automagically using labels (haven semantics) in ggplot2 plots

Question

I'm plotting data marked up using haven semantics, i.e. variables and values have labels defined via attributes.

Often, these labels are also what I want in my axis titles and ticks.

library(ggplot2)
mtcars$mpg = haven::labelled(mtcars$mpg, labels = c("low" = 10, "high" = 30))
attributes(mtcars$mpg)$label = "miles per gallon"
ggplot(mtcars, aes(mpg, cyl)) + geom_point() + 
scale_x_continuous(attributes(mtcars$mpg)$label, 
     breaks = attributes(mtcars$mpg)$labels, 
     labels = names(attributes(mtcars$mpg)$labels))

Could I write a helper that replaces that laborious scale_x_continuous statement with something that can more easily be iterated? E.g. something like scale_x_continuous(label_from_attr, breaks = breaks_from_attr, labels = value_labels_from_attr). Or maybe even + add_labels_from_attributes() to replace the whole thing?

I'm aware that I can write/use helpers like Hmisc::label to slightly shorten the attribute-code above, but that's not what I want here.

take a look at the `Hmisc` package which has quite a few "plotting with labels" features for ggplot. I'm not sure about the haven compatibility, but it uses the same labels as attribute specification, I think. — Matt L., Jan 25 '18 at 15:53

score 3 · Accepted Answer · edited Jun 20 '20 at 09:12

3

I don't have a good scale, but you can use a function like this:

label_x <- function(p) {
  b <- ggplot_build(p)
  x <- b$plot$data[[b$plot$labels$x]]
  
  p + scale_x_continuous(
    attributes(x)$label, 
    breaks = attributes(x)$labels, 
    labels = names(attributes(x)$labels)
  )
}

Then use as (+ won't do):

p <- ggplot(mtcars, aes(mpg, cyl)) + geom_point()
label_x(p)

Alternatively, use a pipe:

mtcars %>% { ggplot(., aes(mpg, cyl)) + geom_point() } %>% label_x()

Old solution

use_labelled <- function(l, axis = "x") {
    if (axis == "x")  {
        scale_x_continuous(attributes(l)$label, 
                           breaks = attributes(l)$labels, 
                           labels = names(attributes(l)$labels))
    } 
    if (axis == "y") {
        scale_y_continuous(attributes(l)$label, 
                          breaks = attributes(l)$labels, 
                          labels = names(attributes(l)$labels))
    }
}

Then you just give:

ggplot(mtcars, aes(mpg, cyl)) + geom_point() + use_labelled(mtcars$cyl)

Or for the y-axis:

ggplot(mtcars, aes(cyl, mpg)) + geom_point() + use_labelled(mtcars$cyl, "y")

edited Jun 20 '20 at 09:12

Community

1
1

answered Jan 25 '18 at 15:38

Axeman

27,115
6
69
82

Thanks. Repeating the data frame and variable name is exactly what I want to avoid. But thanks for sleuthing out that `ggplot_build` kills the attributes we'd need. I also found no info on how to write a new scale. – Ruben Jan 25 '18 at 15:48
Updated with a different solution. – Axeman Jan 25 '18 at 16:16
This is actually pretty cool! So `ggplot_build` doesn't kill the attributes after all, but we cannot get them through `+`? – Ruben Jan 25 '18 at 16:39
It strips them in `data`, but not in `plot$data` it seems. The `+` issue is because `+` doesn't actually make the plot so far available to the function on the right hand side. – Axeman Jan 25 '18 at 16:42

score 0 · Answer 2 · edited Dec 26 '19 at 22:00

Another approach is to write a wrapper for ggplot() that has its own class. Then attributes have full visibility when the corresponding print method is called. See ?ag.print from package 'yamlet' (0.2.1).

library(ggplot2)
library(yamlet)
library(magrittr)

mtcars$disp %<>% structure(label = 'displacement', unit = 'cu. in.')
mtcars$mpg %<>% structure(label = 'mileage', unit = 'miles/gallon')
mtcars$am %<>% factor(levels = c(0,1), labels = c('automatic','manual'))
mtcars$am %<>% structure(label = 'transmission')

agplot(mtcars, aes(disp, mpg, color = am)) + geom_point()

automagically using labels (haven semantics) in ggplot2 plots

2 Answers2

Old solution