In R, there are packages and functions for MCMC's such as jags.model that allows us to control the number of chains we would like to run, i.e., within jags.model, we can set n.chains = 4 for $4$ chains. The resulting traceplot is multi-color as it traces out the posterior estimates for four separate chains. My question is, when we retrieve posterior draws or estimates from the object, which of the $4$ chains was used? Was it an average? Thanks.
Asked
Active
Viewed 1,684 times
9
user321627
- 2,511
- 3
- 13
- 49
1 Answers
12
In general, the burn-in iterations for each chain are dropped and the rest of the iterations are simply combined, as if from one long chain. The main reason for having separate chains is to better diagnose convergence issues.
You may also find this question helpful.
