How to better create stacked bar graphs with multiple variables from ggplot2?

Question

I often have to make stacked barplots to compare variables, and because I do all my stats in R, I prefer to do all my graphics in R with ggplot2. I would like to learn how t

Answer 1

Your second problem can be solved with melt and cast from the reshape package

After you've factored the elements in your data.frame called you can use something like:

install.packages("reshape")
library(reshape)

x <- melt(your.df, c()) ## Assume you have some kind of data.frame of all factors
x <- na.omit(x) ## Be careful, sometimes removing NA can mess with your frequency calculations

x <- cast(x, variable + value ~., length)
colnames(x) <- c("variable","value","freq")
## Presto!
ggplot(x, aes(variable, freq, fill = value)) + geom_bar(position = "fill") + coord_flip() + scale_y_continuous("", formatter="percent")

As an aside, I like to use grep to pull in columns from a messy import. For example:

x <- your.df[,grep("int.",df)] ## pulls all columns starting with "int_"

And factoring is easier when you don't have to type c(' ', ...) a million times.

for(x in 1:ncol(x)) { 
df[,x] <- factor(df[,x], labels = strsplit('
Very Interested
Somewhat Interested
Not Very Interested
Not At All interested
NA
NA
NA
NA
NA
NA
', '\n')[[1]][-1]
}

Answer 2

You don't need prop.tables or count etc to do the 100% stacked bars. You just need +geom_bar(position="stack")

Answer 3

About percentages insted of ..count.. , try:

ggplot(mtcars, aes(factor(cyl), prop.table(..count..) * 100)) + geom_bar()

but since it's not a good idea to shove a function into the aes(), you can write custom function to create percentages out of ..count.. , round it to n decimals etc.

You labeled this post with plyr, but I don't see any plyr in action here, and I bet that one ddply() can do the job. Online plyr documentation should suffice.

Answer 4

If I am understanding you correctly, to fix the axis labeling problem make the following change:

# p<-ggplot(Interest, aes(Interest2, ..count..))
p<-ggplot(Interest, aes(Interest2, ..density..))

As for the second one, I think you would be better off working with the reshape package. You can use it to aggregate data into groups very easily.

In reference to aL3xa's comment below...

library(ggplot2)
r<-rnorm(1000)
d<-as.data.frame(cbind(r,1:1000))
ggplot(d,aes(r,..density..))+geom_bar()

Returns...

alt text http://www.drewconway.com/zia/wp-content/uploads/2010/04/density.png

The bins are now densities...

Answer 5

Your first question: Would this help?

geom_bar(aes(y=..count../sum(..count..)))

Your second question; could you use reorder to sort the bars? Something like

aes(reorder(Interest, Value, mean), Value)

(just back from a seven hour drive - am tired - but I guess it should work)