[R] AFT model

[Thomas Lumley]

On Tue, 17 Mar 2009, tsn4867 wrote:

>    Hi,
> In the package survival, using the function survreg for AFT model, I only see 
> 4 distributions for the response y:  weibull, gaussian, logistic, lognormal and 
> log-logistic, which correspond to certain distributions for the error terms. 
> I'm wondering if there is a package or how to obtain the parameter estimates 
> (the beta's are of great interest) from the AFT model (maximizing 
> log-likelihood with censored data as in Klein and Moeschberger book) if we have
> 1) error ~ Cauchy(0,1)
> 2) error comes from a contaminated gaussian dist., for example,
> density of error, f(x) = 0.9*phi(x, mean=0, sd=1) + 0.1*phi(x, mean=0, sd=3), 
> where phi(.) is the standard gaussian pdf.

If you look at
    ?survreg.distributions
it has an example showing how to specify your own distributions.

      -thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle