Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: bandwidth from gwarpreg and its use in lowess


From   "Lazarte Alcala, Naneida" <naneida@okstate.edu>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   st: RE: bandwidth from gwarpreg and its use in lowess
Date   Mon, 15 Feb 2010 11:21:58 -0600

Dear Nick,
Iam sorry, I'll try to be more clear. I need to estimate a semiparametric model, for which I am using plreg command. This command uses lowess for the estimation of the nonparametric portion of the model. I have looked in stata for a method that allows me to calculate the optimum bandwidth I shold use base on my data, and include that optimum in the lowess part of my command, but I have been unsuccesful. That is why I used a method for kernel estimation in the case of a bivariate model y=f(x) (i.e. gwarpreg); however, I do not know how to translate the results in terms of the percentage of the sample that I shoud use for the lowess. Probably the question here is, is there any command with which I can obtain an optimum bandwidth for lowess?

Thanks and any clue on this will be deeply appreciate

Naneida
________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox [n.j.cox@durham.ac.uk]
Sent: Monday, February 15, 2010 10:12 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: RE: bandwidth from l2cvwarp and its use in lowess

-l2cvwarp- gives one of several competing choices for "optimum"
bandwidth in kernel density estimation, a univariate problem. I don't
see what that has to do with lowess- (a bivariate problem in Stata) or
-plreg- (a trivariate or higher problem).

There's a small industry in such criteria that tests and exhibits the
competence of mathematical statisticians. For practical data analysis a
better test is consistency with scientific or practical needs as given
by the context of your problem.

Nick
n.j.cox@durham.ac.uk

Lazarte Alcala, Naneida

I have used the command l2cvwarp (not part of official Stata, it is in
STB 27/snp6_2) which is the WARPing version of the L2CV (Least squares
cross-validation method) to determine the optimal bandwidth that I need
to use for performing nonparametric estimation. From the command I
obtain the bandwidth that minimizes the L2CV function, which is  4,000,
but now  I need to "translate it" in order to use it in the plreg
command (this one is also a user-written command that performs
semiparametric estimation). The plreg command option bandw() accepts
bandwidths between 0 and 1, e.g. a bandwidth 0.8, which is de default
one, says that 80% of the data is being used to the smoothing at each x.
Can anybody tell me how I should understand the bandwidth from l2cvwarp
and put it in terms of the plreg command?(this command uses lowess to
estimate the nonparametric part of the model).


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index