calc.RL.0 {mixstock}R Documentation

Use Raftery and Lewis diagnostics to calculate MCMC chain lengths

Description

Uses {tt coda}'s raftery.diag implementation of the Raftery and Lewis diagnostic to estimate minimum chain lengths for an MCMC estimate for mixed stock analysis. Runs R&L iteratively until the criteria are satisfied.

Usage

calc.RL.0(data, startfval, pilot=500, maxit=15, verbose=FALSE,
rseed=1001, debug=FALSE)
calc.mult.RL(data,n=50,debug=FALSE,verbose=FALSE)
RL.max(r)

Arguments

data a mixstock.data object, or any list with sourcesamp and mixsamp entries containing source and mixed stock data
startfval starting value of contribution frequencies for the chain, as in gibbs or gibbsC: NULL=random start, 0=equal contributions from all sources, (1..R-1)=95% contribution from one source, with the rest splitting the remainder equally
pilot Chain length to start with (length of "pilot" run)
maxit Max. number of iterations of the Raftery and Lewis procedure
verbose Produce lots of output?
rseed Random-number seed
debug produce debugging output?
n number of different random-number seed chains to try
r the results of a Raftery and Lewis diagnostic test

Details

calc.RL.00 starts by running a Gibbs-sampler chain with the length given by pilot, then repeatedly lengthens the chain until the length is greater than that suggested as the total by the Raftery and Lewis diagnostic. (The next suggested step in the procedure is to run multiple chains of this length and see whether they pass the Gelman and Rubin diagnostic.) calc.mult.RL runs the Raftery and Lewis calculation multiple times, starting each chain from a large contribution from each source in turn, to see if some starting configurations are slower to converge or if there is a lot of variation among chains with different random number seeds. RL.max picks the expected maximum chain length given a set of diagnostics; RL.burn returns the predicted burn-in required.

Value

for calc.RL.00:

current Results of the Raftery and Lewis test on the current iteration
history History of the iterations:

for calc.mult.RL, a matrix giving the maximum expected chain length for each random-number seed/starting point combination

Author(s)

Ben Bolker

Examples

data(bolten98)
b98c <- markfreq.condense(as.mixstock.data(bolten98))

[Package mixstock version 0.9 Index]