In IQSS/Zelig: Everyone's Statistical Software

Built using Zelig version r packageVersion("Zelig")

knitr::opts_knit$set(
    stop_on_error = 2L
)
knitr::opts_chunk$set(
    fig.height = 11,
    fig.width = 7
)

options(cite = FALSE)

---

rm(list=ls(pattern="\\.out"))
suppressWarnings(suppressMessages(library(Zelig)))
set.seed(1234)

Zelig has three available diagnostic tests for MCMC models. Here is an example of their use. First we attach the sample dataset and estimate a linear regression using the MCMC normal.bayes model:

data(macro)
z.out <- zelig(unem ~ gdp + capmob + trade, model = "normal.bayes",
               data = macro, verbose = FALSE)

You can check for convergence before summarizing the estimates with three diagnostic tests.

Geweke Diagnostic

The Geweke diagnostic tests the null hypothesis that the Markov chain is in the stationary distribution and produces z-statistics for each estimated parameter.

z.out$geweke.diag()

Heidelberger-Welch Diagnostic

The Heidelberger and Welch diagnostic first tests the null hypothesis that the Markov Chain is in the stationary distribution and produces p-values for each estimated parameter. Calling this diagnostic also produces output that indicates whether the mean of a marginal posterior distribution can be estimated with sufficient precision, assuming that the Markov Chain is in the stationary distribution.

z.out$heidel.diag()

Raftery-Lewis Diagnostic

The Raftery and Lewis diagnostic indicates how long the Markov Chain should run before considering draws from the marginal posterior distributions sufficiently representative of the stationary distribution.

z.out$raftery.diag()

See also

The convergence diagnostics are part of the CODA package by Martyn Plummer, Nicky Best, Kate Cowles, and Karen Vines. Bayesian normal regression is part of the MCMCpack package by Andrew D. Martin and Kevin M. Quinn.

z.out$references()

IQSS/Zelig documentation built on Dec. 11, 2023, 1:51 a.m.