Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: RE: -weibullfit-, -qweibull-, -pweibull- from SSC


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: RE: -weibullfit-, -qweibull-, -pweibull- from SSC
Date   Tue, 20 Nov 2007 00:12:44 -0000

You are correct. Such distributions would be a challenge. Others can 
speak for themselves, but I think they would require a completely
different 
methodology from the family of programs described here, and they are 
not on any to do list of mine. 

Nick
n.j.cox@durham.ac.uk 

Feiveson, Alan H. (JSC-SK311)

Thanks, Stephen and Nick -  I agree it's nice to have a way of fitting a
Weibull model if you don't have censoring without bothering with _st
commands. 

Since you guys are doing so well, I have a new challenge. Have you or
anyone you know of, considered writing a Stata program for fitting Levy
Stable distributions? These are useful for modeling "heavy" tails and
have the nice property that they are closed under linear combinations.
For example, see
http://en.wikipedia.org/wiki/Levy_skew_alpha-stable_distribution

If you go to John Nolan's web page,
http://academic2.american.edu/~jpnolan/stable/stable.html you will see
he already describes a number of routines for fitting stable
distributions, generating random variates, etc. but to my knowledge, we
have no such thing in Stata. These distributions usually (a Cauchy is a
special case) don't have closed-form expressions for the density or CDF,
so most analysis must be done by inverting their characteristic
function, which does have a closed form. Sounds like a great project for
Mata afficionados.

Nick Cox

Thanks to Kit Baum, a new package -weibullfit- is available from SSC.
Stata 8.1 is required. Use -ssc- to install if interested. 

This package is written by Stephen Jenkins and myself. 

The main program -weibullfit- fits a two-parameter Weibull distribution
by maximum likelihood, optionally as dependent on covariates. 

In addition, existing packages -qweibull- and -pweibull- that plot q-q
and p-p plots for observed data versus results from a fitted Weibull by
maximum likelihood have been revised so that they call -weibullfit-. 

Some of you will know that existing Stata command -streg- offers support
for Weibull distributions, and/or that the existing (but now not
documented) command -weibull- does so too. The main reasons for a
separate -weibullfit- are 

1. Its support of a different parameterisation, one that is popular in
some sciences. It is relatively easy to relate this to Stata's favourite
parameterisation, with or without -nlcom-, but nothing

beats direct support for convenience. 

2. Its support for at least some models not available under -streg-. 

3. Its independence of Stata's survival time facilities. Conversely,
-weibullfit- is innocent of survival complexities and has nothing to
offer -st- users. 

4. Its link to -qweibull- and -pweibull-. 

5. Most users would find -weibull- difficult to use in the absence of
documentation. 

-wbull-, dating from 1998, remains a stand-alone package on SSC for
fitting Weibull distributions (without covariates). 
Only Stata 5 is required for that. 
 
-weibullfit- is the latest in a series of packages for fitting
distributions, optionally as dependent on covariates, using maximum
likelihood. Authors are Stephen Jenkins, Maarten Buis, Philippe Van Kerm
and myself. Distributions treated similarly are 

beta
Dagum
Dirichlet
gamma
generalised beta of the second kind
Gumbel
inverse gamma
inverse Gaussian
lognormal
Singh-Maddala
Type I Pareto

I suspect a few others may follow intermittently as the mood takes one
or more of us. 

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index