\name{diffdic}

\alias{diffdic}

\title{Differences in penalized deviance}

\description{

Compare two models by the difference of two \code{dic} objects.

}

\usage{

\special{dic1 - dic2}

diffdic(dic1, dic2)

}

\arguments{

\item{dic1, dic2}{Objects inheriting from class ``dic''}

}

\details{

A \code{diffdic} object represents the difference in penalized

deviance between two models. A negative value indicates that

\code{dic1} is preferred and vice versa.

}

\value{

An object of class ``diffdic''. This is a numeric vector with

an element for each observed stochastic node in the model.

The \code{diffdic} class has its own print method, which will

display the sum of the differences, and its sample standard

deviation.

}

\note{

The problem of determining what is a noteworthy difference

in DIC (or other penalized deviance) between two models is

currently unsolved. Following the results of Ripley (1996) on

the Akaike Information Criterion, Plummer (2008) argues that

there is no absolute scale for comparison of two penalized

deviance statistics, and proposes that the difference should

be calibrated with respect to the sample standard deviation

of the individual contributions from each observed stochastic

node.

}

\author{Martyn Plummer}

\references{

Ripley, B. (1996)

\emph{Statistical Pattern Recognition and Neural Networks.}

Cambridge University Press.

Plummer, M. (2008)

Penalized loss functions for Bayesian model comparison.

\emph{Biostatistics}

doi: 10.1093/biostatistics/kxm049

}

\seealso{\code{\link{dic}}}

\keyword{models}