kernel smoothing of weighted data
density() in the R-devel version of R allows weights. locfit() in the package of the same name also appears to be documented to.
On Tue, 16 Aug 2005 riap2 at cam.ac.uk wrote:
I want to use kde() or a similar function for kernel smoothing but I want to specify the weight of each of my data points. I do not want to specify the bandwidth on a point by point basis.
The only kde() I found is from the recent package ks, and is for multivariate data -- if you want that, you did not say so and I've not looked for an answer there.
This seems such a simple and obvious thing to want to do I am suspicious that there is not an obvious way to do it. The only discussion I have found is about negative weights(!) and says nothing about implementation. Can anyone suggest a package I have missed or suggest the best starting point for writing my own solution. The reason for wanting this is that I have a number of samples each of ~1000 data points from the same distribution but the samples are of slightly differing statistical weight and eventually each point in each sample may have its own statistical weight. I have searched the list but I am not subscribed to it so please make me an addressee of any reply.
Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595