2

I'm wondering where does the "nonparametric" label of locally weighted regression like LOESS or LOWESS comes from, i.e. why they are nonparametric methods?

Also, I would like to know in general how this locally weighted regression algorithm is implemented in statistical software like R. Specifically, I'm wondering how to choose which x values to be used as the "center" of a window to fit the local weighted regression? Or do we do this for each of the observations?

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
askming
  • 577
  • 1
  • 4
  • 15
  • 3
    They're nonparametric because they don't depend on a finite number of explicit parameters. See part 2. of the definition of *nonparametric* [here](http://en.wikipedia.org/wiki/Nonparametric_statistics#Definitions) You may be thinking of 'nonparametric' in the sense of the probability distribution (sometimes called 'distribution-free', though the two are not quite synonyms), as discussed in part 1. of the same link (if somewhat mangled). The two senses of the word are closely related. See the discussion in the last two thirds of [this answer](http://stats.stackexchange.com/a/58768/805) – Glen_b Feb 21 '15 at 01:19
  • 1
    If you want to know how a procedure is implemented in R, you can just type, say, loess at the prompt (which is not to say that it will always be straightforward to get to the "gist" of the procedure). – Christoph Hanck Feb 24 '15 at 07:20

0 Answers0