Statalist The Stata Listserver


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

RE: st: left censoring in discrete-time duration model


From   "Stephen P. Jenkins" <stephenj@essex.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: left censoring in discrete-time duration model
Date   Fri, 17 Nov 2006 08:56:53 -0000

> Date: Thu, 16 Nov 2006 13:57:13 -0500
> From: Daniel Simon <dhs29@cornell.edu>
> Subject: RE: st: left censoring in discrete-time duration model
> 
> Stephen - first let me say that I have  consulted your 
> website materials 
> for a variety of issues and found them extremely useful. 
> Regarding your 
> comment, this is part of what has had me confused: whether 
> it's truncation 
> or censoring.  As you suggest, magazines become at risk of 
> establishing a 
> website "the later of either the year web
> technology became available or the year when the firm itself was 
> established." But, in some cases, they establish a website 
> before I can 
> observe it. So, this sounds like truncation. If so, my 
> question is: using 
> your easy-estimation methods, which observations do I throw away. In

> particular, in my case, do I throw away all observations for 
> 1996? 1996 is 
> the first year of the analysis period because it is the first 
> year in which 
> I observe which magazines have websites. But, I have no way 
> to distinguish 
> those magazines that had websites before 19996 and those that 
> did not. 
> After 1996, I can tell which magazines adopted new websites 
> (by comparing 
> whether they already had one the year prior). I hope this 
> makes sense. 
> thanks again for the help. Daniel


"establish a website before I can observe it" is a classic case of
left-censoring -- see e.g. Klein & Moeschberger, _Survival Analysis_,
who also provide an expression for the likelihood contribution in this
case.  

This definition of left-censoring differs a bit from the one in common
use in the social sciences -- left-censoring referring to all cases in
which the date of the start of the spell is unobserved.  In this
situation, it is often recommended to drop these cases -- implicitly
appealing to some sort of independence assumption. After all, what
else can one do without the information? But there are potential
problems -- of the sort raised by Jason Yackee <jyackee@law.usc.edu>
-- if spell length (time to event) and start date are systematically
related.

You may have left-censoring (classical sense) and left truncation.
Whatever the case, estimating a model incorporating unobserved
heterogrenity (frailty) is probably harder than you'd hoped. As I
said, I doubt whether an 'easy estimation' method is available.

Stephen
-------------------------------------------------------------
Professor Stephen P. Jenkins <stephenj@essex.ac.uk>
Institute for Social and Economic Research
University of Essex, Colchester CO4 3SQ, U.K.
Tel: +44 1206 873374.  Fax: +44 1206 873151.
http://www.iser.essex.ac.uk  
Survival Analysis using Stata:
http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/ 
Downloadable papers and software: http://ideas.repec.org/e/pje7.html



*
*   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