《Ch10ModelDiagnostics》.pdf

Chapter 10 Bayesian Diagnostics Bayesian Diagnostics • Convergence diagnostics. • Convergence diagnostics. • Posterior predictive checks. • Posterior predictive checks. • DIC, model selection, and complexity. • DIC, model selection, and complexity. • Bayes factors • Bayes factors • Sensitivity analysis • Sensitivity analysis Chapter 10 1 Convergence diagnostics Convergence diagnostics • Primarily, to assess whether the MCMC chain has • Primarily, to assess whether the MCMC chain has converged to a stationary distribution. converged to a stationary distribution. • We will use the CODA package in R. This implements • We will use the CODA package in R. This implements ¯ Gelman-Rubin diagnostic, based on multiple chains. ¯ Gelman-Rubin diagnostic, based on multiple chains. ¯ Geweke diagnostic for stationarity. ¯ Geweke diagnostic for stationarity. ¯ Heidelberger-Welch stationarity and run-length diagnostics. ¯ Heidelberger-Welch stationarity and run-length diagnostics. ¯ Raftery-Lewis run-length diagnostic (for quantiles). ¯ Raftery-Lewis run-length diagnostic (for quantiles). • The CODA package also contains: • The CODA package also contains: ¯ Diagnostic plots ¯ Diagnostic plots ¯ Functions for manipulating BUGS output ¯ Functions for manipulating BUGS output ¯ A function to find HPD intervals ¯ A function to find HPD intervals Chapter 10 2 State-space Schaefer State-space Schaefer surplus prodn model surplus prodn model (Millar and Meyer, 2000) (Millar and Meyer, 2000)