Faulty y-axis for plots based on stat_ecdf with scale_y_log10

[m-pilarski]

When I'm log-scaling the y-axis of CDF-plots (stat_ecdf), the y-values are drawn between 1 and 10 instead of 0 and 1. The y-values of the data extracted from the plot-object (ggplot_build(...)$data[[1]]$y) are ranging from 0 to 1 instead of -1 to 0. When I'm log-scaling the x-axis of CDF-Plots or the x- or y-axis of other plots (e.g. geom_line), everything is drawn as expected.

library(ggplot2)

df <- data.frame(var=1:10/10)

plot_ecdf_ident <- ggplot(df, aes(x=var)) + stat_ecdf()
plot_ecdf_log10_y <- plot_ecdf_ident + scale_y_log10()
plot_ecdf_log10_x <- plot_ecdf_ident + scale_x_log10()

plot_line_ident <- ggplot(df, aes(x=var, y=var)) + geom_line()
plot_line_log10_y <- plot_line_ident + scale_y_log10()
plot_line_log10_x <- plot_line_ident + scale_x_log10()

patchwork::wrap_plots(
  plot_ecdf_ident, plot_ecdf_log10_y, plot_ecdf_log10_x, 
  plot_line_ident, plot_line_log10_y, plot_line_log10_x, 
  nrow=2
)

range(ggplot_build(plot_ecdf_log10_y)$data[[1]]$y)
#> [1] 0 1
range(ggplot_build(plot_line_log10_y)$data[[1]]$y)
#> [1] -1  0

packageVersion("ggplot2")
#> [1] '3.4.0'
packageVersion("scales")
#> [1] '1.2.1'

^{Created on 2022-12-20 with reprex v2.0.2}

[m-pilarski]

I found out that the faulty log-scaling has nothing to do with the default pad=TRUE in stat_ecdf. I also found out that the transformation woks, when I'm feeding the the data extracted from the untransformed plot to a new plot with geom_step.

library(ggplot2)

`%>%` <- magrittr::`%>%`

df <- data.frame(var=1:10/10)

plot_ecdf_ident_nopad_1 <- ggplot(df, aes(x=var)) + stat_ecdf(pad=FALSE)

plot_ecdf_log10_y_nopad_1 <- plot_ecdf_ident_nopad_1 + scale_y_log10()

plot_ecdf_ident_nopad_2 <- 
  ggplot_build(plot_ecdf_ident_nopad_1) %>% 
  purrr::chuck("data", 1) %>% 
  ggplot(aes(x=x, y=y, group=group)) +
  geom_step()

plot_ecdf_log10_y_nopad_2 <- plot_ecdf_ident_nopad_2 + scale_y_log10()

patchwork::wrap_plots(
  plot_ecdf_ident_nopad_1, plot_ecdf_log10_y_nopad_1,
  plot_ecdf_ident_nopad_2, plot_ecdf_log10_y_nopad_2,
  nrow=2
)

^{Created on 2022-12-21 with reprex v2.0.2}