r - Why does JAGS require at least two chains to calculate DIC? -


i'm trying understand how jags calculates deviance , deviance information criterion (dic).

the dic.samples function in rjags in r throws error if have 1 chain.

stop("2 or more parallel chains required") 

the formula calculating dic involves calculating expected deviance , deviance @ expected values of parameters in model. imagine such estimates obtained single chain (albeit there might convergence issues can identified more 1 chain).

the manual states that:

the pd monitor estimates contribution effective number of parameters (pd) [3] observed stochastic node comparing deviance deviance across multiple chains [1]. created using option type(pd). if model has 1 chain pd monitor cannot defined.

so question is:

  • why jags require @ least 2 chains calculate dic?
  • or more specifically, why pd monitor require 2 chains?

there number of different ways calculate pd - method jags uses described martyn plummer in discussion 'bayesian measures of model complexity , fit' paper: http://onlinelibrary.wiley.com/doi/10.1111/1467-9868.00353/abstract

get pdf , search 'plummer' - first result gives details of calculation used in jags.


Comments

Popular posts from this blog

c++ - Delete matches in OpenCV (Keypoints and descriptors) -

java - Could not locate OpenAL library -

sorting - opencl Bitonic sort with 64 bits keys -