dic.samples {rjags} | R Documentation |
Function to extract random samples of the penalized deviance from
a jags
model.
dic.samples(model, n.iter, thin = 1, type, ...) ## S3 method for class 'dic': as.mcmc(x)
model |
a jags model object |
n.iter |
number of iterations to monitor |
thin |
thinning interval for monitors |
type |
type of penalty to use |
x |
An object inheriting from class ``dic'' |
... |
optional arguments passed to the update method for jags model objects |
The dic.samples
function generates penalized deviance
statistics for use in model comparison. The two penalized deviance
statistics generated by dic.samples
are the deviance
information criterion (DIC) and the penalized expected deviance.
These are chosen by giving the values ``pD'' and ``popt'' respectively
as the type
argument.
DIC (Spiegelhalter et al 2002) is calculated by adding the ``effective
number of parameters'' (pD
) to the expected deviance. The
definition of pD
used by dic.samples
is the one proposed
by Plummer (2002) and requires two or more parallel chains in the
model.
DIC is an approximation to the penalized plug-in deviance, which is used when only a point estimate of the parameters is of interest. The DIC approximation only holds asymptotically when the effective number of parameters is much smaller than the sample size, and the model parameters have a normal posterior distribution.
The penalized expected deviance (Plummer 2008) is calculated by adding
the optimism (popt
) to the expected deviance. The popt
penalty is always larger than the pD
penalty, and penalizes
complex models more severely.
An object of class ``dic''. This is a list containing the following elements:
deviance |
A list of mcarray objects, one for each
observed stochastic node, containing samples of the deviance |
penalty |
A list of mcarray objects, one for each
observed stochastic node, containing samples of the penalty
function |
type |
A string identifying the type of penalty: ``pD'' or ``popt'' |
An object of class dic
can be coerced to an mcmc
object
using the as.mcmc
generic function. The resulting mcmc
object has two variables: the mean deviance over all chains and
the penalty.
The popt
penalty is estimated by importance weighting, and may
be numerically unstable. It is recommended to inspect the dic
object after coercing it to a mcmc
object using functions from
the coda
package.
Martyn Plummer
Spiegelhalter, D., N. Best, B. Carlin, and A. van der Linde (2002), Bayesian measures of model complexity and fit (with discussion). Journal of the Royal Statistical Society Series B 64, 583-639.
Plummer, M. (2002), Discussion of the paper by Spiegelhalter et al. Journal of the Royal Statistical Society Series B 64, 620.
Plummer, M. (2008) Penalized loss functions for Bayesian model comparison. Biostatistics doi: 10.1093/biostatistics/kxm049