If your dependent variable is binary and your dataset is
longitudinal--and it sounds like it is--you also might
consider split-population modeling, which treats the prior
problem of exposure to risk as an element to be estimated,
and then controls for the estimated likelihood of membership
in the at-risk group when estimating parameters affecting
the hazard of the event's occurrence.
For the preliminaries on split-population hazard modeling,
see Box-Steffensmeier and Jones's _Event History Modeling: A
Guide for Social Scientists_ (2004: 148-154). For an
interesting application to the occurrence of interstate war,
see David Clark and Patrick Regan, "Opportunities to Fight:
A Statistical Technique for Modeling Unobservable
Phenomena," Journal of Conflict Resolution, Vol. 47, No. 1
(February 2003): 94-115.
Have fun,
Jay
Jay Ulfelder, Ph.D.
Research Director
Political Instability Task Force
http://globalpolicy.gmu.edu/pitf/index.htm
Science Applications International Corporation (SAIC)
(301) 588-8478
[log in to unmask]
----- Original Message -----
From: "Jesse T Richman" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Wednesday, January 30, 2008 10:38 AM
Subject: Re: [POLMETH] Question about Rare Events Logit and
Zero-Inflated Probit Estimator
> Political Methodology Society <[log in to unmask]>
wrote on
> 01/29/2008 10:15:47 AM:
>
> > Hi
> >
> > My first question is does anyone know if rare events
logit (relogit)
> > takes into account the process that generates the zeros
for certain
> > events? For example, some countries have zero for civil
war because
> > they are highly unlikely to have a civil war while
others just had a
> > good decade.
>
> You might want to consider checking for
heteroskedasticity. If this turns
> out to be a problem, STATA does have a packaged
heteroskedastic probit
> approach.
>
> But it sounds like you may have a theory that
distinguishes between
> different kinds of 'no civil war events'. Maybe it is the
following: some
> countries don't have hot civil wars, but civil wars are
likely at any
> moment because there is a residual level of inter-group
violence. Perhaps
> you can recode the dependent variable with three or more
values, leading
> towards a different model that includes zero-inflation, or
more likely an
> ordered probit model that obviates the need for a specific
zero-inflation
> model.
>
> Jesse Richman
> Assistant Professor of Political Science
> Old Dominion University
>
> **********************************************************
> Political Methodology E-Mail List
> Editors: Melanie Goodrich, <[log in to unmask]>
> Delia Bailey, <[log in to unmask]>
> **********************************************************
> Send messages to [log in to unmask]
> To join the list, cancel your subscription, or modify
> your subscription settings visit:
>
> http://polmeth.wustl.edu/polmeth.php
>
> **********************************************************
**********************************************************
Political Methodology E-Mail List
Editors: Melanie Goodrich, <[log in to unmask]>
Delia Bailey, <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|