Help System (web edition)

LOWESS <data>,<smooth_level>,<output_matrix>            / J.Puranen 1991,1998
smooths a variable activated by `Y' according to the values of another
variable activated by `X' using a robust locally weighted regression
method (LOWESS = LOcally WEighted Scatterplot Smooth).
    The smoothened values are saved into a variable activated by `S'. If
no `S' variable exists, the smoothened values (as well as the values of `X'
and `Y' variables) are saved into a matrix file given by <output_matrix>.
The default name of the matrix is SMOOTH.M . The results are saved in
increasing order of the `X' values. The smoothened scatterplot can be
plotted by using this matrix file as an input (see the example below).
    The <smooth_level> gives the level of smoothing, specified as how
many percents of the observations around each point affects the smoothing.
A suitable value is 60, which is also the default.
    <data> must be sorted by `X' variable before smoothing, if `S' mask
is applied.

    The number of iterations can be given by ITERATIONS=1 or 2. The default
value of 2 should be adequate for almost all situations.
    In addition, the deviances from the smoothened values can be saved
into a variable activated by `E'. If there are any missing values in <data>,
the `S' and `E' variables must also be initialized with missing values.
    The IND and CASES specifications can be used to select observations.

Reference: Cleveland, W. (1979). Robust Locally Weighted Regression
           and Smoothing Scatterplots. JASA 74, 829-836.

See also SMOOTH?
                                                 (Example on the next page)

........................................................................
Example:
LOWESS <data>,60,SMOOTH.M / VARS=<xvar>(X),<yvar>(Y)
....................................................
GPLOT SMOOTH.M,X,Y,S / YLINE=0 SLINE=1
........................................................................