1. I can think of 4 plausible parameterizations for the Weibull
distribution: Pr(T>t) = exp(-(t/tau)^a), exp(-(b*t)^a), exp(-c*t^a,
exp(-(t^a)/d). To find out which is being used, consider the following:
> pweibull(a=1, shape=1, scale=1)
[1] 0.6321206
> pweibull(1, 2, 1)
[1] 0.6321206
> pweibull(2, 1, 2)
[1] 0.6321206
> pweibull(2, 2, 2)
[1] 0.6321206
Unless I'm missing something, this looks like Pr(T>q) =
exp(-(q/scale)^shape).
2. If you have censored data, then you need to look for methods
for censored data, e.g., in Venables and Ripley, Modern Applied
Statistics with S. If no observations are censored, then have you
considered "fitdistr" in library(MASS)?
3. If you can't guess "rweibull", then Help -> "Online Manuals"
-> "Guide to Statistics, vol. 1" -> Index of Search -> Weibull will
identify "rweibull" for generating pseudo-random numbers following a
Weibull distribution.
Hope this help. Bon chance.
Spencer Graves
Kamil Toth wrote:
Folks:
I have several questions regarding the Weibull distribution and its
implementation in S+
1. There are known two analytical forms of Weibull distribution.
Which form exactly is implemented in S+ and what is the meaning of
the parameter "scale" in this form?
2. What would be the best method to fit the Weibull distribution? I am
trying to avoid the method of moments, however I was not able to
obtain a fast result with the max likelihood either.
3. What is the method for generating Weibull-distributed random numbers?
Thank you
Kamil Toth
Ecole Normale Superierre
Paris, France
------------------------------------------------------------------------
Do you Yahoo!?
SBC Yahoo!
<http://pa.yahoo.com/*http://us.rd.yahoo.com/evt=24311/*http://promo.yahoo.com/sbc/>
- Internet access at a great low price.
|