SEsamples: Bootstrap Samples for Simple 2-State Model

In hmmr: "Mixture and Hidden Markov Models with R" Datasets and Example Code

Description

Parametric bootstrap samples for a 2-state hidden Markov model used to compute standard errors.

Usage

data("SEsamples")

Format

A matrix with 1000 rows for each sample, 12 columns for each parameter of the model, including the parameters that are fixed at their boundary values.

Details

The bootstrap sample was generated by the following code:

require(depmixS4)
library(hmmr)
data(simplehmm)

# define the model
set.seed(214)
mod1 <- depmix(obs~1,data=simplehmm,nstates=2,
	family=multinomial("identity"), respst=c(.6,0,.4,0,.2,.8), trst=runif(4), inst=c(1,0))

# fit the model
fm1 <- fit(mod1,emcontrol=em.control(random.start=FALSE))

# compute bootstrap samples
nsamples <- 1000
SEsamples <- matrix(0,ncol=npar(fm1),nrow=nsamples)

for(i in 1:nsamples) {
	sample <- simulate(fm1)
	fmsam <- fit(sample,emcontrol=em.control(random.start=FALSE))
	SEsamples[i,] <- getpars(fmsam)
}

Examples

data(SEsamples)
# standard errors
bootses <- apply(SEsamples,2,sd)
bootses[which(bootses==0)] <- NA
bootses
# compare with standard errors from finite differences
library(hmmr)
data(simplehmm)
# define the model
set.seed(214)
mod1 <- depmix(obs~1,data=simplehmm,nstates=2,
	family=multinomial("identity"), respst=c(.6,0,.4,0,.2,.8), trst=runif(4), inst=c(1,0))
# fit the model
fm1 <- fit(mod1,emcontrol=em.control(random.start=FALSE))
ses <- cbind(standardError(fm1),bootses)
ses

hmmr documentation built on May 27, 2021, 9:10 a.m.