(disclaimer: I'm in physics, not stats... )
I have a multivariate problem.
One variable, call it R1, and 3 "predictor" variables, P1, P2, P3.
My goal is to take a load of training data (I know R1,P1,P2,P3 for about
700 total points), and then predict R1 for a new set of data for which I
have all the predictors. Simple, no?
I understand how to calculate bandwidths, and I have a kind of
bastardized way of getting the conditional distribution, i.e.,
f(R1|P1=0.8,P2=0.2,P3=2)
using
fitted(npudens(bw=bw,edat=newdata))
evaluating over a vector of R1.
I have then been using this "density" to get a maximum likelihood
estimator of R1- I have no idea if that is really valid, and if anyone
wants to yell at me go ahead, I want to do this the correct way and I'm
sure I'm making it harder than it is.
Moving past that, the technical problem I am facing is getting a
prediction interval from this.
There's npqreg, and I get how it works when you have one predictor, but
what happens when you have many?
What I want to do is get the 0.05 and 0.95 quantile for a given
P1,P2,P3. to use as my prediction interval.
Thanks,
EM
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
