Choosing and Extending Bayesian Models
Alan Agresti, Maria Kateri, Ranjini Grove et al. · Chapman & Hall/CRC eBooks · 2026
Bayesian model choice and averaging provide a principled framework for selecting and combining models. Bayes factors quantify evidence in favor of competing models, while posterior model probabilities allow for weighted averaging of predictions, reducing the risk of over-relying on a single model. Robust Bayesian analysis provides a systematic framework for assessing and mitigating the impact of uncertainties in the prior and likelihood. Methods for classification of observations into categories of a response variable based on their explanatory variable values include logistic regression, linear discriminant analysis, naive Bayes (which treats the various explanatory variables as conditionally independent of the others for each category), and classification trees that graphically portray classifications. This chapter also introduces Bayesian approaches to modeling network data, providing a flexible framework to account for uncertainty in complex relational structures. When data are missing for some observations, the multiple imputation process uses Monte Carlo to replace missing values multiple times by simulated values from their conditional distribution, given the observed data. These results are not biased when the data are missing at random.
Cited by 0 works. doi:10.1201/9781003715924-7