Joachim Wagner <[email protected]> asks:
> I am reading a paper by Hahn, Todd and Van der Klaauw "Evaluating the effect
> of an antidiscrimination law using a regression-discontinuity desing" (NBER
> Working Paper 7131, May 1999) where the authors estimate the conditional
> means nonparametrically by 'local linear regression' which requires to
> choose a kernel and bandwith. The authors do not inform the reader about the
> software used. Question: Is there a Stata ado-file to perform local linear
> regression? In an up-to-data Stata 8 I typed 'findit local linear
> regression' and got no matches (except for some hints to graph). I am aware
> of the 'lowess' command for locally weighted regression, and I wonder how
> this is related to local linear regression. Any ideas? Many thanks in
> advance.
Have a look at -lp_regress-, available from my user site. From a net-aware
Stata, type
. net from http://www.stata.com/users/rgutierrez
and follow the links.
-lp_regress- is an implementation of local polynomial regression (of which
local linear regression is a special case), and is written entirely in ado
code. A faster implementation, -locpoly-, which utilizes Stata plugin code
will soon become publicly available, once plugin capability itself becomes
available via an executable update.
-lowess- differs from kernel methods (such as that employed by -lp_regress-)
in that -lowess- uses a "nearest neighbor(s)" approach while kernel methods
use those observations within a well-defined window width, which may either be
fixed or varying over the x's. -lp_regress- treats the window width as fixed.
--Bobby
[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/