scales="free_y" in facet_grid has no effect when geom_bar is used

[jiho]

Initially I wanted to plot precomputed bars and was in this case

d = data.frame(x=c(runif(10), runif(10)*4), a=rep(c("foo", "bar"), 5), b=rep(c("one", "two"), each=10))

ggplot(d) + geom_bar(aes(x=a, y=x), stat="identity") + facet_grid(.~b)
ggplot(d) + geom_bar(aes(x=a, y=x), stat="identity") + facet_grid(.~b, scales="free_y")

I supposed that stat="identity" was the problem, but apparently it is the same with an histogram

d = data.frame(x=c(runif(20), runif(20)*4), a=runif(40), b=rep(c("one", "two"), each=20))

ggplot(d) + geom_bar(aes(x=a), binwidth=0.2) + facet_grid(.~b)
ggplot(d) + geom_bar(aes(x=a), binwidth=0.2) + facet_grid(.~b, scales="free_y")

[hadley]

But rows in facet_grid always share scales... Maybe you want facet_wrap?

[jiho]

OK, indeed, the problem is the same with geom_point.

The bottom line is that if I want a 2-dimensional matrix of plots (factor a in line, factor b in columns) with a different y for every subplot, I should use facet_wrap and scale my plotting window appropriately (to have one level of a by line).