Singular prior distributions and ill-conditioning in Bayesian D-optimal design for several nonlinear models

TL;DR

This paper investigates singular prior distributions in Bayesian D-optimal design, characterizes their properties for various models, and develops methods to handle ill-conditioning issues, especially in logistic regression.

Contribution

It introduces the concept of singular priors, characterizes their conditions for several models, and proposes numerical methods to derive Bayesian D-efficient designs under ill-conditioning.

Findings

Some recommended weakly informative priors are singular.

Sufficient conditions for singularity and non-singularity are established.

Numerical methods are demonstrated for logistic regression with heavy-tailed priors.

Abstract

For Bayesian D-optimal design, we define a singular prior distribution for the model parameters as a prior distribution such that the determinant of the Fisher information matrix has a prior geometric mean of zero for all designs. For such a prior distribution, the Bayesian D-optimality criterion fails to select a design. For the exponential decay model, we characterize singularity of the prior distribution in terms of the expectations of a few elementary transformations of the parameter. For a compartmental model and several multi-parameter generalized linear models, we establish sufficient conditions for singularity of a prior distribution. For the generalized linear models we also obtain sufficient conditions for non-singularity. In the existing literature, weakly informative prior distributions are commonly recommended as a default choice for inference in logistic regression. Here it is shown that some of the recommended prior distributions are singular, and hence should not be used for Bayesian D-optimal design. Additionally, methods are developed to derive and assess Bayesian D-efficient designs when numerical evaluation of the objective function fails due to ill-conditioning, as often occurs for heavy-tailed prior distributions. These numerical methods are illustrated for logistic regression.