Hi there,

I am currently writing a manuscript in which I used NUTS to sample the model. I wonder if there are any recipes for reporting the MCMC diagnostics for scientific publication? I have read https://cran.r-project.org/web/packages/bayesplot/vignettes/visual-mcmc-diagnostics.html on Visual MCMC diagnostics using the bayesplot package and https://mc-stan.org/docs/2_25/cmdstan-guide/diagnose.html using cmdstan_diagnose(). However, I lost on the selection of diagnostics because there are so many of them and different ways and different parameters. Iâ€™d not like to distract readers from tons of diagnostic plots. Besides, Bayesian especially using NUTS, is not yet the mainstream inference in my field (neuroimaging).

Do you have any suggestions? What I could think of so far is one figure (2x2) combining

- a plot on divergence using
`mcmc_nuts_divergence(np_cp, lp_cp)`

or`stan_diag(object, "divergence")`

; - a plot on Rhat using
`mcmc_rhat()`

or`stan_rhat`

; - a plot from
`stan_diag(object, "sample")`

to show lp and acceptance; - a plot of
`mcmc_neff()`

to show ESS.

Another question, do you think it is legitimate to show the Rhat and neff/N for only key parameters? I have more than 100 parameters, planning to show key ones in the main context and a version with all parameters rhat and neff/N in supplementary.

Thanks a lot,

ZC