[R] user defined panel function

[Sundar Dorai-Raj]

Martin Henry H. Stevens wrote:
> Mac OS 10.3.9 R framework  v. 2.1.1
> 
> I am attempting to put a fitted curve into each panel of a lattice 
> graph, but am failing to do so. I have tried writing a very 
> sophisticated function to do so. The function seems to work when used 
> with plot(), but does not do so inside a panel function in xyplot().
> Any pointers would be appreciated.
> 
> #The example data
> fact <- gl(2,7)
> x <- rep(1:7,2)
> y <- c(1,1,2,3,2,3,4,1,2,1,2,3,3,4)
> plot(jitter(y/6) ~ x)
> 
> # The following user defined function puts a curve (I believe the 
> correct one) into the scatterplot
> panel.predglm <- function(x, y) {	
> 	model.trial <- glm(cbind(y,6-y) ~ poly(x,2), 
> family=quasibinomial(link="logit"))
> 	xfit <- seq(1,7, length=21)
> 	yfit <- predict(model.trial, newdata=data.frame(x=xfit), 
> type="response")
> 	lines(xfit,yfit)  }
> 
> panel.predglm(x, y)
> 
> 
> # My attempt to use it in a lattice xyplot, however, fails. It draws a 
> curve which in most cases is outside the dimensions of the plot. I 
> suspect that the prediction is on the scale of the link functions.
> 
> library(lattice)
> xyplot(y/6 ~ x|fact, ylim=c(0,.8),
> panel=function(x, y,...) {
> 	panel.xyplot(jitter(x),jitter(y))
> 	panel.predglm(x,y) }
> 	)
> 
> Any thoughts?

Two:

1. The "y" argument in your panel function ranges from 0 to 1 and not 0 
to 6 as your plot example assumes.

2. You need to use llines in your panel function and not lines.

Here's a working example:

library(lattice)
fact <- gl(2,7)
x <- rep(1:7,2)
y <- c(1,1,2,3,2,3,4,1,2,1,2,3,3,4)

# The following user defined function puts a curve (I believe the 
correct one) into the scatterplot
panel.predglm <- function(x, y) {	
	model.trial <- glm(cbind(y,6-y) ~ poly(x,2),
                          family=quasibinomial(link="logit"))
	xfit <- seq(1, 7, length=21)
	yfit <- predict(model.trial, newdata=data.frame(x=xfit), type="response")
	llines(xfit,yfit)
}

xyplot(y/6 ~ x|fact, ylim=c(0,.8),
        panel = function(x, y, ...) {
          panel.xyplot(jitter(x), jitter(y))
          panel.predglm(x, y * 6)
        })