Statalist


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

RE: st: RE: tobit or anything else?


From   "Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: tobit or anything else?
Date   Sun, 13 Apr 2008 11:56:09 -0400

Berthold Hoppe wrote:

>>you are completely right, it is right skewed. I will try it with - 
intcens-.
As this command needs a description of the distribution:
Do you know a way to choose the correct type of distribution?<<

Well, I'm going to have to give the classic weasel answer of "it depends." Much depends on the shape of the distribution at the left. If the distribution is L-shaped, i.e., has a mode at 0, some models can accommodate this and others cannot. As it so happens, I'm writing a paper on this topic. In general I've found that there are reasons due to the nature of the likelihoods to prefer distributions such as the log-logistic or gamma to the log-normal. If you wouldn't mind sharing your data (properly credited), I'll be glad to take a look. A relatively safe method is to try all the distributions and see if it makes any difference. -intcens- won't bite. :) 

A good survival analysis book such as J. F. Lawless, (2003), Statistical Models and Methods for Lifetime Data, Wiley, has a discussion of such variables. I've also found the new book by Marshall and Olkin, Life Distributions, to be very nice. See http://www.springer.com/engineering/production+eng/book/978-0-387-20333-1. 

Jay

<<winmail.dat>>




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