i am relatively new to winBUGS and i am running a meta-regression model for bayesian meta-analysis. This model tracks the posterior distributions of the parameters mean and tau-square. Moreover,which values are non-informative for a uniform prior distribution? Since i am not quite familiar with MCMC and winBUGS i would appreciate any help! Thank you in advance.
1
-
$\begingroup$ Thank you that was very helpful!! Is it preferable to use different initial values for each chain? I've seen other examples using the same initial values from every chain and others choosing different ones for each chain. $\endgroup$– SellyJan 20, 2021 at 21:31
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
In model development, you generally want at least 2 (a common default is 4) chains. The primary reason for having multiple chains is to assess the model performance: you want to make sure that your posterior is being sampled adequately and appropriately, which isn't easy to do if you only had one chain. Practically speaking, there's not usually any advantage to going overboard with the number of chains, and most people's CPUs place an upward bound on that possibility anyway. Once you have a model that you know performs well, you don't actually need multiple chains, but you do still want to have documented evidence that your chains mixed and behaved appropriately.
