• Corpus ID: 236493686

Detecting and diagnosing prior and likelihood sensitivity with power-scaling

@inproceedings{Kallioinen2021DetectingAD,
  title={Detecting and diagnosing prior and likelihood sensitivity with power-scaling},
  author={Noa Kallioinen and Topi Paananen and Paul-Christian Burkner and Aki Vehtari},
  year={2021}
}
Determining the sensitivity of the posterior to perturbations of the prior and likelihood is an important part of the Bayesian workflow. We introduce a practical and computationally efficient sensitivity analysis approach using importance sampling to estimate properties of posteriors resulting from power-scaling the prior or likelihood. On this basis, we suggest a diagnostic that can indicate the presence of prior-data conflict or likelihood noninformativity and discuss limitations to the power… 
Quantification of empirical determinacy: the impact of likelihood weighting on posterior location and spread in Bayesian meta-analysis estimated with JAGS and INLA
TLDR
This work quantified TED, pEDL and pEDS under different modeling conditions such as model parametrization, the primary outcome, and the prior, and clarified to what extent the location and spread of the marginal posterior distributions of the parameters are determined by the data.
Prior knowledge elicitation: The past, present, and future
TLDR
This work analyzes the state of the art of prior elicitation by identifying a range of key aspects of prior knowledge elicitation, from properties of the modelling task and the nature of the priors to the form of interaction with the expert.

References

SHOWING 1-10 OF 102 REFERENCES
Sensitivity analysis for Bayesian hierarchical models
TLDR
This work proposes a novel formal approach to prior sensitivity analysis which quantifies sensitivity without the need for a model re-run, and develops a ready-to-use priorSens package in R which can be used to detect high prior sensitivities of some parameters as well as identifiability issues in possibly over-parametrized Bayesian hierarchical models.
Confronting Prior Convictions: On Issues of Prior Sensitivity and Likelihood Robustness in Bayesian Analysis
TLDR
This review explores issues of the sensitivity of Bayes estimates to the prior and form of the likelihood, and a variety of computational strategies for significantly expanding the maintained sampling model, including the use of finite Gaussian mixture models and models based on Dirichlet process priors.
MCMC-based local parametric sensitivity estimations
Bayesian sensitivity analysis with the Fisher–Rao metric
We propose a geometric framework to assess sensitivity of Bayesian procedures to modelling assumptions based on the nonparametric Fisher–Rao metric. While the framework is general, the focus of this
Global Robust Bayesian Analysis in Large Models
  • P. Ho
  • Economics
    SSRN Electronic Journal
  • 2019
This paper develops tools for global prior sensitivity analysis in large Bayesian models. Without imposing parametric restrictions, the framework provides bounds for a wide range of posterior
Sensitivity and identification quantification by a relative latent model complexity perturbation in Bayesian meta‐analysis
TLDR
A unified method to simultaneously quantify both sensitivity and identification ( S - I) for all model parameters in a Bayesian NNHM, based on derivatives of the Bhattacharyya coefficient with respect to relative latent model complexity (RLMC) perturbations is developed.
Sensitivity of a Bayesian Analysis to the Prior Distribution
TLDR
This work describes the sensitivity of a posterior distribution (or posterior mean) to prior and illustrates the results on two distinct problems: a) determining least-informative (vague) priors and b) estimating statistical quantiles for a problem in analyzing projectile accuracy.
Bayesian estimation - A sensitivity analysis
The robustness of the assigned prior distribution in a Bayesian estimation problem is examined. A Bayesian analysis for a stochastic intensity parameter of a Poisson distribution is summarized in
The Prior Can Often Only Be Understood in the Context of the Likelihood
TLDR
This paper resolves an apparent paradox in prior modeling: a model encoding true prior information should be chosen without reference to the model of the measurement process, but almost all common prior modeling techniques are implicitly motivated by a reference likelihood.
Local Robustness in Bayesian Analysis
TLDR
The use of local analysis to study sensitivity in contexts broader than merely prior uncertainty is discussed, ranging from differentiation of functions to differentiation of functionals (e.g., assessing sensitivity to hyperparameters) and so on.
...
...