ggplot2: facet_wrap strip color based on variable in data set

Question

Is there a way to fill the strips of facets created with facet_wrap based on a variable supplied with the data frame?

Example data:

MYdata <- data.frame(fruit = rep(c("apple", "orange", "plum", "banana", "pear", "grape")), farm = rep(c(0,1,3,6,9,12), each=6), weight = rnorm(36, 10000, 2500), size=rep(c("small", "large")))

Example plot:

p1 = ggplot(data = MYdata, aes(x = farm, y = weight)) + geom_jitter(position = position_jitter(width = 0.3), aes(color = factor(farm)), size = 2.5, alpha = 1) + facet_wrap(~fruit)

I know how to change the background color of the strips (e.g. to orange):

p1 + theme(strip.background = element_rect(fill="orange"))

Is there a way to pass on the values in the variable size in MYdata to the parameter fill in element_rect?

Basically, instead of 1 color for all strips I would like the strip background color of small fruits (apple, plum, pear) to be green and the background color of large fruits (orange, banana, grape) to be red.

Answer 1

With a little bit of work, you can combine your plot with a dummy gtable that has the right grobs,

d <- data.frame(fruit = rep(c("apple", "orange", "plum", "banana", "pear", "grape")), 
                farm = rep(c(0,1,3,6,9,12), each=6), 
                weight = rnorm(36, 10000, 2500), 
                size=rep(c("small", "large")))

p1 = ggplot(data = d, aes(x = farm, y = weight)) + 
  geom_jitter(position = position_jitter(width = 0.3), 
              aes(color = factor(farm)), size = 2.5, alpha = 1) + 
  facet_wrap(~fruit)

dummy <- ggplot(data = d, aes(x = farm, y = weight))+ facet_wrap(~fruit) + 
  geom_rect(aes(fill=size), xmin=-Inf, xmax=Inf, ymin=-Inf, ymax=Inf) +
  theme_minimal()

library(gtable)

g1 <- ggplotGrob(p1)
g2 <- ggplotGrob(dummy)

gtable_select <- function (x, ...) 
{
  matches <- c(...)
  x$layout <- x$layout[matches, , drop = FALSE]
  x$grobs <- x$grobs[matches]
  x
}

panels <- grepl(pattern="panel", g2$layout$name)
strips <- grepl(pattern="strip_t", g2$layout$name)
g2$layout$t[panels] <- g2$layout$t[panels] - 1
g2$layout$b[panels] <- g2$layout$b[panels] - 1

new_strips <- gtable_select(g2, panels | strips)
grid.newpage()
grid.draw(new_strips)

gtable_stack <- function(g1, g2){
  g1$grobs <- c(g1$grobs, g2$grobs)
  g1$layout <- transform(g1$layout, z= z-max(z), name="g2")
  g1$layout <- rbind(g1$layout, g2$layout)
  g1
}
## ideally you'd remove the old strips, for now they're just covered
new_plot <- gtable_stack(g1, new_strips)
grid.newpage()
grid.draw(new_plot)

Answer 2

I would love to know how to do that, it is a great idea. One idea is to generate each chart independently with a different color as you do and then use something like multiplot or viewports to show then side by side - it will require a bit more work.

if you want to extract the legend, which you will need for this approach - here is some code from Hadley that I found a while back

g_legend<-function(a.gplot){
  tmp <- ggplot_gtable(ggplot_build(a.gplot))
  leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
  legend <- tmp$grobs[[leg]]
  return(legend)}

see how it is extracted it from chart p, and then I took it out of the plot legend <- g_legend(p) lwidth <- sum(legend$width) #if you want to define the viewport based on this p <- p + theme(legend.position="none")

then you eventually draw it

grid.newpage()
vp <- viewport(width = 1, height = 1)
#print(p, vp = vp)

submain <- viewport(width = 0.9, height = 0.9, x = 0.5, y = 1,just=c("center","top"))
print(p, vp = submain)
sublegend <- viewport(width = 0.5, height = 0.2, x = 0.5, y = 0.0,just=c("center","bottom"))
print(arrangeGrob(legend), vp = sublegend)

Good luck