1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
|
\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}
|
