*
* Example of non-parametric regression from pp 224-227
*
open data pulppric.dat
data(format=free,org=columns) 1 25 shipments price
*
* The non-parametric regression is done using the instruction NPREG. NPREG
* doesn't do the precise method of locally weighted regression described in the
* text; METHOD=LOWESS takes as its stiffness parameter the fraction of the data
* set to be considered "neighbors" to be included in the local regression. (The
* more distant members of this are then downweighted). Here, the fraction needs
* to be boosted a bit from the default of .5 to get roughly the same fit as shown.
*
* NPREG generates a pair of series - the first ("grid") is the grid of "x" values
* and the second ("fit") are the corresponding estimated curve values.
*
npreg(method=lowess,fract=.75) shipments price / grid fit
*
* To get the smoothed curve on the same graph as the observed points, you need to
* use the options OVERLAY=LINE and OVSAME. OVERLAY allows two different styles to
* share the graph. OVSAME forces both pairs of series to share the same vertical
* axis.
*
scatter(footer="Figure 5-18 Locally Weighted Fit",$
  hlabel="World pulp price",vlabel="Pulp shipments",overlay=line,ovsame) 2
# price shipments
# grid fit