s-news
[Top] [All Lists]

- censorReg parameters

To: s-news@lists.biostat.wustl.edu
Subject: - censorReg parameters
From: AndrewGScott@aol.com
Date: Sun, 31 Dec 2006 17:14:00 -0500
Dear S-News Readers,

Just trying to confirm the output parameter for a survival curve fitted using 
censorReg and an exponential lifespan. An example is as follows:

# Set hazard rate of 0.1
lambda<-0.1

# Generate Exponential distribution of lifetimes
lifetime<-rexp(2500,lambda)

# Fit parametric survival curve
censorReg(censor(lifetime)~1,dist="exponential")

# Results
Call:
censorReg(formula = censor(lifetime) ~ 1, distribution = "exponential")

Distribution: Exponential

Coefficients:
(Intercept)
    17.5609

Dispersion (scale) fixed at 1
Log-likelihood: -42354.75

Observations: 2500 Total; 0 Censored
Parameters Estimated: 1

Question:
What is the relationship between the original hazard rate of 0.1 and the fitted 
parameter of 17.5609?

Interestingly, it seems to make more sense if a weibull distribution is used:

censorReg(censor(lifetime)~1,dist="weibull")

# Results

Call:
censorReg(formula = censor(lifetime) ~ 1, distribution = "weibull")

Distribution: Weibull

Coefficients:
(Intercept)
   2.306047

Dispersion (scale) = 0.9850537
Log-likelihood: -8248.8

Observations: 2500 Total; 0 Censored
Parameters Estimated: 2

The scale is close to 1.0 and the hazard rate of 0.1 is confirmed by 
exp(-2.306047).


Any thoughts would be appreciated.

Best Regards
Andrew G Scott

<Prev in Thread] Current Thread [Next in Thread>
  • - censorReg parameters, AndrewGScott <=