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   Steven Samuels <[email protected]>
To   [email protected]
Subject   Re: st: left censoring in discrete-time duration model
Date   Tue, 28 Nov 2006 18:56:10 -0500

On Nov. 28, 2006, at 2:05 PM, Austin Nicholas wrote:

I would argue that 1992 is a natural "zero" time for Daniel Simon's
project, since I am reasonably certain no magazines had websites
before 1992-

As Austin, Stephen Jenkins, and I have suggested, Daniel is limited to analyses of firms without web sites in 1996 if he relies solely on the web archive. Now I am not sure he can even do that much with safety.

Daniel: Taking a start date of 1992, as Austin suggests, can you identify from other sources which magazines started publication after 1992 and the dates of publication? If you can, then as Stephen stated, then you can use likelihood methods. If you cannot, you have problems for which we have no good solution. It appears from your description that the web archive gives only the date of first appearance of the web-site. If so, what are you going to do about magazines which started publication after 1996? For example, does a magazine with first web-site appearance in 2005 have nine years of "risk" or did the website and magazine start simultaneously in 2005? At the very least you should try to estimate the number of such magazines.

Terminology: I disagree with Stephen that the data are "left- truncated." I believe that they are primarily left-censored with possible left-truncation. With left-truncated data the very existence of some observations is unknown because they take place too early. The classic example is the fetal life table of miscarriages which did not record miscarriages that took place prior to the start of prenatal care. In Daniel's archive, the analogous missed magazines are those which stopped publishing after 1992 and before 1996. (Here the very existence of the magazine, not just the web- site, is unknown.) For left-censored data (or for interval-censored data, in general), the analyst knows that the event took place between two dates. This is the case for the magazines in the archive with web sites in 1996, assuming that the dates for start of risk are known.


Steven Samuels
18 Cantines Island
Saugerties, NY 12477

[email protected]

* For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index