A third possibility is other smoothing problems
with a strictly positive response is to smooth
w.r.t the logarithm of that variable and back-transform.
I have no idea of how far this is a good idea here
with hazard functions, but for density estimation it
often works nicely, and seems a lot less ad hoc than
the surgery of boundary kernels.
Nick
[email protected]
Yulia Marchenko, StataCorp
> William Dupont <[email protected]> asks about behaviour of
> -stcurve- at the boundaries:
>
> >I believe that the smoothing behavior for the hazard
> function plots of
> >-stcurve- is less than ideal near the time boundaries.
> >...
> >I have not attempted to read the -stcurve- code and realize
> that devising
> >smoothing algorithms can be non-trivial. I wondered,
> however, if the program
> >was really working as the authors intended or it there might
> be some way of
> >improving its performance near the time boundaries.
>
> The algorithm in -stcurve- uses the usual smoothing kernel
> technique to
> estimate hazard function as described in [ST] sts graph on
> p.292. Due to the
> symmetry of the kernel, kernel estimators encounter bias at
> the boundary
> points.
>
> Two solutions to this problem would be
>
> 1. Retrict the plot region to not include points near the boundary.
> This is something Bill can do himself, and something we
> are considering
> doing officially.
>
> 2. Use boundary kernels to alleviate the bias. This is
> also something
> we are considering doing officially.
*
* 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/
