Search code examples
rlattice

Fitting smooth through xyplot

This question seems simple but I haven't been able to figure out how to do it. I'm trying to fit a smooth line through longitudinal dataset as illustrated in the following code

library(nlme)
xyplot(conc ~ Time, data = Theoph, groups = Subject, type = c("l", "smooth"))

The output isn't quite what I'm after and there are multiple warnings. I would like to fit a smooth through the entire data. As a bonus, if anyone could also show how to do this using ggplot, that would be great.

Solution

  • To plot the individual Subjects as separate lines and points but plot the overall smooth use either of the two lattices approaches shown or the classic graphics and zoo approach at the end. Also note that we need to order the time points to produce the overall smooth and the nlme package is not used. Also note that no errors are given by the code in the question -- only warnings.

    1) trellis.focus/trellis.unfocus We can use trellis.focus/trellis.unfocus to add an overall smooth:

    library(lattice)

xyplot(conc ~ Time, groups = Subject, data = Theoph, type = "o")

trellis.focus("panel", 1, 1)
o <- order(Theoph$Time)
panel.xyplot(Theoph[o, "Time"], Theoph[o, "conc"], type = "smooth", col = "red", lwd = 3)
trellis.unfocus()

    2) panel function A second way is to define an appropriate panel function:

    library(lattice)

o <- order(Theoph$Time)

xyplot(conc ~ Time, groups = Subject, data = Theoph[o, ], panel =   
  function(x, y, ..., subscripts, groups) {
     for (lev in levels(groups)) {
         ok <- groups == lev
         panel.xyplot(x[ok], y[ok], type = "o", col = lev)
     }
     panel.xyplot(x, y, type = "smooth", col = "red", lwd = 3)
})

    Either of these gives the following output. Note that the overall smooth is the thick red line.

    (continued after chart)

    screenshot

    3) zoo/classic graphics Here is a solution using the zoo package and classic graphics.

    library(zoo)

Theoph.z <- read.zoo(Theoph[c("Subject", "Time", "conc")], 
   index = "Time", split = "Subject")

plot(na.approx(Theoph.z), screen = 1, col = 1:nlevels(Theoph$Subject))

o <- order(Theoph$Time)
lo <- loess(conc ~ Time, Theoph[o, ])
lines(fitted(lo) ~ Time, Theoph[o,], lwd = 3, col = "red")

    screenshot