lplikint {lpint}

Partial likelihood based local polynomial estimators of the counting process intensity function and its derivatives

Description

This local polynomial estimator is based on the (localized) partial likelihood

Usage

lplikint(jmptimes, jmpsizes = rep(1, length(jmptimes)),
         Y = rep(1,length(jmptimes)),
         K = function(x) 3/4 * (1 - x^2) * (x <= 1 & x >= -1),
         bw, adjust = 1, nu = 0, p = 1, Tau = 1, n = 101,
         tseq = seq(from = 0, to = Tau, length = n), tol = 1e-05,
         maxit = 100, us = 10, gd = 5) 

Arguments

jmptimes

a numeric vector giving the jump times of the counting process

jmpsizes

a numeric vector giving the jump sizes at each jump time. Need to be of the same length as jmptimes

Y

a numeric vector giving the value of the exposure process (or size of the risk set) at each jump times. Need to be of the same length as jmptimes

K

the kernel function

bw

a numeric constant specifying the bandwidth used in the estimator. If left unspecified the automatic bandwidth selector will be used to calculate one.

adjust

a positive constant giving the adjust factor to be multiplied to the default bandwith parameter or the supplied bandwith

nu

the degree of the derivative of the intensity function to be estimated. Default to 0 for estimation of the intensity itself.

p

the degree of the local polynomial used in constructing the estimator. Default to 1 plus the degree of the derivative to be estimated

Tau

a numric constant >0 giving the censoring time (when observation of the counting process is terminated)

n

the number of evenly spaced time points to evaluate the estimator at. Not used when tseq is provided.

tseq

the time sequence at which to evaluate the estimator

tol

the parameter error tolerance used to stop the iterations in optimizing the local likelihood

maxit

maximum number of iterations allowed in the optimization used in a single estimation point

us

a numeric constants used together with gd to grid search for a decent start value in solving the local score equation. The starting value is 0 except on its first dimension, which was chosen so that the starting value is the minimizer of the L^1 norm of the score function among the values: average intensity X us^(-gd:gd)

gd

a numeric constant used together with us to search for a decent start value in solving the local score equation

Details

The estimator is based on solving the local score equation using the Newton-Raphson method and extract the appropriate dimension.

Value

a list containing

x

the vector of times at which the estimator is evaluated

y

the vector giving the values of the estimator at times given in x

se

the vector giving the standard errors of the estimates given in y

bw

the bandwidth actually used in defining the estimator equal the automatically calculated or supplied multiplied by adjust

fun

the intensity (or derivative) estimator as a function of the estimation point, which can be called to evaluate the estimator at points not included in tseq

Author(s)

Feng Chen <[e-mail].>

References

Chen, F. (2011) Maximum local partial likelihood estimators for the counting process intensity function and its derivatives. Statistica Sinica 21(1): 107 -128. http://www3.stat.sinica.edu.tw/statistica/j21n1/J21N14/J21N14.html

See Also

lpint

Examples

##simulate a Poisson process on [0,1] with given intensity
int <- function(x)100*(1+0.5*cos(2*pi*x))
censor <- 1
set.seed(2)
N <- rpois(1,150*censor);
jtms <- runif(N,0,censor);
jtms <- jtms[as.logical(mapply(rbinom,n=1,size=1,prob=int(jtms)/150))];

##estimate the intensity
intest <- lplikint(jtms,bw=0.15,Tau=censor)
#plot and compare
plot(intest,xlab="time",ylab="intensity",type="l",lty=1)
curve(int,add=TRUE,lty=2)