Potential regression with filled areas in ggplot2 3.4.0

[cpsievert]

I'm not entirely sure if this should be considered a regression, but I noticed this new surprising behavior with geom_area() in the 3.4.0 release candidate:

library(ggplot2)

df <- aggregate(price ~ cut + carat, data = diamonds, FUN = length)
names(df)[3] <- "n"
temp <- aggregate(n ~ carat, data = df, FUN = sum)
names(temp)[2] <- "sum.n"
df <- merge(x = df, y = temp, all.x = TRUE)
df$freq <- df$n / df$sum.n
p <- ggplot(data = df, aes(x = carat, y = freq, fill = cut)) + 
  geom_area()

Whereas the ggplot2 3.3.6 release yields:

[teunbrand]

Just as a potential workaround, the 3.3.6 behaviour can be restored by using stat = "identity".

[yutannihilation]

This change is intentional. As of v3.4.0, the default stat for geom_area() is align, one of the major new features in the release. So, as @teunbrand suggested, you can explicitly specify stat = "identity", which was the previous default. That said, we could include some explanation about the breakage like this in the NEWS. Thanks for catching.

In this particular case, I'm not sure if the latter plot is the correct one, btw (e.g. is it right that the range from around 3.5 to 4.0 is drawn 100% "Premium"?)

[roelvanklink]

Thanks, this was bugging me for two weeks. A warning message in the help or in the code might also be helpful.
Here's what i got before adding 'stat = "identity"': The two levels (brown and blue) were added up and assigned to the first level

[yutannihilation]

@roelvanklink
I think your one is different problem; what you need is position = "identity". By default, the areas stack, and the only reason you could accidentally get the unstacked result is your data didn't contain the same X values by chance. But, I sympathize with your surprise.

[teunbrand]

Perhaps we can simply add an example to the geom_area() docs on how to turn off the alignment?

From the existing example:

df <- data.frame(
  g = c("a", "a", "a", "b", "b", "b"),
  x = c(1, 3, 5, 2, 4, 6),
  y = c(2, 5, 1, 3, 6, 7)
)
a <- ggplot(df, aes(x, y, fill = g)) +
  geom_area()

# Two groups have points on different X values.
a + geom_point(size = 8) + facet_grid(g ~ .)

# stat_align() interpolates and aligns the value so that the areas can stack
# properly.
a + geom_point(stat = "align", position = "stack", size = 8)

We might add:

# To turn off this alignment, the stat can be set to "identity"
ggplot(df, aes(x, y, fill = g)) + geom_area(stat = "identity")

[yutannihilation]

Sounds good to me!

[thomasp85]

same - an example and perhaps a note in the docs should be sufficient