conditional panel function in lattice for multiple

Question

I have a dataset in which a value (mean) can or cannot fall within an interval given by lower.bound and upper.bound. I would like to plot this using lattice and have achieved something really nice, but there are still three things missing, I don't know how to tackle (I am relatively new to lattice).

df <- read.table("http://pastebin.com/raw.php?i=FQh6F12t")

require(lattice)
lattice.options(default.theme = standard.theme(color = FALSE))

head(df)
##    code topic problem mean lower.bound upper.bound consistent
## 7  A04C  coke      MP 99.5       36.45       95.95          0
## 8  A04C  coke      MT 47.5       22.78      100.00          1
## 11 A04C  girl      MP 50.0        4.75        9.75          0
## 12 A04C  girl      MT 99.5       20.00      100.00          1
## 23 A14G  coke      MP 88.5       21.25       66.75          0
## 24 A14G  coke      MT 82.5       48.36      100.00          1    

dotplot(lower.bound + mean + upper.bound ~ code | problem * topic, 
       data = df, pch = c(6, 3, 2), scales = list(x = list(draw = FALSE)),
       as.table = TRUE)

This produces:

The down-arrows/triangles indicate the lower bound, the up arrows/triangles indicate the upper bound and the + marks the mean. The following things I would like to add to the plot but have no idea how (besides obviously customizing the panel function):

1. Conditional pch based on whether or not a mean value is inside the interval. The variable consistent indicates this (0 = outside the interval). pch should be 1 for values inside and 3 for values outside the interval. (pch for the lower- and upper bound should remain unchanged)
2. Marking the interval. I would like to draw a thicker line between the lower.bound and upper.bound at each x-axis tick.
3. Add the proportion of values outside the interval to the panel headers (e.g., MP; 58.6% to the panel in the upper left corner).

For 1 and 2 my problem obviously is, that I don't know how to deal with custom panel function when having multiple y variables (i.e., how to write conditional panel functions based on this). But I couldn't find anything on it.

For 3, the proportion of values outside the interval is given by something like:

1 - with(df, tapply(consistent, list(topic, problem), mean))
##          MP     MT
## coke 0.5862 0.1724
## girl 0.8276 0.1724

If the answer would furthermore include a nice ordering of levels on the x-axis that would definitely be a plus. The order can change in every panel (i.e., even in panels above each other the same x-axis tick can correspond to a different level of code). But this is not important.

Answer 1

Well, this isn't real pretty, but it should get the real job (showing you how to get this kind of plot working) done.

The basic idea is to rewrite the formula so that it doesn't have a bunch of names on its LHS (i.e. lower.bound + mean + upper.bound). That syntax is equivalent to specifying a groups= term, which ends up triggering panel.superpose() which is kind of a pain to customize in the way you want.

Instead, I just include mean on the LHS, and then use subscripts inside of the custom panel function to pick out in each case the matching elements of upper.bound and lower.bound.

I'm hoping the rest is pretty self explanatory:

LABS <- LETTERS[1:4]
with(df,
     dotplot(mean ~ code | problem * topic,
             lb=lower.bound, ub=upper.bound, mpch = c(3,1)[consistent+1],
             ylim = extendrange(c(0,100)),
             panel = function(x, y, lb, ub, mpch, ..., subscripts) {
                 panel.dotplot(x, y, ..., pch=mpch[subscripts])
                 lpoints(x, lb[subscripts], pch=6)
                 lpoints(x, ub[subscripts], pch=2)
                 lsegments(x,lb[subscripts],x,ub[subscripts],col="grey60")
                 ltext(x=x[3], y=95, LABS[panel.number()], col="red",fontface=2)
             },
             scales = list(x = list(draw = FALSE)), as.table = TRUE)
     )