Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: correct for selection bias in survival analysis
From
"Frederick J. Boehmke" <[email protected]>
To
[email protected]
Subject
Re: st: correct for selection bias in survival analysis
Date
Tue, 15 Mar 2011 08:29:09 -0500 (CDT)
I know this response is terribly late, but since people often find
these threads through web searches, probably better late than never.
For duration and selection in Stata, try -net search dursel-:
-dursel- allows the user to estimate exponential, Weibull or lognormal
duration models accounting for potential non-ignorable selectivity using
maximum likelihood techniques. These models are analogous to Heckman
models for OLS regression.
Someone mentioned Prieger's article already, and here is a second from
which we developed the above command:
Boehmke, Frederick J., Daniel Morey and Megan Shannon. 2006. "Selection
Bias and Continuous-Time Duration Models: Consequences and a Proposed
Solution." American Journal of Political Science 50 (1): 192-207.
(Now back to catching up on my Stata list reading.)
On Wed, 5 May 2010, Antoine Terracol wrote:
Date: Wed, 5 May 2010 06:06:55
From: Antoine Terracol <[email protected]>
Reply-To: [email protected]
To: [email protected]
Subject: Re: st: correct for selection bias in survival analysis
Bonjour Marguerite,
as Maarten said earlier, including the IMR will not work in your setting
because the second stage is not linear, and also because the IMR rests on the
bivariate normality assumption, and your duration model has a log-logistic
distribution.
To my knowledge, there is no "simple" (as in "a couple of lines of code") way
to correct for selection biais in duration models in Stata.
You will need to write your own likelihood function, the simplest way to do
it would be to assume a log-normal distribution for the duration model, and a
Probit for the selection stage, taking advantage of the bivariate normality
of your error terms. For more general setups, you may want to have a look at
Prieger, J. E. "A flexible parametric selection model for non-normal data
with application to health care usage" Journal of Applied Econometrics, 2002,
17(4), 367-392
In addition, it seems you actually have a kind of double hurdle model with
two successive selection stages: having had a civil war and, conditional on
civil war, that peace was settled.
Given your research question, I assume you do not have many data points, so
you should try to stick to the simplest setup possible, and forget about the
double selection...
Antoine
PS: I have some old code for the log-normal duration model with selection,
feel free to send me an email privately if you're interested in it (ou de
passer à la MSE dans le courant de la semaine prochaine)
On 05/05/2010 12:44, Marguerite Duponchel wrote:
I am working on the duration of post civil war peace so I want to correct
for
the fact that my sample is restricted to countries who 1/had a civil war,
2/
the peace was settled.
Thanks.
Quoting Maarten buis<[email protected]>:
--- On Wed, 5/5/10, Marguerite Duponchel wrote:
It's just that as I am the one creating the selection bias
on a full available sample, I thought there might be a way
to correct it while using the info in the complete data
which frailty would not integrate.
The easiest solution is not to make the selection or, if you
want to work with a smaller dataset, make the selection
random. Otherwise I don't understand why you would want to
make a selection and than "correct" for the fact that you
made a selection.
Hope this helps,
Maarten
--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://www.maartenbuis.nl
--------------------------
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
--
Ce message a ete verifie par MailScanner
pour des virus ou des polluriels et rien de
suspect n'a ete trouve.
--
Marguerite Duponchel
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
- Fred
< ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Frederick J. Boehmke
< Associate Professor of Political Science
> University of Iowa
<
> 308 Schaeffer
< 319-335-2342
> http://www.fredboehmke.net
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^