svylme (version 1.5-1)

boot2lme: Resampling variances for svy2lme

Description

Computes variance estimates for the weighted-pairwise-likelihood linear mixed models fitted by svy2lme using replicate weights. The replicate weights for a pair are obtained by dividing by the sampling weight and then multiplying by the replicate weight. There will be a warning if the ratio of replicate weight to sampling weight differs for observations in the same pair.

Usage

boot2lme(model, rdesign,  verbose = FALSE)
# S3 method for boot2lme
vcov(object,
   parameter=c("beta", "theta","s2", "relSD" ,"SD","relVar","fullVar"),
   ...)

Value

For boot2lme, an object of class boot2lme with components

theta

replicates of variance parameters (on Cholesky square root scale)

beta

replicates of regression parameters

D

replicates of relative variance matrix

scale,rscales

from the input

formula

model formula from the input

For the vcov method, a variance matrix.

Arguments

model

A model returned by svy2lme with the devfun=TRUE option

rdesign

replicate-weights design corresponding to the design used to fit the model, see example

verbose

print progess information?

object

returned by boot2lme

...

for method compatibility

parameter

Variance matrix for: regression parameters, relative variance parameters on Cholesky square root scale, residual variance, relative standard errors of random effects, standard errors of random effects, entire relative variance matrix, entire variance matrix

Details

The variance is estimated from the replicates thetastar and original point estimate theta as scale*sum(rscales* (thetastar-theta)^2).

Examples

data(api, package="survey")

# two-stage cluster sample
dclus2<-svydesign(id=~dnum+snum, fpc=~fpc1+fpc2, data=apiclus2)

m0<-svy2lme(api00~(1|dnum)+ell+mobility, design=dclus2,return.devfun=TRUE)
jkdes<-as.svrepdesign(dclus2, type="mrb")
jkvar<-boot2lme(m0,jkdes)

SE(jkvar, "beta")
SE(jkvar, "SD")
SE(jkvar,"s2")


m1<-svy2lme(api00~(1|dnum)+ell+mobility,
design=dclus2,return.devfun=TRUE, all.pairs=TRUE, subtract.margins=TRUE)
jk1var<-boot2lme(m1,jkdes)

SE(jk1var, "beta")
SE(jk1var, "SD")


# \donttest{
##takes a few minutes
data(pisa)

pisa$w_condstuwt<-with(pisa, w_fstuwt/wnrschbw)
pisa$id_student<-1:nrow(pisa)

dpisa<-survey::svydesign(id=~id_school+id_student, weight=~wnrschbw+w_condstuwt, data=pisa)

m<-svy2lme(isei~(1+female|id_school)+female+high_school+college+one_for+both_for+test_lang,
	design=dpisa, return.devfun=TRUE,method="nested")

bpisa<-as.svrepdesign(dpisa, type="bootstrap", replicates=100)

b<-boot2lme(m, bpisa, verbose=TRUE)
str(b)

vcov(b,"beta")
vcov(b,"s2")

## SE() inherits the parameter= argument
SE(b,"beta")
SE(b,"theta")
SE(b,"SD")

# }