Re: st: Mixed Poisson models with degenerate conditional distribution for a given value of a covariate

[Alexander Limonov <alexlimonov@yahoo.com>]

Hmm, 

I'm not sure the particular package you suggest would do it, but the other component of your answer (about wage models), got me thinking -

First, a ZIP model is not appropriate, because in the ZIP model, the state (perfect or imperfect) you're in isn't actually observed, and even if it were, there is no strict segregation based on an exogenous variable.

Second, if I make the assumption that the variable in question is completely exogenous (it is not, of course), then the way to proceed would be to simply throw out the observations where the "enabling" covariate is 0 (and so the response is induced to be 0). An analogy would be - if I were studying the choice of prizes, out of some available selection, among lottery winners, I wouldn't be including the people who haven't won anything in the analysis (assuming the lottery is truly random).

However, the problem becomes interesting again because the subjects who are more likely to provide a stronger (higher) response, are also more likely to have a non-zero enabling covariate in the first place (and this covariate may further be dependent on the remaining covariates).

So...I seem to have answered my own question (the answer being what you indirectly suggested) - this is fairly similar to a wage model, where people likely to receive higher wages are more likely to be employed, and, in turn, wage determinants also determine employment. 

Does this make sense? (also, a citation or two of "classical" papers dealing with such wage models would be much appreciated)

Best,

Alexander Limonov

--- On Wed, 3/2/11, Maarten buis <maartenbuis@yahoo.co.uk> wrote:

> From: Maarten buis <maartenbuis@yahoo.co.uk>
> Subject: Re: st: Mixed Poisson models with degenerate conditional distribution for a given value of a covariate
> To: statalist@hsphsun2.harvard.edu
> Date: Wednesday, March 2, 2011, 3:30 AM
> --- On Wed, 2/3/11, Alexander Limonov
> wrote:
> > I'm working with a count data-generating process where
> the
> > dependent variable is restricted to 0, if one of the
> > covariates (let's say it's nonnegative and discrete)
> is 0.
> > This occurs in a nontrivial number of cases.
> Otherwise, the
> > dependent variable may be assumed to follow any
> conditional
> > discrete distribution of your choice, but Poisson
> could be a
> > start.
> <snip>
> > -Does the ZIP model fit the process I'm describing,
> and
> > what modifications are necessary? 
> > 
> > -If yes, is there a way to "hack" this together with
> > existing Stata procedures? 
> > 
> > -If no, what model would be appropriate?
> 
> Sounds to me similar to a selection model. The classic
> example
> would be the case where we are explaining women's income
> and
> homemakers have no income. -cmp- seems to allow for a wide
> range of selection models which could include count
> models.
> You can download that package by typing in Stata 
> -ssc install cmp-, a paper explaining it can be found
> here:
> <http://ideas.repec.org/p/cgd/wpaper/168.html>
> 
> 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/
> 


      

*
*   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/