R help - Jul 2003 - Multipanel weighted regression lines.

Dear R-users,

I am trying to create a multipanel scatter plot which displays weighted 
regression lines for each subgroup. I am having diffuculty figuring out 
how the Trellis multipanel plotting functions work, in general. Does 
someone know where there are some beginners examples on how to create 
and use  panel.xxx functions in conjunction with the xyplot functions?

Back to the matter at hand - There are two different ways that I could 
add these lines, I think.

The first, and simplest would be to first find the coeffecients of the 
weighted regression line for each panel ( I can do this). Then I would 
add these coeffecients into a vector/matrix (i'm not sure what form of 
vector would be suitable) and add them to each panel (I do not know how 
to do this).

The second way would be to fiddle around with the function below to add 
the "weights" to the lm function call. I do not know how to change
this
"plot.regression" function to do that.

plot.regression = function(x,y)
{
panel.xyplot(x,y)
panel.abline(lm(y~x))
}

xyplot(yg~xg | g, panel="plot.regression")

Any help would be great!
Thanks,
Kevin Brown.

Kevin Brown asks:
> Does someone know where there are some beginners examples on how to
> create and use panel.xxx functions in conjunction with the xyplot
> functions? 
I don't know of any formal resources other than the help page. I worked
out about panel functions by experimenting with functions of the form
  panel=function(...) print(...)
!
> The second way would be to fiddle around with the function below to add 
> the "weights" to the lm function call. I do not know how to
change this
> "plot.regression" function to do that.
I would use the subscripts=TRUE argument to xyplot. Then your panel
function will receive an argument named subscripts, which indicates which
rows of your data are being plotted in this panel. For example, something
like

xyplot ( yg~xg | g, subscripts=TRUE, panel=function(x,y,subscripts,...) {
  panel.xyplot(x,y)
  thisPanel <- unique(g[subscripts])
  fit <- lm(y~x and something involving thisPanel)
  panel.abline(fit)
  })

Damon Wischik.

Dear Kevin,

If you look at the examples under ?xyplot, you'll get some idea
about using panel functions. In your case, assume your data frame
has variables x, y, w (for weight), G (for panel) and g (for sub-
group). You could plot this using something like the following
clumsy effort:

   x <- rnorm(500)
   y <- 0.9*x + sqrt(1-0.9*0.9)*rnorm(500)
   g <- sample(x=1:2, size=500, replace=TRUE)
   G <- sample(x=c("A", "B", "C"), size=500,
replace=TRUE)
   w <- runif(500)
   tmp <- data.frame(x=x, y=y, w=w, g=g, G=G)
   #
   xyplot(y ~ x | G, data=tmp, groups=g, weights=w, 
          panel=function(x,y,groups,subscripts=subscripts,
                         weights,...) {
      t1 <- unique(groups)
      for (i in 1:length(t1)) {
         w <- is.element(groups[subscripts], t1[i])
         lpoints(x[w], y[w], pch=i, col=i)
         panel.abline(lm(y[w] ~ x[w], weights=(weights,
               [subscripts])[w]), lty=i, col=i)
      }
   }, key=list(space="top", columns=2, points=list(pch=1:2,
               col=1:2), text=list(paste("Group", 1:2),
               col=1:2)))
    


Regards,

Andrew C. Ward

CAPE Centre
Department of Chemical Engineering
The University of Queensland
Brisbane Qld 4072 Australia
andreww at cheque.uq.edu.au


Quoting Kevin Brown <kevin.brown at mail.mcgill.ca>:
> Dear R-users,
> 
> I am trying to create a multipanel scatter plot which displays
> weighted 
> regression lines for each subgroup. I am having diffuculty
> figuring out 
> how the Trellis multipanel plotting functions work, in general.
> Does 
> someone know where there are some beginners examples on how to
> create 
> and use  panel.xxx functions in conjunction with the xyplot
> functions?
> 
> Back to the matter at hand - There are two different ways that
> I could 
> add these lines, I think.
> 
> The first, and simplest would be to first find the coeffecients
> of the 
> weighted regression line for each panel ( I can do this). Then
> I would 
> add these coeffecients into a vector/matrix (i'm not sure what
> form of 
> vector would be suitable) and add them to each panel (I do not
> know how 
> to do this).
> 
> The second way would be to fiddle around with the function
> below to add 
> the "weights" to the lm function call. I do not know how to
> change this 
> "plot.regression" function to do that.
> 
> plot.regression = function(x,y)
> {
> panel.xyplot(x,y)
> panel.abline(lm(y~x))
> }
> 
> xyplot(yg~xg | g, panel="plot.regression")
> 
> Any help would be great!
> Thanks,
> Kevin Brown.
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>