bnormnlr (version 1.0)

infocrit: Expected Number of Parameters, DIC, AIC and BIC for bnlr fit.

Description

Function to calculate Expected Number of Parameters, DIC, AIC and BIC for bnlr output.

Usage

infocrit(model, burn)

Arguments

model

A list derived from bnlr function

burn

A vector indicating which samples must be discarded from the mcmc simulation

Value

a vector with:

pd

Expected Number of Parameters

DIC

Deviance Information Criterion

AIC

Akaike Information Criterion

BIC

Bayesian Information Criterion

References

Carlin, B. P. & Louis, T. A. (2009), Bayesian Methods for Data Analysis, 3rd edn, CRC Press, New York.

Gamerman, D. & Lopes, H. F. (2006), Markov Chain Monte Carlo: Stochastic Simulation for Bayesian Inference, 2nd edn, CRC Press, New York.

Examples

#######################################
###Simulation of heteroscedastic model
#######################################
utils::data(muscle, package = "MASS")
###mean and variance functions
fmu<-function(param,cov){ param[1] + param[2]*exp(-cov/exp(param[3]))}
fsgma<-function(param,cov){drop(exp(cov%*%param))}

###simulate heteroscedastic data
muscle$Length<-fmu(c(28.9632978, -34.2274097,  -0.4972977),muscle$Conc)+
rnorm(60,0,sqrt(exp(log(2)+.8*muscle$Conc)))

##Note: use more MCMC chains (i.e NC=10000) for more accurate results.
m2b<-bnlr(y=muscle$Length,f1=fmu,f2=fsgma,x=muscle$Conc,
z=matrix(rep(1,length(muscle$Length)),ncol=1),bta0=c(20,-30,-1),gma0=2,Nc=650)
m1b<-bnlr(y=muscle$Length,f1=fmu,f2=fsgma,x=muscle$Conc,z=cbind(1,muscle$Conc),
bta0=c(20,-30,0),gma0=c(.5,.5),Nc=650)

chainsum(m1b$chains,burn=1:65)
chainsum(m2b$chains,burn=1:65)
infocrit(m1b,1:65)
infocrit(m2b,1:65)