[R] confidence interval as shaded band (lme)

I don't know lme models very well, but if you have standard errors for your
values, this shouldn't be too hard (as a last resort) using polygon()

For example

x = 1:10
y = x^2
y.Err = 2*x
y.Up = y + y.Err; y.Dn =y-y.Err

# This graph is actually quite ugly so don't copy the formatting....
plot(x,y,type="n")
polygon(c(x,rev(x)),c(y.Up,rev(y.Dn)),col="grey",border="red")
# Use the rev commands so the border moves logically around the shaded area
lines(x,y,type="b",lwd=3) # put the means back on top of the polygon

Still, this is a little brute force and I'd imagine that someone else will
shortly let you know how R can already do automatically.

Michael Weylandt

Hi:
Which fitted values? The ones conditional on the random effects or the
ones averaged over the random effects? The standard errors of the two
sets of predictions are not the same.
You can fake it with geom_ribbon(), but it would be convenient to have
the endpoints of the CIs in a data frame you could input into ggplot2.

HTH,
Dennis

Would someone tell me how they propose to go from "standard errors" to
confidence intervals*. I suspect Doug Bates would probably like to
know, also, as he has expended a lot of effort on this over the years,
I believe. :-)

-- Bert

* Note +/- 2 std errors is almost certainly not the right answer --
how far wrong it is ... that's the question.


Cheers,
Bert