Re: st: computing elasticities after using lpoly

[Austin Nichols <austinnichols@gmail.com>]

Bert Lloyd <bert.lloyd.89@gmail.com>:
There is no way to get the slope terms or SE or CI from -lpoly- and it
is not easy to calculate them; but bootstrapping should be easy
enough, and work well provided you have a large N relative to a small
number of fixed values at which you want to compute slopes and SE or
CI.

On Sun, Oct 28, 2012 at 9:51 PM, Bert Lloyd <bert.lloyd.89@gmail.com> wrote:
> Dear Austin,
>
> Sorry to be imprecise and thanks for allowing me to revise-and-resubmit.
>
> My question is about inference on the estimated slope, rather than the
> estimated conditional mean.
>
> In terms of your example, are the standard errors given by _se[time]
> likely to be consistent?
>
> If not, is there a way to obtain the analytical standard errors for
> the slope term directly from lpoly? Or would a bootstrapping approach
> be preferable? (If I am reading the help file for R's np package
> correctly, npreg and related functions compute analytical standard
> errors for the gradient but also provide bootstrap estimates -- see
> citation below.)
>
> Thanks,
>
> BL
>
> Hayfield and Racine, "The np package," v 0.40-13, 2012; pages 9-11.
> http://cran.r-project.org/web/packages/np/vignettes/np.pdf
>
> Li and Racine, Nonparametric Econometrics, 2007, provide a formula for
> the asymptotic variance of the local polynomial estimator (Theorem
> 2.10, pg. 90), although it is not obvious to me whether this is for
> data assumed to be i.i.d. (homoskedastic) or i.n.i.d.
> (heteroskedastic).
>
> On Sun, Oct 28, 2012 at 7:11 PM, Austin Nichols <austinnichols@gmail.com> wrote:
>> Bert Lloyd <bert.lloyd.89@gmail.com>:
>> I do not understand your question; are you asking whether -lpoly-
>> computes consistent standard errors? If so, please read the manual
>> entry for  -lpoly- and the references therein.  If you mean CIs in my
>> example, there are none!
>>
>> On Sat, Oct 27, 2012 at 10:43 PM, Bert Lloyd <bert.lloyd.89@gmail.com> wrote:
>>> Dear Austin,
>>>
>>> This looks like a very clever solution. Do you have a sense of whether
>>> the confidence intervals are likely to be consistent? If not, would
>>> you recommend a bootstrapping approach?
>>>
>>> Thanks,
>>> BL
>>>
*
*   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/