embryogrowth (version 5.1)

tsd_MHmcmc: Run the Metropolis-Hastings algorithm for tsd

Description

Run the Metropolis-Hastings algorithm for tsd. Deeply modified from a MCMC script by Olivier Martin (INRA, Paris-Grignon). The number of iterations is n.iter+n.adapt+1 because the initial likelihood is also displayed. I recommend that thin=1 because the method to estimate SE uses resampling. If initial point is maximum likelihood, n.adapt = 0 is a good solution. Note that resultLnL has all the likelihoods, not only those defined by n.adapt and thin. To get the SE from result_mcmc <- embryogrowth_MHmcmc(result=try), use: result_mcmc$BatchSE or result_mcmc$TimeSeriesSE The batch standard error procedure is usually thought to be not as accurate as the time series methods. Based on Jones, Haran, Caffo and Neath (2005), the batch size should be equal to sqrt(n.iter). Jones, G.L., Haran, M., Caffo, B.S. and Neath, R. (2006) Fixed Width Output Analysis for Markov chain Monte Carlo , Journal of the American Statistical Association, 101:1537-1547. coda package is necessary for this function.

Usage

tsd_MHmcmc(result = stop("An output from tsd() must be provided"),
  n.iter = 10000,
  parametersMCMC = stop("Priors must be given. Use tsd_MHmcmc_p()"),
  n.chains = 1, n.adapt = 0, thin = 1, trace = FALSE,
  batchSize = sqrt(n.iter))

Arguments

result

An object obtained after a SearchR fit

n.iter

Number of iterations for each step

parametersMCMC

A set of parameters used as initial point for searching with information on priors

n.chains

Number of replicates

n.adapt

Number of iterations before to store outputs

thin

Number of iterations between each stored output

trace

True or False, shows progress

batchSize

Number of observations to include in each batch fo SE estimation

Value

• A list with resultMCMC being mcmc.list object, resultLnL being likelihoods and parametersMCMC being the parameters used

Details

tsd_MHmcmc runs the Metropolis-Hastings algorithm for tsd (Bayesian MCMC)

Examples

library(embryogrowth)
eo <- subset(STSRE_TSD, Species=="Emys orbicularis", c("Males", "Females",
                                       "Incubation.temperature"))
eo_logistic <- tsd(eo)
pMCMC <- tsd_MHmcmc_p(eo_logistic, accept=TRUE)
# Take care, it can be very long; several days
result_mcmc_tsd <- tsd_MHmcmc(result=eo_logistic,
		parametersMCMC=pMCMC, n.iter=10000, n.chains = 1,
		n.adapt = 0, thin=1, trace=TRUE)
# summary() permits to get rapidly the standard errors for parameters
summary(result_mcmc_tsd)
# They are store in the result also. Two SE are estimated using or
# batch method or time-series SE:
# The batch standard error procedure is usually thought to be not
# as accurate as the time series methods.
se1 <- result_mcmc_tsd$BatchSE
se2 <- result_mcmc_tsd$TimeSeriesSE
plot(result_mcmc_tsd, parameters="S", scale.prior=TRUE, xlim=c(-3, 3), las=1)
plot(result_mcmc_tsd, parameters="P", scale.prior=TRUE, xlim=c(25, 35), las=1)
plot(eo_logistic, se=se2)