Notes on MCMC convergence

1
(No Transcript)
2
Notes on MCMC convergence

• Mainly taken directly from Kate Cowles notes

http//www.stat.uiowa.edu/ftp/kcowles/s138/
3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
Let d denote the quantity of interest and m
denote the number of parallel chains, each run
for 2n iterations discarding the first n Idea
before convergence, the individual sequences will
be much less variable than the converged
sequences. Between-sequence Variability, B
Within-sequence Variability, W
9
will overestimate the true variance if the
starting points are overdispersed
W will underestimate the true variance since
individual sequences will have ranged over only
part of their support GR Stat is
generally run the simulations until all GR
values are less than 1.1 or 1.2
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)