ICE (version 0.69)

ickde: Interval-Censored Kernel Density Estimation

Description

Iterated conditional expectation kernel density estimation using a local constant. The bandwidth is assumed fixed. (See the example for a way to get a quick ballpark estimate of the bandwidth.) The gaussian, epanechnikov and biweight kernels can be used. Note that the bandwidth estimate would have to be adjusted before using with epanechnikov or biweight.

Usage

ickde(I, h, f, m, n.iterations = 10, x1, xm, right.limit = 10000,kernel="gaussian", old=TRUE)

Arguments

I

A matrix with two columns, consisting of left and right endpoints of the interval data

h

A scalar bandwidth

f

An initial estimate of the density at a sequence of grid points (optional; if this is used, do not specify m)

m

The number of (equally-spaced) grid points at which the density is to be estimated

n.iterations

The maximum number of iterations allowed

x1

The left-most grid point (optional)

xm

The right-most grid point (optional)

right.limit

For right-censored data, the value to be used as an artificial right endpoint for the intervals

kernel

character argument indicated choice of kernel; current choices are "gaussian", "epanechnikov", "biweight"

old

logical value, indicating whether denominators in conditional expectation calculation should use the previous value of the density estimate.

Value

IC

References

Braun, J., Duchesne, T. and Stafford, J.E. (2005) Local likelihood density estimation for interval censored data. Canadian Journal of Statistics 33: 39-60.

Examples

 tmp <- apply(ICHemophiliac, 1, mean)
 h <- try(dpik(tmp), silent=T) # dpik() will work if KernSmooth is loaded
 if (class(h) !="numeric" ) h <- .9  # this makes the example work 
                       # if KernSmooth is not loaded
 estimate <- ickde(ICHemophiliac, m=200, h=h)
 plot(estimate, type="l")