ELIR ESS computation

[Chiam97430]

Hello,
The ess() function may require some adjustments in cases of large between-trial heterogeneity.
I have conducted a simulation for binary responses using this package and, in some scenarios, I obtained negative or implausibly high ELIR ESS values.

example:
data <- data.frame(
study = 1:4,
n = rep(30, 4),
r = c(2, 18, 0, 2)
)
map_mcmc <- gMAP(cbind(r, n-r) ~ 1 | study,
data=data,
tau.dist="HalfNormal",
tau.prior=1,
beta.prior=2,
family=binomial(link="logit"))
map_fit <- automixfit(map_mcmc)
map_robust <- robustify(map_fit, weight=0.1, mean=1/2)
ess(map_robust)

For the same data and model configuration, set.seed(36546) results in a negative ELIR ESS, while set.seed(37500) results in a large positive ELIR ESS.

[weberse2]

Thanks for reporting. I can confirm the large and even negative ESS values. This is rooted in this case in the behaviour at the boundary. Whenever there is no clear peak of the posterior, then I would not consider ESS routines to be useful. A possibility is to consider the problem on the logit scale, which I need to investigate.

[Chiam97430]

Thank you for the prompt reply! I look forward to your updates on the logit scale approach, or hopefully, it can at least show an error when the calculation is not reliable.

Thanks again for your attention to this!