Closed n-kall closed 7 months ago
I would not change that, as it's still telling whether the error can be small. Capping with 0.7 is related to reliability of MCSE estimate which is not good if bias dominates, but then if sample size is big enough, bias can be small too. We can talk more when I'm backl to campus
Ok, that sounds good. Closing this
Currently
ps_khat_threshold
is just returning1 - 1 / log10(S)
. As the recommendation in PSIS paper is to use 0.7 as an upper bound on the threshold, should this be changed tomin(1 - 1 / log10(S), 0.7)
as it is inloo
? @avehtari