Nonlinear mixed effects models for data in the form of continuous, repeated measurements on each of a number of individuals, also known as hierarchical nonlinear models, are a popular platform for analysis when interest focuses on individual-specific characteristics. This framework first enjoyed widespread attention within the statistical research community in the late 1980s, and the 1990s saw vigorous development of new methodological and computational techniques for these models, the emergence of general-purpose software, and broad application of the models in numerous substantive fields. This article presents an overview of the formulation, interpretation, and implementation of nonlinear mixed effects models and surveys recent advances and applications.

Abstract. Various noninformative prior distributions have been suggested for scale parameters in hierarchical models. We construct a new folded-noncentral-t family of conditionally conjugate priors for hierarchical standard deviation parameters, and then consider noninformative and weakly informative priors in this family. We use an example to illustrate serious problems with the inverse-gamma family of “noninformative ” prior distributions. We suggest instead to use a uniform prior on the hierarchical standard deviation, using the half-t family when the number of groups is small and in other settings where a weakly informative prior is desired. We also illustrate the use of the half-t family for hierarchical modeling of multiple variance parameters such as arise in the analysis of variance.

Hierarchical linear and generalized linear models can be fit using Gibbs samplers and Metropolis algorithms; these models, however, often have many parameters, and convergence of the seemingly most natural Gibbs and Metropolis algorithms can sometimes be slow. We examine solutions that involve reparameterization and over-parameterization. We begin with parameter expansion using working parameters, a strategy developed for the EM algorithm by Meng and van Dyk (1997) and Liu, Rubin, and Wu (1998). This strategy can lead to algorithms that are much less susceptible to becoming stuck near zero values of the variance parameters than are more standard algorithms. Second, we consider a simple rotation of the regression coefficients based on an estimate of their posterior covariance matrix. This leads to a Gibbs algorithm based on updating the transformed parameters one at a time or a Metropolis algorithm with vector jumps; either of these algorithms can perform much better (in terms of total CPU time) than the two standard algorithms: one-at-a-time updating of untransformed parameters or vector updating using a linear regression at each step. We present an innovative evaluation of the algorithms in terms of how quickly they can get away from remote areas of parameter space, along with some more standard evaluation of computation and convergence speeds. We illustrate our methods with examples from our applied work. Our ultimate goal is to develop a fast and reliable method for fitting a hierarchical linear model as easily as one can now fit a non-hierarchical model, and to increase understanding of Gibbs samplers for hierarchical models in general. Keywords: Bayesian computation, blessing of dimensionality, Markov chain Monte Carlo, multilevel modeling, mixed effects models, PX-EM algorithm, random effects regression, redundant

We present a Bayesian mixture model for probabilistic latent semantic analysis of documents with images and text. The Bayesian perspective allows us to perform automatic regularisation to obtain sparser and more coherent clustering models. It also enables us to encode a priori knowledge, such as word and image preferences. The learnt model can be used for browsing digital databases, information retrieval with image and/or text queries, image annotation (adding words to an image) and text illustration (adding images to a text).

Data visualizatio n has the poE tialto assist humans in analyzing and co prehending large vo umes o data, andto detect patterns, clusters ando utliers that are n o o vi oq using noq graphical f ms o presentatiot Fo this reas oq data visualizatio ns have an impo rtant ro le to play in a diverse range o applied pro blems, including data explo atio n and mining, info rmatio n retrieval, and intelligence analysis.

How do we interpret findings that are intriguing, potentially important, but not statistically significant? We discuss in the context of a series of papers in the Journal of Theoretical Biology that reported evidence that beautiful parents have more daughters, violent men have more sons, and other sex-ratio patterns (Kanazawa, 2005, 2006, 2007). These papers have been shown to have statistical errors, but the more general research questions remain. From a classical statistical perspective, these studies have insufficient power to detect the magnitudes of effects (on the order of 1 percentage point) that could be expected based on earlier studies of sex ratios. The anticipated small effects can also be handled using a Bayesian prior distribution. These concerns are relevant to other studies of small effects and also to the reporting of such studies.

Figueiredo (2001) and Figueiredo and Jain (2001) described a particular sparseness-inducing Bayesian model for probit regression. For several standard datasets, they reported...

Bayesian inference requires all unknowns to be represented by probability distributions, which awkwardly implies that the probability of an event for which we are completely ignorant (e.g., that the world’s greatest boxer would defeat the world’s greatest wrestler) must be assigned a particular numerical value such as 1/2, as if it were known as precisely as the probability of a truly random event (e.g., a coin flip). Robust Bayes and belief functions are two methods that have been proposed to distinguish ignorance and randomness. In robust Bayes, a parameter can be restricted to a range, but without a prior distribution, yielding a range of potential posterior inferences. In belief functions (also known as the Dempster-Shafer theory), probability mass can be assigned to subsets of parameter space, so that randomness is represented by the probability distribution and uncertainty is represented by large subsets, within which the model does not attempt to assign probabilities. Through a simple example involving a coin flip and a boxing/wrestling match, we illustrate difficulties with robust Bayes and belief functions. In short: pure Bayes does not distinguish ignorance and randomness; robust Bayes allows ignorance to spread too broadly, and belief functions inappropriately collapse to simple Bayesian models. Keywords: Belief functions, Dempster-Shafer theory, epistemic and aleatory uncertainty, foundations of probability, ignorance, robust Bayes, subjective prior distribution

Introduction The Bayesian approach to data analysis computes conditional probability distribu- tions of quantities of interest (such as future observables) given the observed data. Bayesian analyses usually begin with a .full probability model - a joint probability dis- tribution for all the observable and unobservable quantities under study - and then use Bayes' theorem (Bayes, 1763) to compute the requisite conditional probability distributions (called poster'Joy distributions). The theorem itself is innocuous enough. In its simplest form, if Q denotes a quantity of interest and D denotes data, the theorem states: P(ql D) P(;lq) X P(q)/P(). This theorem prescribes the basis for statistical learning in the probabilistic frame- work. With p(Q) regarded as a probabilistic statement of prior knowledge about Q before obtaining the data D, p(QI D) becomes a revised probabilistic statement of our knowledge about Q in the light of the data (Bernardo and Smith, 1994, p.2). The marginal lik

Hierarchical linear and generalized linear models can be fit using Gibbs samplers and Metropolis algorithms; these models, however, often have many parameters, and convergence of the seemingly most natural Gibbs and Metropolis algorithms can sometimes be slow. We examine solutions that involve reparameterization and over-parameterization. We begin with parameter expansion using working parameters, a strategy developed for the EM algorithm by Meng and van Dyk (1997) and Liu, Rubin, and Wu (1998). This strategy can lead to algorithms that are much less susceptible to becoming stuck near zero values of the variance parameters than are more standard algorithms. Second, we consider a simple rotation of the regression coefficients based on an estimate of their posterior covariance matrix. This leads to a Gibbs algorithm based on updating the transformed parameters one at a time or a Metropolis algorithm with vector jumps; either of these algorithms can perform much better (in terms of total CPU time) than the two standard algorithms: one-at-a-time updating of untransformed parameters or vector updating using a linear regression at each step. We present an innovative evaluation of the algorithms in terms of how quickly they can get away from remote areas of parameter space, along with some more standard evaluation of computation and convergence speeds. We illustrate our methods with examples from our applied work. Our ultimate goal is to develop a fast and reliable method for fitting a hierarchical linear model as easily as one can now fit a non-hierarchical model, and to increase understanding of Gibbs samplers for hierarchical models in general. Keywords: Bayesian computation, blessing of dimensionality, Markov chain Monte Carlo, multilevel modeling, mixed effects models, PX-EM algorithm, random effects regression, redundant