehicular ingress into the neighborhood and internal system counts, during a day in May 2000. Some of the data are accurate (video recordings), but some are quite inaccurate (observer counts of vehicles). Previous use of the full dataset involved “tuning” the parameters of CORSIM—in an ad hoc fashion—until CORSIM output was reasonably close to the actual data. This common approach, of simply tuning a complex computer model to real data, can result in poor parameter choices and completely ignores the often considerable unc...
26 citations
23 Jul 2005
TL;DR: A complete Bayesian approach to answering this question is developed for the challenging practical context in which the computer model produce functional data and is illustrated through study of a computer model developed to model vehicle.
Abstract: A key question in evaluation of computer models is Does the computer model adequately represent reality? A complete Bayesian approach to answering this question is developed for the challenging practical context in which the computer model (and reality) produce functional data. The methodology is particularly suited to treating the major issues associated with the validation process: quantifying multiple sources of error and uncertainty in computer models; combining multiple sources of information; and being able to adapt to different – but related – scenarios through hierarchical modeling. It is also shown how one can formally test if the computer model reproduces reality. The approach is illustrated through study of a computer model developed to model vehicle
15 citations
TL;DR: It is shown how many pseudo-Bayes factors proposed behave quite differently from ordinary Bayes factors, and that arguments of predictive optimality, based on simply inserting the empirical distribution in place of the 'true predictive distribution', can be misleading.
13 citations
TL;DR: In this paper, for exchangeable hierarchical multivariate normal models, the authors determine when a standard class of hierarchical priors results in proper or improper posteriors, and then determine which elements of this class lead to admissible estimators of the mean under quadratic loss.
Abstract: Hierarchical modeling is wonderful and here to stay, but hyperparameter priors are often chosen in a casual fashion. Unfortunately, as the number of hyperparameters grows, the effects of casual choices can multiply, leading to considerably inferior performance. As an extreme, but not uncommon, example use of the wrong hyperparameter priors can even lead to impropriety of the posterior. For exchangeable hierarchical multivariate normal models, we first determine when a standard class of hierarchical priors results in proper or improper posteriors. We next determine which elements of this class lead to admissible estimators of the mean under quadratic loss; such considerations provide one useful guideline for choice among hierarchical priors. Finally, computational issues with the resulting posterior distributions are addressed.
1 citations