The required R packges are Rcpp and RcppArmadillo. You can install these two R packages by
install.packages(c("RcppArmadillo","Rcpp"))
The necessary files are "VCM.cpp" and "Functions.R". Please download the two files to the directory of your R/Rstudio. Use the following codes to source both of them.
library(Rcpp)
library(RcppArmadillo)
sourceCpp("VCM.cpp")
source("Functions.R")
Now you are ready to fit our varying-coefficient model.
Individual.fit=IndT(u_grid, Y, U, X, W, c(a,b))
# u_grid is a grid of u-values
# Y: a size-J vector of individual-level measurements
# U: a size-J vector of indivdiual-level U-covariates
# X: a J*(p+1) matrix of indiviudal-level X-covariates
# W: a size-J vector of individual-level sampling weights
# (a,b): the cross-validation will search h within the interval (a,b)
#The output Individual.fit is a list of the selected bandwidth and the estimates
Individual.fit$h # a single value
Individual.fit$fit # a (p+1)*length(u_grid) matrix
# You can plot the estimates by
par(mfrow=c(ceiling(nrow(Individual.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Individual.fit$fit))
{
plot(u_grid,Individual.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Download the "individual.csv" file to the directory. Run the following programs.
individual.data=read.csv("individual.csv")
u_grid=seq(-1.5,1.5,length=400)
Y=individual.data$Y
U=individual.data$U
X=cbind(individual.data$X1,individual.data$X2,individual.data$X3,individual.data$X4)
W=individual.data$W
Individual.fit=IndT(u_grid,Y,U,X,W,c(0.01,1))
Individual.fit$h
par(mfrow=c(ceiling(nrow(Individual.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Individual.fit$fit))
{
plot(u_grid,Individual.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Output are
> Individual.fit$h
[1] 0.5753412
Random.fit=RT(u_grid,Z,U,X,POOLID,W,c(a,b))
# u_grid is a grid of u-values
# Z: a size-N vector of pooled-level measurements for each individual,
# if two indiviudals are in the same pool, the corresponding Z-value are the same
# U: a size-N vector of indivdiual-level U-covariates
# X: a N*(p+1) matrix of indiviudal-level X-covariates
# PoolID: a size-N vector of individuals' pool ID,
# if two individuals are in the same pool, their pool IDs are the same
# W: a size-N vector of individual-level sampling weights
# (a,b): the cross-validation will search h within the interval (a,b)
#The output Random.fit is a list of the selected bandwidth and the estimates
Random.fit$h # a single value
Random.fit$fit # a (p+1)*length(u_grid) matrix
# You can plot the estimates by
par(mfrow=c(ceiling(nrow(Random.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Random.fit$fit))
{
plot(u_grid,Random.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Download the "random.csv" file to the directory. Run the following programs.
random.data=read.csv("random.csv")
u_grid=seq(-1.5,1.5,length=400)
Z=random.data$Z
U=random.data$U
X=cbind(random.data$X1,random.data$X2,random.data$X3,random.data$X4)
W=random.data$W
PoolID=random.data$poolID
Random.fit=RT(u_grid,Z,U,X,PoolID,W,c(0.01,1))
Random.fit$h
par(mfrow=c(ceiling(nrow(Random.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Random.fit$fit))
{
plot(u_grid,Random.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Output are
> Random.fit$h
[1] 0.7529637
Homogenous.fit=HT(u_grid,Z,U,X,POOLID,W,c(a,b))
# u_grid is a grid of u-values
# Z: a size-N vector of pooled-level measurements for each individual,
# if two indiviudals are in the same pool, the corresponding Z-value are the same
# U: a size-N vector of indivdiual-level U-covariates
# X: a N*(p+1) matrix of indiviudal-level X-covariates
# PoolID: a size-N vector of individuals' pool ID,
# if two individuals are in the same pool, their pool IDs are the same
# W: a size-N vector of individual-level sampling weights
# (a,b): the cross-validation will search h within the interval (a,b)
#The output Homogeneous.fit is a list of the selected bandwidth and the estimates
Homogeneous.fit$h # a single value
Homogeneous.fit$fit # a (p+1)*length(u_grid) matrix
# You can plot the estimates by
par(mfrow=c(ceiling(nrow(Homogeneous.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Homogeneous.fit$fit))
{
plot(u_grid,Homogeneous.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Download the "homogeneous.csv" file to the directory. Run the following programs.
homogeneous.data=read.csv("homogeneous.csv")
u_grid=seq(-1.5,1.5,length=400)
Z=homogeneous.data$Z
U=homogeneous.data$U
X=cbind(homogeneous.data$X1,homogeneous.data$X2,homogeneous.data$X3,homogeneous.data$X4)
W=homogeneous.data$W
PoolID=homogeneous.data$poolID
Homogeneous.fit=HT(u_grid,Z,U,X,PoolID,W,c(0.01,1))
Homogeneous.fit$h
par(mfrow=c(ceiling(nrow(Homogeneous.fit$fit)/2),2))
par(mar=c(4,4,.1,.1))
for(k in 1:nrow(Homogeneous.fit$fit))
{
plot(u_grid,Homogeneous.fit$fit[k,],type="l",xlab="u",ylab=expression(beta(u)))
}
Output are
> Homogeneous.fit$h
[1] 0.3141536
