BayHaz (version 0.1-3)

BPSpostSample: Function to Sample Hazard Rates from BPS Posteriors

Description

A function to generate a random sample of hazard rates from the posterior distribution originated by a first order autoregressive BPS prior through the observation of a sequence of possibly right censored times to event.

Usage

BPSpostSample(hyp, times, obs = NULL, mclen = 10, burnin = 0, thin = 1, df = 10, etastar = NULL)

Arguments

hyp

list of hyperparameters (as generated by BPSpriorElicit)

times

vector of (possibly right censored) times to event

obs

vector of censoring indicators (0 = censored, 1 = exact)

mclen

requested sample size

burnin

burn-in parameter

thin

thinning parameter

df

degrees of freedom for the multivariate Student-t proposal distribution

etastar

posterior mode and corresponding hessian in list format (as generated by optim with hessian = TRUE)

Value

hyp

list of hyperparameters identifying the BPS prior that originated the posterior distribution from which the sample was extracted (copy of the input argument)

dat

dataframe with two variables (times and obs) containing the observations on which the posterior distribution is based

burnin

burn-in parameter used (copy of the input argument)

thin

thinning parameter used (copy of the input argument)

df

degrees of freedom used for the multivariate Student-t proposal distribution (copy of the input argument)

etastar

posterior mode and corresponding hessian in list format (copy of the input argument or computed via optim if the input argument was NULL)

eta

matrix with mclen rows (and length(hyp$knots)-hyp$ord columns) containing the spline weights

Details

A Markov chain sample of length mclen from the posterior distribution originated by hyp through the observation of times and obs is generated using a taylored proposal density Metropolis-Hastings sampler (starting at the posterior mode); see Chib \& Greenberg (1995).

The first burnin states of the Markov chain are discarded, then one every thin is kept.

If obs is NULL, it is assumed that all observations are exact (no censoring).

References

Chib, S. \& E. Greenberg (1995). Understanding the Metropolis-Hastings algorithm. American Statistician 49, 327--335.

See Also

BayHaz-package, BPSevalHR, BPSplotHR, BPSpost2mcmc

Examples

# set RNG seed (for example reproducibility only)
set.seed(1234)

# select a BPS prior distribution
hypars<-BPSpriorElicit(r0 = 0.1, H = 1, T00 = 50, ord = 4, G = 3, c = 0.9)
# load a data set
data(earthquakes)

# find the posterior mode
postmode<-BPSpostSample(hypars, times = earthquakes$ti, obs = earthquakes$ob, mclen = 0)
# evaluate the posterior mode hazard rate at year multiples
BPSevalHR(time = seq(0,50), sample = postmode)

# generate a posterior sample
post<-BPSpostSample(hypars, times = earthquakes$ti, obs = earthquakes$ob, etastar = postmode$etastar)
# plot some posterior hazard rate summaries
BPSplotHR(post, tu = "Year")