R help - Aug 2004 - Dotplot with nested factors

I am using the dotplot function from the lattice package to
display a quantitative variable versus two factors, say 'a' and
'b'.  The levels of 'a' are nested within levels of 'b'.
The
issue is that dotplot includes all the levels of 'a' in each panel
(conditioning on 'b'), even though many are empty in any given 
panel.  A toy example is

dat <- data.frame(a=letters[1:5],
b=c("A","A","A","B","B"),
y=1:5)
dotplot(y ~ a | b, data=dat)

(In the real data, there are far more levels of 'a' and 'b', so
each
panel ends up with all its data squashed into a small portion of 
the available space.)

I would like to show only the levels of 'a' actually present 
in a given panel.  I hoped that setting 'relation="free"'
would
cause each panel to adjust its axis to match the set of 'a' values 
observed in that panel, but unfortunately it seems 'relation' is 
ignored for factors.  Would it make sense to have 'relation' work
for factors?  Is there some other way to accomplish the same effect?

Rich Raubertas
Merck & Co.

On Monday 16 August 2004 17:06, Raubertas, Richard
wrote:
> I am using the dotplot function from the lattice package to
> display a quantitative variable versus two factors, say 'a' and
> 'b'.  The levels of 'a' are nested within levels of
'b'.  The
> issue is that dotplot includes all the levels of 'a' in each panel
> (conditioning on 'b'), even though many are empty in any given
> panel.  A toy example is
>
> dat <- data.frame(a=letters[1:5],
b=c("A","A","A","B","B"),
y=1:5)
> dotplot(y ~ a | b, data=dat)
>
> (In the real data, there are far more levels of 'a' and
'b', so each
> panel ends up with all its data squashed into a small portion of
> the available space.)
>
> I would like to show only the levels of 'a' actually present
> in a given panel.  I hoped that setting 'relation="free"'
would
> cause each panel to adjust its axis to match the set of 'a' values
> observed in that panel, but unfortunately it seems 'relation' is
> ignored for factors.  Would it make sense to have 'relation' work
> for factors?  Is there some other way to accomplish the same effect?
Yes and no, respectively.

This was recently brought up by John Maindonald off-list, and I had a 
tentative fix planned. Unfortunately, I just realized that it would 
only work when the 'present' levels within each panel are contiguous 
levels of the factor. I'll have to think a bit more about it. Whatever 
the fix, it would have to break the current API, so it's not going to 
be available before R 2.0.0. 

Here's a bad workaround (bad because it would give wrong answers if 
relation != "free"):

dotplot(y ~ a | b, data=dat,
        scales = list(x = list(relation = "free")),
        prepanel = function(x, y, ...) {
            ans <- list()
            if (is.factor(x)) {
                x <- x[drop = TRUE]
                ans$xlim <- levels(x)
            }
            if (is.factor(x)) {
                y <- y[drop = TRUE]
                ans$ylim <- levels(y)
            }
            ans
        },
        panel = function (x, y, ...) {
            x <- x[drop = TRUE]
            y <- y[drop = TRUE]
            panel.dotplot(x, y, ...)
        })


Deepayan