[R] Adding locfit confidence intervals in trelis xyplot

Well, panel.locfit doesn't have any options to draw confidence bands,
so you'll have to write a panel function that does. Shouldn't be hard
to extend panel.locfit if you know how to extract that information
from a locfit object.

-Deepayan

Deepayan Sarkar <deepayan.sarkar <at> gmail.com> writes:
...
plot.locfit has bands. I think Juan mixed trellis and standard plot
documentation.


x <- rnorm(100)
y <- dnorm(x) + rnorm(100) / 5
plot(locfit(y~x), band="global")

But it is *within* a trellis plot when I cannot plot the bands!

Building up on Dieter's example:

x <- rnorm(100)
y <- dnorm(x) + rnorm(100) / 5
plot(locfit(y~x), band="global")

Plots two nice confidence bands.

But try this
z <- dnorm(y) + rnorm(100) / 3
Z <- equal.count(z, number = 4, overlap = .1)
xyplot(z ~ y|Z, alpha = 1,band = "global",panel = panel.locfit)

and you will not see the bands,

indeed band is completely ignored as
xyplot(z ~ y|Z, alpha = 1,band = "rubbish",panel = panel.locfit)

works fine but without the bands!!

- juan

And why does that come as a surprise to you?

Since you seem to have missed the point of the previous replies, let
me spell it out for you. A function does what it is instructed to do,
not what you _want_ it to do. The most reliable way to figure out what
those instructions are is to look at the source code of that function.
Since this is not very user friendly, R also requires that all
functions be documented, and specifically, that the documentation for
any function, say foo, can be accessed by typing help(foo). Since you
are using the function panel.locfit, you are supposed to look at
help(panel.locfit), which does not claim to have any support for
bands.

In this case, the source of panel.locfit is actually simpler:

panel.locfit
function (x, y, subscripts, z, xyz.labs, xyz.axes, xyz.mid, xyz.minmax,
xyz.range, col.regions, at, drape, contour, region, groups,
...)
{
panel.xyplot(x, y, ...)
args <- list(x = x, y = y, ...)
ok <- names(formals(locfit.raw))
llines.locfit(do.call("locfit.raw", args[ok[ok %in% names(args)]]))
}

(removing the irrelevant part)

-Deepayan