R hat of 1.01

Hello everyone

I came across a recent study entitled Rank-normalization, folding, and localization: An improved R hat for assessing convergence of MCMC, by Vehtari, Gelman, Simpson, Bob Carpenter, and Bürkner (2020).

What I understand, that it is recommended to use R hat of 1.01 in order to ensure chain convergence instead of the previous recommended value of 1.10 suggested by Gelman and Rubin (1992).

My question is, will this new value also be applied to the traditional computation of R hat? Because as far as I understand, the new value of 1.01 is associated with the new calculation of R hat that is based on rank normalizing.
Is this new version of R hat incorporated in R program?

Thank you, and I am sorry if I misunderstand any part of this paper.

Yeah it’s available in our new package posterior:

It’s not on CRAN quite yet but you can install it from github:


The new r-hat is computed by the rhat() function. The old version of r-hat is available via rhat_basic().


Many thanks, Johan, for sharing this information.
It helps a lot.

1 Like