Re: st: difference between -heckman- and -treatreg-

[Shehzad Ali <sia500@york.ac.uk>]

Thanks, Martin and Mark. I appreciate that the two phenomena of sample 
selection and endogeneity bias are different in terms of whether or not the 
outcome variable is observed. But are formulae for hazard/inverse Mills 
ratios the same for the two cases since they both use a probit model to 
generate the ratio? Obviously, as you mentioned, the outcome variable and 
covariates will be different in the probit models for the -heckman- and 
-treatreg-.

Secondly, once we have the ratio, do both heckman and treatreg use the 
ratio as an additional regressor in the same way? If this is true, then the 
difference between the two is only in the probits, and once we have the 
probit then IMR or Hazard would be calculated and used in the outcome 
equation in the same way. Is that right?

Finally, why do we call the ratio 'hazard' in -treatreg- and 'imr' in 
-heckman-?

Thank you,

Shehzad


On Aug 31 2008, Martin Weiss wrote:


> Well, start from the examples in -h heckman- and -h treatreg-, and do  
> not be fooled by the similarity with respect to computation: There is  
> a reason why Stata supplies two estimators.
>
> In - h heckman-, the wage that is supposed to be modelled is missing  
> in 657 cases (-ta wage, m- to see that). Heckman allows one to take  
> into account the mechanism that determines the censoring of 657 cases,  
> i.e. the labor supply behavior of the women in the dataset. So the  
> selection equation models this question with -possibly- different  
> covariates from the outcome equation - the determination of the wage  
> itself.
>
> -h treatreg-, on the other hand, shows the effect of the -enodgenous-  
> choice of attending college on earnings. There are no missing cases  
> here (-ta ww, m- to see that) but the choice of a higher degree  
> impacts earnings. As more able students tend to choose this career  
> track, the decision is endogenous and must be explicitly modelled.
>
>
> HTH
> Martin
>
> Quoting Shehzad Ali <sia500@york.ac.uk>:
>
>
> > Hi,
> >
> > I was wondering if someone can explain the difference between how
> > -heckman- and -treatreg are estimated. I understand that analysts
> > usually prefer -heckman- for sample selection bias and -treatreg- for
> > endogeneity bias. But I was not sure how the two models are different
> > computationally because they both use hazard ratio (or inverse Mills).
> > Is hazard ratio different from IMR? Can anyone direct me to an article
> > that explains the computational and theoretical difference between the
> > two models?
> >
> > Thank you,
> >
> > Shehzad
> > *
> > *   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/
*
*   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/