diffdic: Differences in penalized deviance

In rjags: Bayesian Graphical Models using MCMC

Differences in penalized deviance

Description

Compare two models by the difference of two dic objects.

Usage

dic1 - dic2
diffdic(dic1, dic2)

Arguments

dic1, dic2

Objects inheriting from class “dic”

Details

A diffdic object represents the difference in penalized deviance between two models. A negative value indicates that 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 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(s)

Martyn Plummer

References

Ripley, B. (1996) Statistical Pattern Recognition and Neural Networks. Cambridge University Press.

Plummer, M. (2008) Penalized loss functions for Bayesian model comparison. Biostatistics doi: 10.1093/biostatistics/kxm049

See Also

dic

rjags documentation built on Sept. 11, 2024, 6:31 p.m.