new geom suggestion: separate lines for upper and lower intervals

[weiyangtham]

I'm wondering if there's room for a new geom that can produce the following graph

library(tidyverse)

# Example data
df <- tibble(y = c(0, 0, 0, 0, 0, 1, 1.5, 1.7, 1.8), 
             upper_y = y + 0.2, 
             lower_y = y - 0.2,
             year = -4:4)

p <- ggplot(df, aes(year, y)) + geom_line()

# geom_errorline would simplify this
p + 
  geom_line(aes(y = upper_y), linetype = 2) +
  geom_line(aes(y = lower_y), linetype = 2)

So instead of having to use geom_line twice, you could instead do

ggplot(df, aes(year, y, ymax = upper_y, ymin = lower_y)) + geom_errorline()

and this would fit more neatly into the idea of mapping the upper and lower intervals to the ymin and ymax aesthetics.

I imagine there's a way to achieve that same plot with geom_ribbon or geom_area but I haven't been able to do so so I'm guessing it wouldn't be straightforward even if there is.

Here is a prototype I wrote a few years ago: https://github.com/weiyangtham/econothemes

Thanks!

[yutannihilation]

I imagine there's a way to achieve that same plot with geom_ribbon or geom_area but I haven't been able to do so so I'm guessing it wouldn't be straightforward even if there is.

I agree with you. I don't come up with some nice idea for now, but I don't think we need a new geom as this won't be so difficult. e.g.

library(ggplot2)

# Example data
df <- tibble::tibble(
  y = c(0, 0, 0, 0, 0, 1, 1.5, 1.7, 1.8), 
  upper_y = y + 0.2, 
  lower_y = y - 0.2,
  year = -4:4
)

StatErrorLine <- ggproto("StatErrorLine", Stat,
  required_aes = c("x", "ymin", "ymax"),
  default_aes = aes(y = stat(y)),
  
  compute_group = function(data, scales) {
    idx <- names(data) %in% c("y", "ymin", "ymax")
    data_min <- data_max <- data[!idx]
    data_min$y <- data$ymin
    data_min$group <- paste0(data$group, "-min")
    data_max$y <- data$ymax
    data_max$group <- paste0(data$group, "-max")
    rbind(data_min, data_max)
  }
)

p <- ggplot(df, aes(year, y)) + geom_line()
p + 
  geom_line(aes(ymax = upper_y, ymin = lower_y), linetype = 2, stat = StatErrorLine)

^{Created on 2019-08-29 by the reprex package (v0.3.0)}

[weiyangtham]

I'd thought of it as a geom since I was thinking the more common use case would be one where the upper/lower bounds are provided in the data (e.g. confidence intervals from a model that has been passed through broom::tidy(model, conf.int = TRUE)) rather than being a transformation. It would also be more directly analogous to the other "interval" geoms like geom_errorbar or geom_ribbon, and therefore maybe more intuitive?

[yutannihilation]

Now I come to wonder if we can just write this with geom_ribbon():

ggplot(df, aes(year, y)) + geom_line() + 
  geom_ribbon(aes(ymax = upper_y, ymin = lower_y),
              linetype = 2, fill = NA, colour = "black")

I mean, it's just an implementational detail that GeomRibbon draws polygons, and it might be able to draw lines and polygons separately. For example, 017992b:

devtools::load_all("~/Documents/repo/ggplot2")
#> Loading ggplot2

df <- tibble::tibble(
  y = c(0, 0, 0, 0, 0, 1, 1.5, 1.7, 1.8), 
  upper_y = y + 0.2, 
  lower_y = y - 0.2,
  year = -4:4
)

# can divide the ribbon
df[4, "upper_y"] <- NA 
df[4, "lower_y"] <- NA

ggplot(df, aes(year, y)) + geom_line() + 
  geom_ribbon(aes(ymax = upper_y, ymin = lower_y),
              linetype = 2, fill = "red", alpha = 0.1, colour = "black")

^{Created on 2019-09-02 by the reprex package (v0.3.0)}

I'm not sure if this really will be an acceptable change, though...

[weiyangtham]

Something like this would be great. Is the worry that users might expect geom_ribbon() to always draw polygons?

[yutannihilation]

users might expect geom_ribbon() to always draw polygons?

I expected there's effectively no such users, but I'm not fully sure... Just curious, have you ever seen such a usage?

[weiyangtham]

Just to make sure we're on the same page, are you asking if I've seen someone use geom_ribbon with a transparent fill but they still want the polygon? I don't think I've seen that before.

[yutannihilation]

Yes, that's what I meant to ask. Thanks.