Empirical Determinacy of Posterior Location and Scale in Bayesian Hierarchical Models
summary
Claims a concrete diagnostic and provides code: the R package ed4bhm computes a power-weighted likelihood perturbation (w ≈ 1 ± Δ), estimates overlap via a Bhattacharyya-like coefficient and reports total empirical determinacy plus separate location/spread contributions. Implements MCMC (JAGS, Stan) and INLA shortcuts to avoid reruns. Applied to Bayesian logistic regression on bacterial resistance, results show the spread-driven determinacy falls from near 1 to near 0 as separation increases, flagging parameters that remain prior-dominated.
AI-written summary · about
chapters (13)
- 0:00Bayesian modeling challenges
- 2:00Empirical determinacy in inference
- 4:34Hierarchical model identification
- 6:17Measuring posterior sensitivity
- 8:28Total empirical determinacy
- 10:57Location and scale effects
- 13:09MCMC and Laplace approximations
- 17:02Software for determinacy analysis
- 17:23Logistic regression separation
- 24:57Separation experiments results
- 26:47Hemophilia aging analysis
- 28:54Method strengths and limits
- 29:35Q&A
related talks
36mClusterf*ck: A Practical Guide to Bayesian Hierarchical Modeling in PyMC3
01h 06mNonparametric Bayesian Methods: Models, Algorithms, and Applications I
48mLocally Private Bayesian Inference for Count Models
47mScale Calibration for High-Dimensional Robust Regression
57mA New Perspective on High-Dimensional Causal Inference
01h 24mBayesianism and Survival Analysis
01h 04mNonparametric Bayesian Methods: Models, Algorithms, and Applications II
25mScaling Bayesian Experimentation