Using WinBUGS, I fit several Bayesian hierarchical logistic regression models for the mean of a binary response variable conditional on a set of criteria.
I am now using CODA in R to determine if my simulations have converged. I've visually inspected trace plots and density plots as a first step, and things look reasonable (though I am quite new to this and not completely confident in my assessment).
I've also used Gelman and Rubin's diagnostic. (I have multiple chains for each of the model variations I fit). All of the parameters here have scale reductions factors below 1.1.
I've looked at the Geweke test, and the use of Z-scores makes me uncertain if it is appropriate for non-normal models. Similarly, the normality test used in the Heidelberger & Welch test leaves me uncertain about whether I should be using it. Should I include these in my convergence diagnostic tests or are they not appropriate for my model? Are there any other convergence tests I should be including that are appropriate for my binary parameter?
