Multirow axis labels with nested grouping variables

A very simple solution which gives a similar (though not identical) result is to use faceting. The downside is that the Category label is above rather than below.

ggplot(data=data, aes(x=Group, y=Value, fill=Group)) +
  geom_bar(position = 'dodge', stat="identity") +
  geom_text(aes(label=paste(Value, "%")), position=position_dodge(width=0.9), vjust=-0.25) + 
  facet_grid(. ~ Category) + 
  theme(legend.position="none")

You can create a custom element function for axis.text.x.

library(ggplot2)
library(grid)

## create some data with asymmetric fill aes to generalize solution 
data <- read.table(text = "Group Category Value
                   S1 A   73
                   S2 A   57
                   S3 A   57
                   S4 A   57
                   S1 B   7
                   S2 B   23
                   S3 B   57
                   S1 C   51
                   S2 C   57
                   S3 C   87", header=TRUE)

# user-level interface 
axis.groups = function(groups) {
  structure(
    list(groups=groups),
    ## inheritance since it should be a element_text
    class = c("element_custom","element_blank")  
  )
}
# returns a gTree with two children: 
# the categories axis
# the groups axis
element_grob.element_custom <- function(element, x,...)  {
  cat <- list(...)[[1]]
  groups <- element$group
  ll <- by(data$Group,data$Category,I)
  tt <- as.numeric(x)
  grbs <- Map(function(z,t){
    labs <- ll[[z]]
    vp = viewport(
             x = unit(t,'native'), 
             height=unit(2,'line'),
             width=unit(diff(tt)[1],'native'),
             xscale=c(0,length(labs)))
    grid.rect(vp=vp)
    textGrob(labs,x= unit(seq_along(labs)-0.5,
                                'native'),
             y=unit(2,'line'),
             vp=vp)
  },cat,tt)
  g.X <- textGrob(cat, x=x)
  gTree(children=gList(do.call(gList,grbs),g.X), cl = "custom_axis")
}

## # gTrees don't know their size 
grobHeight.custom_axis = 
  heightDetails.custom_axis = function(x, ...)
  unit(3, "lines")

## the final plot call
ggplot(data=data, aes(x=Category, y=Value, fill=Group)) + 
  geom_bar(position = position_dodge(width=0.9),stat='identity') +
  geom_text(aes(label=paste(Value, "%")),
            position=position_dodge(width=0.9), vjust=-0.25)+
  theme(axis.text.x = axis.groups(unique(data$Group)),
        legend.position="none")

An alternative to agstudy's method is to edit the gtable and insert an "axis" calculated by ggplot2,

p <- ggplot(data=data, aes(x=Category, y=Value, fill=Group)) + 
  geom_bar(position = position_dodge(width=0.9),stat='identity') +
  geom_text(aes(label=paste(Value, "%")),
            position=position_dodge(width=0.9), vjust=-0.25)

axis <- ggplot(data=data, aes(x=Category, y=Value, colour=Group)) +
  geom_text(aes(label=Group, y=0),
            position=position_dodge(width=0.9))

annotation <- gtable_filter(ggplotGrob(axis), "panel", trim=TRUE)
annotation[["grobs"]][[1]][["children"]][c(1,3)] <- NULL #only keep textGrob

library(gtable)
g <- ggplotGrob(p)
gtable_add_grobs <- gtable_add_grob # let's use this alias
g <- gtable_add_rows(g, unit(1,"line"), pos=4)
g <- gtable_add_grobs(g, annotation, t=5, b=5, l=4, r=4)
grid.newpage()
grid.draw(g)

The strip.position argument in facet_wrap() and switch argument in facet_grid() since ggplot2 2.2.0 now makes the creation of a simple version of this plot fairly straightforward via faceting. To give the plot the uninterrupted look, set the panel.spacing to 0.

Here's the example using the dataset with a different number of Groups per Category from @agtudy's answer.

• I used scales = "free_x" to drop the extra Group from the Categories that don't have it, although this won't always be desirable.
• The strip.position = "bottom" argument moves the facet labels to the bottom. I removed the strip background all together with strip.background, but I could see that leaving the strip rectangle would be useful in some situations.
• I used width = 1 to make the bars within each Category touch - they'd have spaces between them by default.

I also use strip.placement and strip.background in theme to get the strips on the bottom and remove the strip rectangle.

The code for versions of ggplot2_2.2.0 or newer:

ggplot(data = data, aes(x = Group, y = Value, fill = Group)) + 
    geom_bar(stat = "identity", width = 1) +
    geom_text(aes(label = paste(Value, "%")), vjust = -0.25) +
    facet_wrap(~Category, strip.position = "bottom", scales = "free_x") +
    theme(panel.spacing = unit(0, "lines"), 
         strip.background = element_blank(),
         strip.placement = "outside")

You could use space= "free_x" in facet_grid() if you wanted all the bars to be the same width regardless of how many Groups per Category. Note that this uses switch = "x" instead of strip.position. You also might want to change the label of the x axis; I wasn't sure what it should be, maybe Category instead of Group?

ggplot(data = data, aes(x = Group, y = Value, fill = Group)) + 
    geom_bar(stat = "identity", width = 1) +
    geom_text(aes(label = paste(Value, "%")), vjust = -0.25) +
    facet_grid(~Category, switch = "x", scales = "free_x", space = "free_x") +
    theme(panel.spacing = unit(0, "lines"), 
         strip.background = element_blank(),
         strip.placement = "outside") + 
    xlab("Category")

Older code versions

The code for ggplot2_2.0.0, when this feature was first introduced, was a little different. I've saved it below for posterity:

ggplot(data = data, aes(x = Group, y = Value, fill = Group)) + 
    geom_bar(stat = "identity") +
    geom_text(aes(label = paste(Value, "%")), vjust = -0.25) +
    facet_wrap(~Category, switch = "x", scales = "free_x") +
    theme(panel.margin = unit(0, "lines"), 
         strip.background = element_blank())