# Re: st: Re: Tobit coefficients

 From Lisa Powell <[email protected]> To [email protected] Subject Re: st: Re: Tobit coefficients Date Thu, 30 Dec 2004 12:55:47 -0500 (EST)

```Greetings,

Following up on this earlier exchange over the marginal effects from the
tobit model, I have a question regarding the STATA output:

For the following command:

tobit facil4p \$demo5, ll(0);
mfx compute,  predict(e(0,.));

does this give marginal effects in terms of

1) E(y|x,y>0)

OR

2) E(y|x) = P(y>0) * E(y|x,y>0)

If it is (1) could someone please tell me how to specify (2).

Many thanks,
Lisa

On Tue, 23 Mar 2004, Scott Merryman wrote:

> Stefan,
>
> The output of -tobit- gives the estimated coefficients that are related to the
> latent variable y*. It shows the effect of a change in a given x variable on the
> expected value of the latent variable, holding all other x variables constant.
> In Stata Manual example, the OLS coefficients of the latent model are compared
> to the estimated latent effects of the tobit model E[y*|x].  But this is just
> for illustration (see technical note, page 253) is you had uncensored sample
> there would be no reason to use a tobit model.
>
> If you are comparing the OLS marginal effects estimated on the observed data,
> than you would want use the expected unconditional of the observed data E[y|x].
> Usually, the marginal effects of the latent variable, y*, are not of interest
> since it is unobserved -- we only observe y* if it above a threshold.  Though
> why would you want to compare them to OLS coefficients, since the OLS
> coefficients will be inconsistent?
>
> Hope this helps,
> Scott
>
> ----- Original Message -----
> From: "LACHENMAIER" <[email protected]>
> To: <[email protected]>
> Sent: Tuesday, March 23, 2004 8:11 AM
> Subject: st: Tobit coefficients
>
>
> > Hi all,
> >
> > I have some problems with the output of the Tobit command. I always
> > thought that the coefficients presented after the tobit command are the
> > "pure" tobit coefficients, i.e. if I want to compare them to OLS
> > coefficients I have to rescale them by an adjustment factor (which
> > depends on whether I want to get the unconditional effect E(y | y,x), or
> > the conditional effect E(y | y>0,x), like described e.g. in Wooldridge
> > (2003, ch. 17.2).
> >
> > But then I had a closer look at the Stata manual (Stata 8) because I
> > predict(). But in the first tobit example in the manual (p.252) the
> > tobit coefficient is directly compared to the OLS coefficient. So I am a
> > bit confused what the output of the tobit command really is now and how
> > to calculate which effects....
> >
> > Maybe someone can help me here.
> >
> > Stefan
> >
> > PS: Sorry for posting this problem again, but I have not solved the
> > problem yet. Thanks!
> >
> > *
> > *   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/
>
> *
> *   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/
>

Lisa M. Powell
Research Associate Professor, University of Illinois at Chicago
Fellow, School of Policy Studies, Queen's University
tel: 1-312-413-8468; fax: 312-355-2801
email: [email protected]  [email protected]

*
*   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/
```