reorder barchart by sum of bar segments with ggplot/plyr - r

Question

I need the 11 bars in the following stacked barplot to be reordered by the sum of the first two segments of each bar, i.e. sorted by the (red+green) segments in the plot.
> dput(q1m.bl)
structure(list(ItemA = structure(c(1L, 2L, 3L, 4L, 1L, 2L, 3L,
4L, 1L, 2L, 3L, 4L, 2L, 3L, 4L, 1L, 2L, 3L, 4L, 1L, 2L, 3L, 4L,
1L, 2L, 3L, 4L, 2L, 3L, 4L, 1L, 2L, 3L, 4L, 1L, 2L, 3L, 4L, 1L,
2L, 3L, 4L), .Label = c("sehr wichtig", "wichtig", "unwichtig",
"keine Angabe"), class = "factor"), ItemQ = structure(c(1L, 1L,
1L, 1L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 5L, 5L, 5L,
5L, 6L, 6L, 6L, 6L, 7L, 7L, 7L, 7L, 8L, 8L, 8L, 9L, 9L, 9L, 9L,
10L, 10L, 10L, 10L, 11L, 11L, 11L, 11L), .Label = c("PUSHERS_AA",
"PUSHERS_COM", "PUSHERS_BED", "PUSHERS_SEC", "PUSHERS_STAB",
"PUSHERS_COST", "PUSHERS_INNO", "PUSHERS_VAL", "PUSHERS_INDEP",
"PUSHERS_STDS", "PUSHERS_SRC"), class = "factor"), Counts = c(1L,
3L, 4L, 1L, 3L, 3L, 2L, 1L, 4L, 2L, 2L, 1L, 3L, 5L, 1L, 1L, 1L,
6L, 1L, 5L, 1L, 2L, 1L, 1L, 1L, 6L, 1L, 2L, 6L, 1L, 2L, 4L, 2L,
1L, 3L, 3L, 2L, 1L, 2L, 1L, 5L, 1L), blpos = c(0.111111111111111,
0.444444444444444, 0.888888888888889, 1, 0.333333333333333, 0.666666666666667,
0.888888888888889, 1, 0.444444444444444, 0.666666666666667, 0.888888888888889,
1, 0.333333333333333, 0.888888888888889, 1, 0.111111111111111,
0.222222222222222, 0.888888888888889, 1, 0.555555555555556, 0.666666666666667,
0.888888888888889, 1, 0.111111111111111, 0.222222222222222, 0.888888888888889,
1, 0.222222222222222, 0.888888888888889, 1, 0.222222222222222,
0.666666666666667, 0.888888888888889, 1, 0.333333333333333, 0.666666666666667,
0.888888888888889, 1, 0.222222222222222, 0.333333333333333, 0.888888888888889,
1)), .Names = c("ItemA", "ItemQ", "Counts", "blpos"), row.names = c(NA,
-42L), class = "data.frame")
The plot ...
ggplot(q1m.bl, aes(x = ItemQ, y = Counts, fill = ItemA)) +
geom_bar(stat="identity", position="fill") +
geom_text(aes(y = blpos, label = Counts), hjust = 1) +
theme(axis.text.x=element_text(angle=90, hjust = 0), text = element_text(size=10)) +
coord_flip()
Ugh, not enough rep points to embed images. Sorry for the inconvenience. Plot is here: http://i.stack.imgur.com/am0Ud.png
I played around with arrange() and after checking the data frame itself, I thought the following sorting should do the trick. (Note: blpos means "bar label position" and are the positions of the various numbers in the plot.) But plotting this "sorted" data frame leads to the identical plot as above. I do not understand which information to change to change the plotting order of the ItemQ column.
q1m.bl.s <- arrange(q1m.bl, ItemA, desc(blpos))
ggplot(q1m.bl.s, ....
What's the best approach anyway? Should I manipulate the df (using ddply/arrange/reorder/etc.) prior to plotting? Because I tend to think this is a presentation issue and should be done inside ggplot. Does it even matter? The "ggplot ordered barchart" questions I found on SO seem to use both approaches; yet none I found was referring to stacked bar segments and using factor data... hence this new question.
Thank you very much for enlightening me!

Answer 1

It's all about re-ordering the factor levels of the ItemQ variable.
d <- subset(q1m.bl, ItemA %in% c("sehr wichtig", "wichtig"))
totals <- aggregate(d$Counts, list(ItemQ = d$ItemQ), sum)
ItemQ.order <- as.character(totals[order(-totals$x), ]$ItemQ)
q1m.bl$ItemQ <- factor(q1m.bl$ItemQ, levels = ItemQ.order)
Then you should be able to run the code exactly as you provided it and it will produce this:
EDIT (digisus): konvas, I am just re-adding your first answer showing the use of ddply because even I do not feel comfortable with it/do not fully get it, I am sure others can benefit from it. :-) So, with your permission I repost it here:
library(plyr)
ItemQ.order <- q1m.bl %>%
group_by(ItemQ) %>%
filter(ItemA %in% c("sehr wichtig", "wichtig")) %>%
summarise(total = sum(Counts)) %>%
arrange(-total) %>%
select(ItemQ) %>%
unlist %>%
as.character
q1m.bl$ItemQ <- factor(q1m.bl$ItemQ, levels = ItemQ.order)

Answer 2

library(ggplot2)
fac_ord <- function(seed){
set.seed(seed)
return(sample(letters[1:4]))
}
# this seed simulates arbitrary sortings
seed <- 2
fac_ord(seed)
val = c(1,2,3,4,2,2,2,2)
fac = factor(c("a","b","c","d","a","b","c","d"),
levels=fac_ord(seed),
labels=fac_ord(seed),
ordered=FALSE)
dif = c(rep("x",4),rep("y",4))
df = data.frame(val = val, fac = fac)
ggplot(df, aes(x=fac, y=val, fill=dif)) +
geom_bar(stat="identity") +
labs(title = sprintf("seed = %d / %s", seed, paste(fac_ord(seed),collapse=",")))
As the example shows - ggplot will use same ordering for fac in the plot as the internal order of fac. So to influence the plotted order you have to write a function which returns the intended order - in dependence on whatever facts and values - and use this to create the factor fac - and then use this propperly-ordered factor for the plotting.
The intended result can also be reached by application of reorder() for reordering the levels of the factor.