[R] R function for censored linear regression

Andreas Wittmann andreas_wittmann at gmx.de
Tue Sep 24 11:34:00 CEST 2013


Dear R-useRs,

I'm looking for an R-function for censored linear regression. I have the 
following data

x1 <- rnorm(100)
x2 <- rnorm(100)
y <- x1 + 2*x2 + rnorm(100,0,0.5)
stat <- rep(1,100)
stat[50:100] <- 0
data <- data.frame(y,x1,x2,stat)

y is the dependent variable, x1 and x2 are the independent variables in 
a linear model. the variable y could be right-censored, this information 
is in the variable stat, where 1 denotes observed and 0 denotes 
censored. If stat is 0, then the value in y is the observed 
right-censored value and could be greater. Using the Tobit-model would 
not be the right thing here because the Tobit model assumes the same 
limit for all observations, in my data each value of y[50:100] could 
have a different limit.

If i use linear regression

lm1 <- lm(y ~ x1 + x2, data=data)
summary(lm1)

the censoring is not incorporated, so my idea is to use survreg from the 
survival package

library(survival)
s1 <- survreg(Surv(y, stat) ~ x1 + x2, data, dist='gaussian')
summary(s1)

my question is, is this the right approach for my aim? Is it right, that 
here each censored observations could have its own limit?

Thanks and best regards
Andreas



More information about the R-help mailing list