[R] ex-Gaussian survival distribution

Thomas Lumley tlumley at u.washington.edu
Wed Mar 1 16:21:17 CET 2006


On Tue, 28 Feb 2006, Jonathan Williams wrote:

> Dear R-Helpers,
>
> I am hoping to perform survival analyses using the "ex-Gaussian"
> distribution.
> I understand that the ex-Gaussian is a convolution of exponential and
> Gaussian
> distributions for survival data.
>
> I checked the "survreg.distributions" help and saw that it is possible to
> mix
> pre-defined distributions. Am I correct to think that the following code
> makes
> the ex-Gaussian:-
>
> exGauss=survreg.distributions$exponential
> exGauss$name='exGaussian'
> exGauss$dist=survreg.distributions$gaussian

No. That just produces an object with some pieces from the lognormal and 
some from the exponential model.  All the functions described in 
help(survreg.distributions) have to be written to make a new distribtuion

Also, survreg.distributions will only handle location-scale families, and 
I don't think the "ex-Gaussian" is one of these.

> Am I further correct to think that I can compare the fits of the ex-Gaussian
> and Weibull distributions to the data via:-
>
> fit1=survreg(Surv(response)~1+frailty(unit), data=dat, dist=exGauss)
> fit2=survreg(Surv(response)~1+frailty(unit), data=dat, dist='weibull')
> anova(fit1, fit2)

No. Even if the models had been fitted correctly they aren't nested 
models.

 	-thomas

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


> Finally, am I further correct to think that the output from this anova means
> that
> the Weibull distribution fits the data worse than the exGauss distribution
> that I
> made?
>
>              Terms Resid. Df    -2*LL Test       Df Deviance P(>|Chi|)
> 1 1 + frailty(unit)  4229.778 63129.46            NA       NA        NA
> 2 1 + frailty(unit)  4228.020 58426.27    = 1.757815 4703.190         0
>
> Many thanks for your help with these questions. I have a feeling they are
> trivial,
> but I am a psychiatrist so I need to check that I am not barking up the
> wrong tree
> (or simply barking...)!
>
> Jonathan Williams
>
> PS why does "weibull" need quotes in the survreg procedure, while exGauss
> does not?
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>

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




More information about the R-help mailing list