[R] Survreg function for loglogistic hazard estimation

[chenjiakai]

Dear Terry,

Thanks a lot for your reply.  The result of survreg reads like: 

Call:
survreg(formula = Surv(end, status) ~ fico_demean, data = raw, 
    dist = "loglogistic", model = TRUE)
               Value Std. Error      z         p
(Intercept)  2.98365   1.34e-03 2233.2  0.00e+00
fico_demean -0.00183   5.36e-05  -34.2 4.18e-256
Log(scale)  -1.41070   2.65e-03 -532.9  0.00e+00

So given your transform the intercept term and log(scale) matches the a and
b respectively - thanks a lot.  The effect of fico_demean, however, still
seems not close.  I will try to find the K&P book. 

Yes, the function which(rnos <= bhaz * exp(dfico[i]*dficoeff))[1] is
identical with min(which(rnos <= bhaz * exp(dfico[i]*dficoeff))), because
the vector of subscription returned by 'which' is always ascending.  bhaz is
not a constant but a vector of baseline hazard rate, so 'which' will do a
element by element comparison.  If bhaz is a constant, then endtime[i] will
follow a geometric distribution. 

I appreciate your help.  Thanks a lot.

Jiakai Chen


Terry Therneau wrote:
> 
> 
>  The survreg function uses the location-scale parameterizations found in 
> Kalbfleisch and Prentice's book "The statistical analysis of failure time
> data". 
>  Looking at your code, you are using something different
>  
>   You     K&P
>   1/a     lambda
>   b       gamma
>   
> And from K&P: lambda = exp(-alpha), gamma= 1/sigma, where
>                alpha = linear predictor from the fit
> 	       sigma = scale factor from the fit
> I would suggest checking out a copy of the book to verify the above.	       
> 	       
>  Second, your method of constructing a random time from the distribution
> is a 
> mystery to me.  I believe your code is equivalent to
>       y <- min(which(runif(n) < constant))
> for any given y (with changing values of the constant).
> This is a value from the geometic distribution.  
> 
> Terry Therneau
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> 

-- 
View this message in context: http://www.nabble.com/Survreg-function-for-loglogistic-hazard-estimation-tp23907598p23927718.html
Sent from the R help mailing list archive at Nabble.com.