Statalist


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

Re: st: tobit probit censored data confusion


From   nicola.baldini2@unibo.it
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: tobit probit censored data confusion
Date   Wed, 03 Oct 2007 18:07:03 +0200

You may have a look at Stata's FAQ at http://www.stata.com/support/statalist/faq. That on selection models, if I remember correctly.
Basically:
- Sample selection bias refers to problems where the y is observed only for a restricted, nonrandom sample. If you believe that the coefficients of the covariates differ according to the selection variable, then you need -heckman-. The selection equation requires to specify independent variables that affect the probability to observe y, but not its level estimated in the regression equation (conditional on a non-zero observation). If you assume that the independent variables affect both the incidence and the level of y, the Heckman model reduces to a tobit specification (-tobit-).
- Under treatment-effects models (-treatreg-), the coefficients of the covariates are restricted to be the same for treated and non-treated. 
Note that only tobit has a built-in command for panel data in Stata 9 (-xttobit-). However, you may use 
-xtivreg- (or -xtivreg2- from ssc) in lieu of -heckman-. Also, you may have a look at http://ideas.repec.org/p/boc/usug07/12.html

Nicola

At 02.33 03/10/2007 -0400, you wrote:
>Dear all,
>
>I have a question regarding statistical models. I have a panel data
>with obs on people's background information (age, wealth from other
>sources, etc), and wage if the person decides to work. A person may
>choose to work, if so, there is an observable wage; if not, wage is 0.
>
>I am confused about what models to use. It looks like a standard tobit
>model, with wage censored at 0. Below is an alternative way to model.
>
>A: Probit model of the decision to work or not.
>B: For wage>0 observations, use OLS regression.
>
>I have seen both in practice. But while using two models, the authors
>mention that they are estimating both model A and model B at the same
>time. What does it mean "at the same time"?
>
>Bottom line:
>If both tobit and the joint estimation of models A and B are
>appropriate, which is better?
>
>Thanks for clarifying this for me! 

*
*   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