In stc04003/tranSurv: Transformation Model Based Estimation of Survival and Regression Under Dependent Truncation and Independent Censoring

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = NA, 
  prompt = TRUE,
  fig.path = "README-"
)

Install tranSurv package from GitHub using

devtools::install_github("stc04003/tranSurv")

Load tranSurv and copula packages.

library(tranSurv)
library(copula)

Generating correlated data:

set.seed(1)
rho <- iTau(normalCopula(dim = 2), .5) ## convert kendall's tau to pearson's rho
u <- rCopula(2000, normalCopula(rho, dim = 2)) ## generate correlated data
u <- qweibull(u, 2, 1) ## assumes t1 and t2 follows some weibull distribution
colnames(u) <- c("t1", "t2")

This gives a Kendall's tau of 0.5 (between t1 and t2)

cor(u, method = "kendall")
kendall(u)

Now apply the truncation

u <- subset(as.data.frame(u), t1 < t2) 
dim(u) ## ~50% truncation rate because t1 and t2 follows the same distribution
head(u)

Arguments for wKendall:

args(wKendall)

When there is no perturbation weights, wKendall is equivalent to condKendall.

attach(u)
condKendall(t1, t2)$PE
wKendall(t1, t2)
detach(u)

wKendall with perturbation weight, which assumes to be a standard exponential distribution.

with(u, wKendall(t1, t2, NULL, rexp(length(t1))))

Use perturbation weights for standard error estimation:

attach(u)
set.seed(2)
sd(replicate(500, wKendall(t1, t2, NULL, rexp(length(t1))))) ## 0.018
condKendall(t1, t2)$SE ## 0.208
detach(u)

Small scale simulation:

do <- function() {
    u <- rCopula(2000, normalCopula(rho, dim = 2)) ## generate correlated data
    u <- qweibull(u, 2, 1) ## assumes t1 and t2 follows some weibull distribution
    colnames(u) <- c("t1", "t2")
    u <- subset(as.data.frame(u), t1 < t2)
    out <- with(u, c(sd(replicate(500, wKendall(t1, t2, NULL, rexp(length(t1))))) ,
                     condKendall(t1, t2)$SE))
    out
}

set.seed(3)
rowMeans(replicate(100, do())) 

stc04003/tranSurv documentation built on July 21, 2021, 6:46 p.m.