Jennifer A. Hoeting

Bayesian Model Averaging for Linear Regression Models

Abstract We consider the problem of accounting for model uncertainty in linear regression models. Conditioning on a single selected model ignores model uncertainty, and thus leads to the underestimation of uncertainty when making inferences about quantities of interest. A Bayesian solution to this problem involves averaging over all possible models (i.e., combinations of predictors) when making inferences about quantities of interest. This approach is often not practical. In this article we offer two alternative approaches. First, we describe an ad hoc procedure, “Occams window,” which indicates a small set of models over which a model average can be computed. Second, we describe a Markov chain Monte Carlo approach that directly approximates the exact solution. In the presence of model uncertainty, both of these model averaging procedures provide better predictive performance than any single model that might reasonably have been selected. In the extreme case where there are many candidate predictors but ...

A structured and dynamic framework to advance traits‐based theory and prediction in ecology

Predicting changes in community composition and ecosystem function in a rapidly changing world is a major research challenge in ecology. Traits-based approaches have elicited much recent interest, yet individual studies are not advancing a more general, predictive ecology. Significant progress will be facilitated by adopting a coherent theoretical framework comprised of three elements: an underlying trait distribution, a performance filter defining the fitness of traits in different environments, and a dynamic projection of the performance filter along some environmental gradient. This framework allows changes in the trait distribution and associated modifications to community composition or ecosystem function to be predicted across time or space. The structure and dynamics of the performance filter specify two key criteria by which we judge appropriate quantitative methods for testing traits-based hypotheses. Bayesian multilevel models, dynamical systems models and hybrid approaches meet both these criteria and have the potential to meaningfully advance traits-based ecology.

MODEL SELECTION FOR GEOSTATISTICAL MODELS

We consider the problem of model selection for geospatial data. Spatial correlation is often ignored in the selection of explanatory variables, and this can influence model selection results. For example, the importance of particular explanatory variables may not be apparent when spatial correlation is ignored. To address this problem, we consider the Akaike Information Criterion (AIC) as applied to a geostatistical model. We offer a heuristic derivation of the AIC in this context and provide simulation results that show that using AIC for a geostatistical model is superior to the often-used traditional approach of ignoring spatial correlation in the selection of explanatory variables. These ideas are further demonstrated via a model for lizard abundance. We also apply the principle of minimum description length (MDL) to variable selection for the geostatistical model. The effect of sampling design on the selection of explanatory covariates is also explored. R software to implement the geostatistical model selection methods described in this paper is available in the Supplement.

An improved model for spatially correlated binary responses

In this paper, we use covariates and an indication of sampling effort in an autologistic model to improve predictions of probability of presence for lattice data. The model is applied to sampled data where only a small proportion of the available sites have been observed. We adopt a Bayesian set-up and develop a Gibbs sampling estimation procedure. In four examples based on simulated data, we show that the autologistic model with covariates improves predictions compared with the simple logistic regression model and the basic autologistic model (without covariates). Software to implement the methodology is available at no cost from StatLib.

A method for simultaneous variable selection and outlier identification in linear regression

We suggest a method for simultaneous variable selection and outlier identification based on the computation of posterior model probabilities. This avoids the problem that the model you select depends upon the order in which variable selection and outlier identification are carried out. Our method can find multiple outliers and appears to be successful in identifying masked outliers. We also address the problem of model uncertainty via Bayesian model averaging. For problems where the number of models is large, we suggest a Markov chain Monte Carlo approach to approximate the Bayesian model average over the space of all possible variables and outliers under consideration. Software for implementing this approach is described. In an example, we show that model averaging via simultaneous variable selection and outlier identification improves predictive performance and provides more accurate prediction intervals as compared to any single model that might reasonably be selected.

Linking Chronic Wasting Disease To Mule Deer Movement Scales: A Hierarchical Bayesian Approach

Observed spatial patterns in natural systems may result from processes acting across multiple spatial and temporal scales. Although spatially explicit data on processes that generate ecological patterns, such as the distribution of disease over a landscape, are frequently unavailable, information about the scales over which processes operate can be used to understand the link between pattern and process. Our goal was to identify scales of mule deer (Odocoileus hemionus) movement and mixing that exerted the greatest influence on the spatial pattern of chronic wasting disease (CWD) in northcentral Colorado, USA. We hypothesized that three scales of mixing (individual, winter subpopulation, or summer subpopulation) might control spatial variation in disease prevalence. We developed a fully Bayesian hierarchical model to compare the strength of evidence for each mixing scale. We found strong evidence that the finest mixing scale corresponded best to the spatial distribution of CWD infection. There was also evidence that land ownership and habitat use play a role in exacerbating the disease, along with the known effects of sex and age. Our analysis demonstrates how information on the scales of spatial processes that generate observed patterns can be used to gain insight when process data are sparse or unavailable.

The importance of accounting for spatial and temporal correlation in analyses of ecological data.

nonlinear inversion: how many parameters can we estimate and which measurements are most useful? Global Change Biology 7:495-510. Williams, M., P. A. Schwarz, B. E. Law, J. Irvine, and M. R. Kurpius. 2005. An improved analysis of forest carbon dynamics using data assimilation. Global Change Biology 11:89-105. Xu, T., L. White, D. Hui, and Y. Luo. 2006. Probabilistic inversion of a terrestrial ecosystem model: analysis of un certainty in parameter estimation and model prediction. Global Biogeochemical Cycles 20. [doi: 10.1029/2005GB002468]

Bayesian Variable and Transformation Selection in Linear Regression

This article suggests a method for variable and transformation selection based on posterior probabilities. Our approach allows for consideration of all possible combinations of untransformed and transformed predictors along with transformed and untransformed versions of the response. To transform the predictors in the model, we use a change-point model, or “change-point transformation,” which can yield more interpretable models and transformations than the standard Box–Tidwell approach. We also address the problem of model uncertainty in the selection of models. By averaging over models, we account for the uncertainty inherent in inference based on a single model chosen from the set of models under consideration. We use a Markov chain Monte Carlo model composition (MC3) method which allows us to average over linear regression models when the space of models under consideration is very large. This considers the selection of variables and transformations at the same time. In an example, we show that model averaging improves predictive performance as compared with any single model that might reasonably be selected, both in terms of overall predictive score and of the coverage of prediction intervals. Software to apply the proposed methodology is available via StatLib.

Autoregressive models for capture-recapture data: A Bayesian approach

In this article, we incorporate an autoregressive time-series framework into models for animal survival using capture-recapture data. Researchers modeling animal survival probabilities as the realization of a random process have typically considered survival to be independent from one time period to the next. This may not be realistic for some populations. Using a Gibbs sampling approach, we can estimate covariate coefficients and autoregressive parameters for survival models. The procedure is illustrated with a waterfowl band recovery dataset for northern pintails (Anas acuta). The analysis shows that the second lag autoregressive coefficient is significantly less than 0, suggesting that there is a triennial relationship between survival probabilities and emphasizing that modeling survival rates as independent random variables may be unrealistic in some cases. Software to implement the methodology is available at no charge on the Internet.

A clipped latent variable model for spatially correlated ordered categorical data

We propose a model for a point-referenced spatially correlated ordered categorical response and methodology for inference. Models and methods for spatially correlated continuous response data are widespread, but models for spatially correlated categorical data, and especially ordered multi-category data, are less developed. Bayesian models and methodology have been proposed for the analysis of independent and clustered ordered categorical data, and also for binary and count point-referenced spatial data. We combine and extend these methods to describe a Bayesian model for point-referenced (as opposed to lattice) spatially correlated ordered categorical data. We include simulation results and show that our model offers superior predictive performance as compared to a non-spatial cumulative probit model and a more standard Bayesian generalized linear spatial model. We demonstrate the usefulness of our model in a real-world example to predict ordered categories describing stream health within the state of Maryland.