Re: st: mfx failing with tobit?

[Maarten Buis <maartenlbuis@gmail.com>]

To add to Nick's repetition of my answer:

"The" marginal effect for a Tobit model does not exist: there are
several marginal effects depending on what your outcome of interest
is. The outcome of interest could be the expectation of the latent
variable, the expectation of the censored variable where the censored
values are replaced by 0 (i.e. the values one would actually observe
in the field), the expectation of the censored variable conditional on
not being censored, or the probability of being censored. To learn
more about these you could look at (Breen 1996) The default is the
expectation of the latent variable, and this is a linear function of
the explanatory variables, so the these marginal effects will be the
same as the coefficients. If you want one of the other marginal
effects, you will have to specify the -predict()- option in -margins-.
This will also require that you study the -help tobit_postestimation-
helpfile, and probably the manual entry for -tobit- and -tobit
postestimation-.

Hope this helps,
Maarten

Richard Breen (1996) "Regression Models: Censored, Sample Selected, or
Truncated Data." Thousand Oaks: Sage.


On Tue, Mar 25, 2014 at 5:42 PM, Nick Cox <njcoxstata@gmail.com> wrote:
> Maarten has already addressed this directly. As tobit is a linear
> predictor, the derivatives are equal to the coefficients of the
> predictors. It's like asking: if I differentiate the equation of a
> hyperplane, will I get more information out? No; the information is
> already given in the coefficients.
>
> Nick
> njcoxstata@gmail.com
>
> On 25 March 2014 16:34, Roger Cremades Rodeja <ecosafor@gmail.com> wrote:
>
>> using stata 12, and trying to obtain the marginal estimates after a
>> Tobit regression, I write:
>>
>> tobit indepvar depvar1 depvar2 .. depvar9, ll
>> margins, dydx (*)
>>
>> the second command is supposed to provide the "estimate marginal
>> effect of variables in varlist", after checking the help of margins,
>> but I only get again the same coefficients as in the Tobit
>> regressions, but labelled as dy/dx
>>
>> I do not know what should I do to obtain the marginal estimates of a
>> Tobit, I understood that was the right way,
>
> 2014-03-17 17:30 GMT+01:00 Roger Cremades Rodeja <ecosafor@gmail.com>:
>
>>> it was extremely helpful, I will see what can I do with stata 10, or
>>> use the 12th of a colleague,
>
> 2014-03-17 14:26 GMT+01:00 Maarten Buis <maartenlbuis@gmail.com>:
>
>>>> -mfx- is a general purpose program, so it needs to choose a default
>>>> outcome for which to compute the derivative. It chooses that default
>>>> to be the default chosen for the predict option of the command, which
>>>> in the case of -tobit- happens to be the linear predictor. So the
>>>> default marginal effect computed for -tobit- is just the coefficients.
>>>> If you want to change that, you need to specify the -predict()- option
>>>> to whatever outcome you are interested in.
>>>>
>>>> Notice however, that since you did not specify the version of Stata
>>>> you are using we must, per the Statalist FAQ, assume you use the
>>>> latest version of Stata (13). In that case the -mfx- command is
>>>> hopelessly outdated, and you should use the -margins- command instead.
>
> On Mon, Mar 17, 2014 at 11:13 AM, Roger Cremades Rodeja
>
>>>>> after running a Tobit regression, I introduce the mfx command and
>>>>> surprisingly I get exactly the same coefficient estimates I had from the
>>>>> regression, although labelled as dy/dx. Does anybody had a similar problem?
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/



-- 
---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/