st: RE: trying to understand -lpoly- with degree zero

[philippe van kerm <philippe.vankerm@ceps.lu>]

Dan,

The kernel function you are using is what Stata would typically refer to as the 'alternative Epanechnikov kernel'. See under -kdensity- in the manual for kernel definitions.

Try adding the -kernel(epan2)- option to your -lpoly- call.

Philippe


> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-
> statalist@hsphsun2.harvard.edu] On Behalf Of Dan Weitzenfeld
> Sent: Tuesday, April 13, 2010 1:20 AM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: trying to understand -lpoly- with degree zero
>
> Hi All,
> I'm trying to reverse-engineer the lpoly command, so that I can code it
> in a different language.
> Specifically, I'm trying to mimic the command with the default options
> of degree(0) and kernel(epanechnikov).
> Within -help lpoly-, it says:
>
>
>     degree(#) specifies the degree of the polynomial to be used in the
> smoothing.  The default is degree(0), meaning local-mean
>         smoothing.
>
> I understand this to mean that at each point in the smoothing grid, a
> kernel-weighted mean is taken.  For my purposes, it makes sense to use
> every point in the dataset as a point in the grid.  But when I tried to
> code that - see below - my resulting curve is less smooth than the
> output of -lpoly- with degree 0, same kernel, same half-width.
>
> According to wikipedia, the epanechnikov kernel is:
> K(u) = (3/4)(1-u^2)
> Am I scaling the kernel incorrectly?
>
> Thanks in advance,
> Dan
>
> *----------------------------------------------------
>
> webuse motorcycle, clear
> *my dataset is a constant 60 hz; changing this dataset to mimic
> replace time = _n*(1/60)
> local  halfwidth = .04
> local  halfwidth_squared = `halfwidth'*`halfwidth'
> qui gen fake_lpoly = .
> qui count
> local vals = r(N)
> forvalues i = 1(1)`vals' {
>  local time = time[`i']
>  capture drop distance* kernel product
>  qui gen distance_from_point = time-`time'
>  qui gen distance_sq =  distance_from_point*distance_from_point
> *generating epanechnikov kernel, scaled to local halfwidth
>  qui gen kernel = (3/4)*(1- (distance_sq/`halfwidth_squared'))
>  qui gen product = kernel*accel if kernel>=0
>  qui sum product
>  local numerator = r(sum)
>  qui sum kernel if kernel>=0
>  local denominator = r(sum)
>  qui replace fake_lpoly = `numerator'/`denominator' in `i'
> }
>
> *compare with -LPOLY-
> sort time
> capture drop x real_lpoly
> lpoly accel time, bw(`halfwidth') n(`vals') gen(x real_lpoly) nogra
> twoway (line  real_lpoly  fake_lpoly time) (scatter  accel time)
>
> *--------------------------------------------------
>
> This email message is confidential and is intended only for the named
> recipient(s) above, and may contain information that is privileged,
> attorney work product or exempt from disclosure under applicable law.
> The contents of this email are for discussion purposes only and shall
> not create a legally binding commitment or agreement of EmSense
> Corporation, which shall require EmSense Corporation management
> approval and a formal written agreement signed by both parties.  If you
> have received this message in error, or are not the named recipient(s),
> please immediately notify the sender and delete this message from your
> computer.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

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