.-
help for ^gwarpreg^                                               [STB-30: snp9]
.-

Adjusted prediction error for WARP regression (Nadaraya-Watson estimator)
-------------------------------------------------------------------------

        ^gwarpreg^ yvar xvar [^if^ exp] [^in^ range],
                ^d^elta^(^#^)^ ^k^ercode^(^#^)^ ^s^elect^(^#^)^ [ ^bou^nd^(^#^)^
                ^g^en^(^score mvalue bwidth^)^ ^me^nd^(^#^)^ ^ms^tart^(^#^)^ ]


Description
-----------

^gwarpreg^ program calculates the adjusted prediction error of the WARP
approximation of the Nadaraya-Watson estimator for several penalty functions
in order to find an optimal bandwidth for kernel regression.  By default,
^gwarpreg^ draws two graphs, the score versus M and the score versus
bandwidth, and displays a table with the five smallest scores and their 
corresponding M's and bandwidths.


Options
-------

^d^elta^(^#^)^ specifies the desired accuracy for estimation (equivalent to the
    small bin width in WARP density estimation).

^k^ercode^(^#^)^ specifies the weight function (kernel) used to calculate the
    required univariate densities.  The following codes are used:

	1 = Uniform
 	2 = Triangle 
 	3 = Epanechnikov
 	4 = Quartic (Biweight)
 	5 = Triweight
 	6 = Gaussian

^s^elect^(^#^)^ specifies the penalty function as follows:

	1 = Shibata's model selector
	2 = Generalized cross-validation
	3 = Akaike's information criterion
	4 = Finite prediction error
	5 = Rice's T

^bou^nd^(^#^)^ specifies the fraction of the data to be excluded from the
   estimation. By default it is set to 0.1.

^g^en^(^score mvalue bandwidth^)^ generates new variables to store the adjusted
    prediction error (score) and corresponding M and bandwidth values.

^ms^tart^(^#^)^ specifies the initial value for the range of M
    used for the minimum score search.  The default is 5.

^me^nd^(^#^)^ specifies the final value for the range of M used for
    the minimum score search.  The default is int(0.3*(max(x)-min(x)/delta).


Remarks
-------

^d^elta, ^s^elect, and ^k^ercode, are not optional. If the user does not provide
them, the program halts and displays an error message on screen.

This program uses modified versions of the ado and Turbo Pascal executable
files for kernel density estimation presented in Salgado-Ugarte, et al.
(1993).  The program is based on the algorithms and programs of Haerdle (1991)
and Scott (1992).


Examples
--------

  . ^gwarpreg wait dura, delta(0.05) selec(2) kercode(4) mend(40)^

  . ^gwarpreg accel time, d(0.2) s(1) k(4) me(20) gen(score mvalue bandw)^


References
----------

Haerdle, W. 1991. Smoothing techniques with implementation in S.
    Springer-Verlag.

Salgado-Ugarte, I. H., M. Shimizu, and T. Taniuchi 1993. snp6: Exploring
    the shape of univariate data using kernel density estimators. Stata 
    Technical Bulletin 16: 8-19.

Salgado-Ugarte, I. H., M. Shimizu, and T. Taniuchi 1995. snp6.1: ASH,
    WARPing, and kernel density estimation for univariate data. Stata
    Technical Bulletin 26: 23-31.

Salgado-Ugarte, I. H., M. Shimizu, and T. Taniuchi 1995. snp6.2: Practical
    rules for bandwidth selection in univariate density estimation. Stata
    Technical Bulletin 27: 5-19.

Scott, D. W. 1992. Multivariate density estimation: Theory, practice,
    and visualization. John Wiley & Sons.

Silverman, B. W. 1986. Density estimation for statistics and data analysis. 
    Chapman and Hall.


Authors
-------

        Isaias H. Salgado-Ugarte, Makoto Shimizu and Toru Taniuchi
        University of Tokyo, Faculty of Agriculture,
        Department of Fisheries, Yayoi 1-1-1, Bunkyo-ku
        Tokyo 113, Japan.
        fes01@@tzetzal.dcaa.unam.mx


Also see
--------

    STB: STB-30 snp9, STB-27 snp6.2, STB-26 snp6.1, STB-16 snp6
On-line: ^help^ for @kerneld@, @warpden@, @warpdens@, @warpreg@, @kernreg@